There is a line the startup ecosystem loves because it sounds wise, generous, and vaguely inevitable: venture capital is the real deal.

It is one of those sentences that travels well at demo days, on panels, inside accelerator programs, and across LinkedIn. It suggests that capital is freedom. More money means more runway, more experiments, more hires, more shots on goal. In software, that can be true often enough to become a believable default.

In deep tech, it can be dangerously wrong.

The first institutional check does not simply give a science company more oxygen. It can change what the company is allowed to become. It can force a wide scientific possibility space into a narrow company narrative before the technology has finished revealing its strongest markets, its most valuable applications, or even its most rational business model.

Deep tech is not a more complex version of SaaS with more expensive equipment. It is not software wearing a lab coat. It is not “AI, but with atoms.” Deep tech companies are built around technologies whose real economic value often emerges slowly, indirectly, and non-linearly.

• The first use case is often not the best use case.

• The first market is often not the largest market.

• The first product story is often not the strategy.

Sometimes it is just the first story that was legible enough to raise interests.

A material invented for one industrial application may become more valuable in aerospace, defense, batteries, semiconductors, medical devices, or process heat. ‘Small’ changes in surface functionalization can capture entirely new markets with significant higher shares. Except, those ‘small’ changes need years to evolve, still, and may have not been visible in the first years of the company operating at all.

A synthetic biology platform may begin with materials and later reveal therapeutic relevance. An mRNA platform may begin as one vaccine story and later expand into respiratory disease, oncology, rare disease, and latent viruses. A quantum technology company may think it is selling hardware, only to discover that the real value sits in sensing, simulation, infrastructure, standards, or data.

This is why early deep tech companies should not be understood primarily as “startups with products.” At their earliest stages, the best ones are option spaces. They are living portfolios of technical, commercial, regulatory, manufacturing, and partnership possibilities. The value is not only in what has already been built. It is in the right to keep learning before the future is prematurely selected.

That is exactly where early VC can become dangerous.

Capital is not neutral. Capital has a governance logic.

And in deep tech, the wrong governance logic at the wrong time can destroy value long before anyone calls it failure.

The company is not the product. The option space is the product.

The lazy version of this debate is “VC good” versus “VC bad.” That is not the argument. The serious argument is that early priced venture capital can create path-dependent governance and commercialization commitments before the company’s full opportunity set is knowable.

Path dependence is not a buzzword. It describes a situation in which earlier choices reshape later possibilities. Once a system starts moving down one route, self-reinforcing dynamics make it harder to reverse. In organizational theory, this has been studied for decades. Sydow, Schreyögg and Koch describe organizational path dependence as a process in which early choices can produce lock-in through self-reinforcing mechanisms. Their work is useful because it gives us a language for something founders often feel but cannot quite name: a financing event can become an organizational architecture, not just a cash event. You can read the paper on organizational path dependence here.

Real options theory gives us the second half of the argument. Under uncertainty, value does not only come from current cash flows. It also comes from the ability to wait, stage, expand, redirect, pause, or abandon decisions as new information arrives. That logic is especially relevant for R&D, where information does not arrive all at once. Research creates future decision rights. It gives the company the possibility to discover, over time, which applications deserve more capital and which should be abandoned. This is why real options approaches are frequently used to understand R&D investment under uncertainty. See, for example, this paper on real options and R&D.

Now put those two ideas together and the problem becomes obvious. Deep tech companies are path-dependent real option systems. Early decisions matter because they change the future decision space. Financing decisions matter even more because they do not only bring money into the company. They bring expectations, control rights, board dynamics, valuation pressure, milestone logic, and exit assumptions.

A deep tech company may not know its best market in year one. It may not know whether it should sell a product, license the technology, build infrastructure, enter a regulated market, partner with incumbents, create application-specific spinouts, or become a platform company. That uncertainty is not a defect. In many cases, it is the asset.

The problem starts when venture capital treats this uncertainty as something that must be compressed into a fundable story too early.

Because VC rarely asks only, “What could this technology become?” It also asks, “What can this become within our fund logic?”

• Can it grow fast enough?

• Can it raise the next round?

• Can it become venture-scale?

• Can it fit a category investors understand?

• Can it be priced?

• Can it exit?

Those are legitimate questions. But they are not the same as asking where the maximum long-term value of the science might sit.

Subscribe now

The asymmetry nobody wants to discuss

There is an uncomfortable asymmetry in deep tech financing that many ecosystem conversations politely avoid: a company can often move from organic growth into VC, but it is very hard to move from VC back into organic growth.

Those two paths are not symmetrical.

A science company that grows through grants, early revenue, strategic partnerships, project finance, patient capital, or carefully structured non-dilutive funding can still choose later to raise VC. It can wait until the technology is clearer. It can learn which markets are real. It can preserve optionality while the science matures. Then, once the company knows what kind of company it actually is, it can decide whether venture capital is the right layer of the capital stack.

But once a company has taken significant early priced institutional VC, it has usually absorbed an entirely different architecture. It has a valuation to defend, a board to manage, investor rights to respect, a burn profile to feed, a growth story to maintain, employees hired against a specific roadmap, and a next-round narrative that begins shaping decisions long before the next round actually happens.

That is why the sentence “you can always find alternative financing later” is misleading. Later alternatives may technically exist, but the company that needs them is no longer the same company. Its cap table has changed. Its governance has changed. Its cost base has changed. Its investor base has changed. Its strategic freedom has changed.

Going from organic to VC is often a strategic choice.

Going from VC back to organic is often a correction, a recap, a down round, a distressed bridge, a painful acquisition, or a quiet collapse.

This asymmetry matters more in deep tech than in other ventures because deep tech timelines are longer, uncertainty resolves more slowly, and the most valuable applications are often not obvious at incorporation. In software, a wrong market assumption may be painful but relatively fast to test. In deep tech, the wrong market assumption can consume years of lab work, regulatory preparation, manufacturing design, partner development, and capital structure.

The tragedy is not only that some companies fail. The deeper tragedy is that some technologies may be scientifically right but financially trapped because the company around them was forced into the wrong shape too early.

Founder-friendly terms are not the whole story

A common response to this argument is that venture terms have become cleaner. That is partly true.

According to Cooley’s Q4 2024 Venture Financing Report, 96% of reported US deals had a 1x liquidation preference and 96% used non-participating preferred stock. In Europe, Orrick’s 2025 European Deal Flow report reported that 77% of deals featured a 1x non-participating liquidation preference and 79% included broad-based weighted-average anti-dilution.

On the surface, this sounds reassuring. The most visibly aggressive terms are not everywhere. Not every term sheet is full of 2x participating preferences, brutal ratchets, and founder-hostile traps. But this is exactly where the debate becomes too shallow.

Optionality is not destroyed only by ugly liquidation preferences. It is destroyed by the whole governance system that assembles around the financing.

Board seats matter. Protective provisions matter. Consent rights matter. Milestone framing matters. Hiring plans matter. Burn expectations matter. Category bias matters. Next-round dependency matters. Exit assumptions matter.

Kaplan and Strömberg’s classic work on venture capital contracts shows that VC financings separate cash-flow rights, board rights, voting rights, liquidation rights, and other control rights. These are not legal decorations. They determine who controls what when the company performs well and, more importantly, who controls what when it performs poorly.

That last part is critical for deep tech. In a software company, “poor performance” may mean the product is not selling. In a deep tech company, “poor performance” may simply mean that the biology took longer, the physics was harder, the manufacturing process needed another iteration, the first market was wrong, the regulator moved slowly, or the most valuable application turned out to be different from the one in the pitch deck.

And if investor control increases precisely when technical uncertainty is still resolving, the company may be pushed toward the wrong decision at the exact moment when patience would have created more value.

The valuation is not the strategy

The ecosystem loves valuation because valuation is simple. Big number good. Higher valuation better. Unicorn status best.

Unfortunately, this is also how intelligent people get fooled.

The valuation is not the strategy. The valuation is not the value of the science. The valuation is not even necessarily the value of the company to its founders, employees, or common shareholders. It is a price attached to a particular security with particular rights, preferences, protections, and expectations.

Gornall and Strebulaev’s paper Squaring Venture Capital Valuations with Reality found that reported post-money valuations of unicorns averaged 48% above fair value once the economic rights of preferred shares were accounted for. That does not mean every high valuation is fake. It means headline post-money valuation can materially overstate the economic value of common equity because preferred shares often carry richer rights.

In plain language: the latest investor may have bought a better instrument than everyone else.

That matters in deep tech because a company can appear successful on paper while quietly becoming strategically trapped. A founder may celebrate a high valuation without fully appreciating what happens if the next round is delayed, if the company needs to pivot, if the best market requires five more years of development, or if a strategic acquisition clears the preference stack but undersells the long-term platform.

When a deep tech company says, “We raised at a $500 million valuation,” the serious response is not simply “Congratulations.” The serious response is: what kind of $500 million?

Who controls the board? What happens in a $200 million exit? What happens in a $600 million exit? What happens if the next round is a down round? What happens if the technically rational move is to license rather than build? What happens if the best application was not in the original deck? What happens if the company is scientifically right but venture-wrong?

These are not cynical questions. They are the questions that determine whether the science survives its own financing.

Zymergen, or what optionality compression looks like in public

The most useful cautionary case is Zymergen. Not because it proves that early VC always destroys deep tech companies. It does not. But because it shows how brutally a broad platform story can collapse when commercial reality fails to match the financed narrative.

Zymergen presented itself as a platform company with potential across advanced materials, automation, drug discovery, and biomolecular design. Its public filings described a large universe of biomolecular possibilities and multiple future markets. That is exactly the kind of language deep tech investors love when the platform story is still expanding.

Then the narrative narrowed.

After a portfolio review, Zymergen disclosed that it would focus on a smaller number of programs, cut staff, and discontinue programs including Hyaline after concluding that the near-term market opportunity was smaller than previously expected. The company’s own public filing is available here.

In 2024, the SEC charged Zymergen with misleading IPO investors about Hyaline’s market potential, revenue prospects, and customer pipeline. The SEC noted that Zymergen raised approximately $530 million in its April 2021 IPO. You can read the SEC release here. In July 2022, Ginkgo announced an all-stock transaction valuing Zymergen at approximately $300 million market capitalization. The transaction announcement is here.

This sequence is not a clean causal proof. It would be irresponsible to pretend it is. But as a public warning, it is powerful. A broad technical platform was translated into a commercial narrative. The narrative broke. The platform was narrowed, repriced, and absorbed into a strategic transaction.

That is optionality compression becoming visible.

The lesson is not “never build a platform company.” The lesson is that platform companies are fragile when their financing model demands premature proof from the wrong application. The platform may still contain scientific value, but the company can lose the right to discover where that value should live.

Oxford Nanopore, or what preserved optionality can look like

The strongest argument is never built only on failures. So we should also look at a more constructive case: Oxford Nanopore.

Oxford Nanopore’s 2021 prospectus described a platform that could analyze DNA and RNA, with potential extension toward proteins and other molecules, across both research and applied markets. The prospectus is available here. By 2021, its technology had already appeared in more than 2,100 publications across human, cancer, plant, animal, pathogen, and environmental science. By 2024, the company reported more than 14,000 cumulative peer-reviewed publications, including around 3,000 in 2024 alone, according to its annual report here.

This is what preserved platform optionality can look like. The company did not become interesting because it guessed one perfect application on day one. It became interesting because the technology was allowed to diffuse across scientific and applied use cases long enough for the ecosystem to discover where the value sat.

That is the deep tech logic. Sometimes the market has to be discovered through the technology. Sometimes the application space is created by users. Sometimes the most valuable path looks like a distraction in year two. Sometimes the route that creates the most value is the route that would have looked too slow, too broad, or too unfocused to an impatient financing model.

Moderna shows why this is not an anti-capital argument

Moderna adds the nuance that keeps this argument honest.

Moderna is not a story about avoiding outside capital. It is a story about capital structure fit. Its 2024 annual report describes an mRNA platform spanning four franchises: respiratory vaccines, latent virus vaccines, oncology therapeutics, and rare disease therapeutics. The filing is available here.

That breadth matters because it shows the platform logic in action. A technology that begins with one dominant commercial narrative can branch across multiple markets if the company retains enough room to learn, redirect, and scale.

But Moderna also shows that optionality can be constrained by strategic alliances, not only by VC. Capital always has a governance effect. Partnerships can create restrictions. Licensing deals can shape development choices. Public markets can impose their own pressure. There is no pure form of capital untouched by incentives.

That is why the correct question is not “capital or no capital?” The correct question is: what kind of capital, with what rights, at what moment, against what uncertainty?

In 2024, Blackstone agreed to provide up to $750 million to fund Moderna’s flu program while Moderna retained program rights and control, according to reporting by Fierce Biotech. That kind of structure points toward the real issue. Deep tech does not need less capital. It needs capital that does not amputate the option space before the science has finished speaking.

The Apple argument is not an argument

Whenever someone criticizes venture logic, someone else eventually says something like “Apple was also once a startup.”

This sounds clever for about three seconds. Then it collapses.

Apple is not a dataset. Apple is a survivor story. It is not representative evidence for early VC as a financing model for deep tech. It is not even a good analogy for most science-based companies today.

The capital markets are different. The fund structures are different. The cost of scaling is different. The regulatory environments are different. The technology cycles are different. The timelines for biology, hardware, energy, robotics, quantum, semiconductors, and advanced materials are different.

Using Apple as proof that early VC is structurally right for deep tech is like using Michael Jordan as proof that basketball is a safe career plan. It is winner-tail storytelling.

And winner-tail storytelling is one of the main ways ecosystems avoid looking at structural failure.

The real question is not whether a few extraordinary companies became enormous after taking venture money. Of course they did. The real question is how many science-based companies took early institutional capital before technical uncertainty had resolved, narrowed their application space too soon, became dependent on the next round before market clarity existed, and then exited below their long-term industrial value or died before the underlying technology found its correct home.

That is the dataset we need.

And it is exactly the dataset the ecosystem does not like to discuss.

Europe’s problem is not only “too little capital”

Europe loves to talk about the deep tech funding gap. The gap is real. The European Investment Bank has repeatedly emphasized that Europe needs more innovation and scale-up financing. See the EIB’s investment report here and related press release here.

BCG has argued that deep tech has grown from roughly 10% of venture funding a decade earlier to around 20%. See BCG’s report here. McKinsey has also argued that Europe’s share of global deep tech funding increased from around 10% in 2019 to 19% in 2023. See McKinsey’s European deep tech report here.

So yes, Europe needs more capital. But that is only half the sentence.

Europe also needs better-shaped capital. That’s the rest of the sentence.

If we simply import Silicon Valley venture logic into science-heavy EU-based ecosystems, we may solve the funding gap while creating a governance mismatch. We may produce more rounds, more valuations, more press releases, more “deep tech unicorns,” and still lose the underlying industrial value.

The question is not simply, “How do we get more VC into European deep tech?” The better question is,

“How do we finance scientific optionality without prematurely collapsing it into fund-compatible narratives?”

That is harder. It is also much more important.

What good deep tech capital should do

Good deep tech capital preserves optionality while reducing uncertainty. Bad deep tech capital reduces optionality before it reduces uncertainty.

That is the distinction.

Good capital asks what technical risks need to be resolved next, what information would change the strategy, which applications should remain open, which partnerships increase strategic freedom, and which milestones create real learning rather than fundraising theatre.

Bad capital asks what story can be funded, what category can be sold, what valuation can be marked, what milestone can be announced, and what exit can be underwritten.

Deep tech needs capital stacks, not capital ideology.

That capital stack may include a combination of grants, revenue, strategic partnerships, milestone-based technical financing, project finance, royalty finance, venture debt after technical de-risking, patient equity, corporate joint ventures, public co-investment, licensing structures, application-specific spinouts, and yes, at times, under specific conditions, also VC.

Because in deep tech, validation is not “someone priced the company.” Validation is technical progress, customer relevance, manufacturing feasibility, regulatory clarity, strategic fit, and the preservation of enough optionality to let the best market emerge.

The question every founder should ask before taking the first check

Before taking early VC, every deep tech founder should ask one question:

What future am I making impossible by accepting this money now?

Not just how much runway the round buys. Not just whether the valuation is high. Not just whether the investor is prestigious. Not just whether the fund has a good brand.

The real question is what paths become harder after the round.

Can the company still license if licensing becomes the best route? Can it still go slower if technical learning requires it? Can it still explore a second application that was not in the deck? Can it still work with strategic partners without triggering investor resistance? Can it still survive without a fast Series A? Can it still reject an acquisition that clears the preference stack but undersells the platform? Can it still be scientifically right if it becomes temporarily venture-wrong?

If the answer is no, the round is not just financing.

It is selection.

The question institutions should ask

Institutions need a different question. Are we building deep tech ecosystems that maximize scientific and industrial value, or are we building venture pipelines that only recognize value when it looks like a fundable startup?

Those are not the same thing.

A university spinout may create more value as an industrial platform than as a venture-backed standalone company. A materials technology may be better commercialized through licensing than through a forced growth-company structure. A biotech platform may need a portfolio architecture rather than a single-asset story. A quantum company may need infrastructure financing rather than SaaS-style milestones. A climate hardware company may need project finance before venture equity. A dual-use company may need procurement access before it needs another priced round.

If the ecosystem has only one dominant success template, it will force too many technologies into the wrong shape.

That is how value dies. Not dramatically, not visibly, not always through spectacular bankruptcy, but through a thousand reasonable decisions made too early.

What should change now

First, we should stop calling early VC “validation” in deep tech. It is not validation. It is a financing choice with embedded constraints.

Second, we should separate technical milestones from commercial theatre. A good milestone should increase knowledge. It should not merely make the company easier to pitch.

Third, every financing round should include an optionality impact assessment. Which markets remain open? Which are now harder? Which partnerships become impossible? Which technical detours can still be funded? Which exit paths become implicitly preferred?

Fourth, governance should be treated as product architecture. In deep tech, board rights and consent rights are not legal details. They shape the future of the technology.

Fifth, capital stacks should be built around uncertainty rather than against it. A platform company should not be financed as if it were already a single-product company. A science company should not be punished for learning that the first market was wrong. A deep tech company should not be forced into capital instruments that panic when nature refuses to obey a fund cycle.

The first check is not just money

The first check is not just money. It is a constraint.

Sometimes that constraint is worth it. Sometimes it focuses the company. Sometimes it brings discipline. Sometimes it helps a technology scale faster than it ever could otherwise.

But sometimes the first check selects the wrong future.

And in deep tech, that mistake can be irreversible.

The tragedy is not only that companies fail. The deeper tragedy is that technologies can succeed scientifically while failing financially because they were forced into the wrong company shape too early.

We keep saying Europe needs more science-based companies. Then we push them into venture templates before we know whether venture is the right commercialization route.

We keep saying founders need to move faster. But maybe the real problem is that capital is moving too early, too narrowly, and with too much confidence.

The future of deep tech will not be won by whoever writes the first check.

It will be won by whoever understands what that check destroys.

Subscribe now

The Part Nobody Admits Out Loud

There is a conversation that happens in corridors, over coffee, between people who have spent real time inside deep tech programmes. Founders who have been through multiple cohorts. Programme managers who started with genuine intent. Funders who keep seeing the same pattern repeat.

The conversation goes something like this:

It keeps running because the people inside it need it to keep running. Not necessarily out of cynicism. But because the structure they are operating within rewards continuation, not conversion. Activity, not outcome. Visibility, not industrial deployment.

This is not a local problem. It is not a problem that more funding solves, or that better management fixes, or that a new cohort of founders will finally overcome.

It is a structural problem. And it has a structural solution.

The Label Justified the Budget. The Budget Built the System. The System Protects the Budget.

When deep tech became the policy priority — and it became one quickly, as AI enthusiasm started colliding with the recognition that software alone does not build industrial capacity — money followed.

Public programmes were funded. Accelerators were launched. Incubators rebranded. Technology transfer offices expanded. A new category of support infrastructure materialised, staffed by people hired to deliver it, measured by metrics inherited from the programme generation before.

This is where the structural problem begins. Not in the people. In the sequence.

When you hire staff to run a programme, those staff need the programme to continue in order to have jobs. When you measure a programme by activity outputs — cohorts completed, founders supported, investor meetings facilitated — those staff will produce activity outputs. When the programme’s renewal depends on its reporting, and the reporting measures activity rather than outcome, the programme will become very good at generating reportable activity.

None of this requires bad intent. It requires only that people respond rationally to the incentive environment they are placed in. Which they do. Always.

What Gets Measured Is What Gets Built — And It Is Not the Technology

Here is the mechanism, stated precisely.

A programme is designed with a mandate — support deep tech ventures, accelerate commercialisation, build industrial capacity. That mandate is translated into KPIs. Those KPIs are inherited, almost universally, from one of two sources: the consumer startup ecosystem, or public grant reporting requirements. Neither was calibrated for deep tech development timescales.

So the KPIs look like this:

• Follow-on funding secured.

• Investor meetings held.

• Cohort completions.

• Spinout entities created.

• Media mentions.

• Pitch event participations.

• Jobs created within mandate period.

These are not outcome indicators. They are activity proxies. They measure what is happening inside the programme. They do not measure whether what is happening inside the programme will produce industrial market outcomes at the timescales deep tech development requires.

But here is the critical part: they shape behaviour.

When a programme rewards follow-on funding secured, founders optimise for investor readiness — narrative development, pitch polish, network density — rather than for evidence depth and technical reproducibility. When a programme rewards cohort completions, programme managers fill cohorts — including with ventures that are programme-experienced rather than technically substantive. When a programme rewards spinout entity creation, TTOs create entities — regardless of whether those entities have the foundation to survive without continued programme support.

The measurement system does not passively observe the programme. It actively constructs it.

And what it constructs, progressively, is a programme optimised for its own metrics.

Which are not the metrics of industrial conversion.

The Budget Flows to the Intermediary — By Design

There is a political economy to this that makes it more durable than it should be.

Public and semi-public funds move more easily into programme infrastructure than into direct technology support. The reason is structural: programmes generate accountability artefacts. Reports. Event documentation. Cohort lists. Participation certificates. KPI dashboards. These are easy to produce, easy to audit, easy to present to oversight bodies, and easy to use as justification for renewal.

Direct technology support — funding a founder to move a material from TRL 3 to TRL 5 over eighteen months — is harder to document in ways that satisfy quarterly reporting requirements. The evidence that matters is scientific and industrial. It does not fit neatly into a programme reporting template.

So the budget flows to the intermediary. Not because the intermediary produces better outcomes. Because the intermediary produces better paperwork.

The result is a layer of infrastructure — incubators, accelerators, TTOs, programme offices — that sits between public funding and the technologies it is nominally meant to support, consuming a substantial share of that funding in its own operation, and justified by metrics that measure its operation rather than its effect.

In the UK alone, one recent analysis estimated the publicly funded TTO and accelerator infrastructure costs in the order of £300 million annually. As documented in the UK version of this in forensic detail — the equity stakes, the deal timelines, the budget flows, the spinout numbers that don't survive scrutiny. The data is damning. But the more important question is not what the numbers show. It is why the system keeps producing them.

Against that expenditure, the sector’s own data shows spinout activity heavily concentrated in a handful of institutions, with the majority of deep tech companies building themselves entirely outside the formal support system. The 66% who never went through a TTO. The founders who didn’t wait six months for a licence negotiation. The people who just built something.

The system is not failing to reach them. It was not designed for them. It was designed for the metrics it reports.

This Is What the Compounding Looks Like — Before It Becomes Visible

What makes this particularly difficult to address is the temporal structure of the damage.

The KPI configuration is set at programme design. The behavioural adaptations it induces begin immediately — but they are invisible in early reporting, because the metrics look fine. Founders are attending workshops. Investor meetings are happening. Cohort numbers are healthy.

The structural distortion accumulates underneath. Founders are being trained, implicitly, to optimise for programme participation rather than industrial readiness. The selection mechanism is progressively filtering toward ventures that perform well in programme contexts — which is not the same as ventures that will convert industrially. Capital is being mis-sequenced: equity conversations opening before the evidence base can support them, because the programme’s KPIs reward funding signals over validation milestones.

By the time the consequences are visible — stalled Series A rounds, ventures in grant-dependent survival, technologies that never reached deployment — the causal chain runs back years. And the standard attribution is wrong: the failure gets assigned to the venture, or to market conditions, or to the inherent difficulty of deep tech.

Never to the programme structure that shaped the trajectory.

This is the accountability gap that makes the system self-perpetuating. The damage is real. The cause is invisible within the reporting architecture. The programme renews.

What the Relief Looks Like — And Why It Has to Start Before Execution

If you are running a programme, funding one, or designing the next mandate — this is the section that is actually for you.

Not because what follows is a framework to adopt or a methodology to implement. But because the structural leverage point is specific, and it is earlier than most institutions think.

The measurement architecture that determines long-term programme outcomes is set before execution begins. Before the first cohort is selected. Before the first reporting cycle runs. Before the first behavioural adaptation has a chance to become structurally embedded.

Once execution starts, each reporting cycle under a misaligned KPI configuration induces adaptations that compound. Founders learn what the programme rewards. Programme managers learn what renewal requires. The system learns to produce what it is measured for. And each cycle makes the correction more expensive — not because the problem grows linearly, but because the behavioural patterns it induces become harder to reverse.

What that realignment requires — specifically, which indicators need to shift, in which sequence, and against which baseline — is what distinguishes a programme structurally capable of producing industrial outcomes from one that is not.

That question has a measurable answer. But the determining variables are typically embedded before the first cohort is ever selected. It is not the same answer for every programme. It depends on the TRL range of the intake, the mandate horizon, the capital and industrial architecture of the ecosystem the programme operates within, and the behavioural incentives the current KPI configuration is already creating.

Getting that answer — before the next mandate cycle locks in — is the intervention point.

The Three Levels at Which the System Reproduces Itself

The self-perpetuating quality of structurally misaligned programmes is not accidental. It operates at three distinct levels, each reinforcing the others.

These dynamics are not merely cultural or organisational. They produce predictable behavioural adaptation patterns across founders, programme operators, capital allocators, and institutional reporting structures.

Over time, those adaptations compound into measurable shifts in venture selection, capital timing, evidence sequencing, and industrial conversion probability.

In other words: programme KPI architectures do not simply measure ecosystems.
They shape their outcome distributions.

Read more

Something shifted in the past eighteen months.

If you have been working in deep tech for more than a few years — as a founder, as a funder, as someone inside a programme — you have probably noticed it. The language changed. The room changed. Suddenly, everyone is in deep tech.

The incubator that spent five years running SaaS cohorts updated its website. The accelerator whose partners have never seen the inside of a laboratory added “deep tech” to its mandate. The public programme that used to count follow-on funding rounds as its primary success metric is now talking about industrial sovereignty and technological resilience.

The conferences filled with new speakers. The LinkedIn ecosystem produced new experts. The funding calls started citing quantum, advanced materials, biotech — not because the programme structure changed, but because the vocabulary did.

If you felt something was off, you were right.

And if you stayed quiet about it — because the enthusiasm seemed genuine, or because you didn’t want to be the one who said it — you are not alone.

The Label Rotated. The Structure Didn't.

What happened is structurally predictable. It happens in every domain when a label acquires funding gravity.

The AI boom demonstrated something important: software scales fast, attracts capital quickly, and generates enormous visibility. It also demonstrated something that took longer to become obvious: a lot of it burns money without creating proportional industrial or societal value. As that recognition spread, a reorientation began. Deep tech — hardware-intensive, science-based, slow, expensive, regulation-gated — started to look like the correction. The real thing.

And into that space moved the same actors, with the same infrastructure, carrying the same assumptions.

This is not a story about bad intentions. Most of the people running these programmes genuinely believe they are doing something useful. The problem is structural. When an actor optimised for one type of activity — short cycles, visible outputs, investor-facing milestones — rebrands for a domain that operates on fundamentally different physics, the mismatch is not visible on day one. It compounds.

Deep tech does not mature on reporting cycles. It matures on scientific, industrial, and regulatory timescales. A programme whose KPIs were calibrated for SaaS — follow-on funding secured, investor meetings held, cohort completions counted — does not become a deep tech programme by adding the words to its mandate. It becomes a deep tech programme by rebuilding its measurement logic from the ground up.

Almost none of them have done that.

What you see instead is a structural reproduction: the same demo days, the same pitch coaching, the same investor readiness modules, the same quarterly reporting frameworks — applied to founders who are trying to move a technology from TRL 3 to TRL 5, which takes years of laboratory work that no pitch competition accelerates.

The programme looks like deep tech support. The activities look like deep tech support. The KPIs confirm that deep tech support is happening.

And the technologies stall anyway.

This Would Be Fine If It Were Simply Ineffective.

This would be a contained problem if it were simply ineffective.

The more serious issue is what it actively produces — in the ventures it touches, in the founder pipeline it shapes, and in the industrial capacity it claims to be building.

Because the consequences do not stay inside the programme. They compound. They move through the ecosystem. And by the time they are visible in outcomes — in stalled Series A rounds, in technologies that never reached industrial deployment, in founders who burned out or gave up — the causal chain runs back years to decisions that looked, at the time, entirely reasonable.

What that chain looks like, and where it leads for European industrial capacity specifically, is what the rest of this piece covers.

If You're Inside This System…

If you are running an initiative, funding one, designing one, or sitting inside one wondering why the outcomes don’t match the effort — the second half of this piece is for you.

Not more diagnosis. What you get is:

A clear account of the specific mechanisms that convert well-intentioned programme activity into structural outcome failure — and the minimal correctives that change the trajectory before the next decision locks it in.

The question is not whether your efforts are well managed. The question is what the current structure is mathematically biased to produce. That is a different question. And it has a specific answer.

What It Actually Produces — At Three Levels, In That Order.

The consequences operate at three levels. They are not independent — they compound across each other.

At the venture level:

Read more

[Previous piece: Activity ≠ Outcome: Why we have been measuring the wrong thing for deep tech for years]

There is a number that changes how you see the entire European deep tech ecosystem. Not a funding number. Not a failure rate.

A timing number.

The median time from founding to a hard commercialisation outcome — IPO, acquisition, industrial deployment, recurring revenues at scale — in a pilot sample of 20 European deep tech companies is 9.5 years.

Several took fifteen. Some took twenty-two.

The typical programme measurement window — the period during which a programme tracks, reports on, and claims credit for the ventures it supports — is two to four years.

That gap — between 4 and 9.5, between 4 and 22 — is not a rounding error.

It is the structural blind spot of the entire system.

And almost everything that goes wrong in European deep tech commercialisation lives inside it.

What happens in the gap

When a programme ends, tracking stops.

Not because anyone decided to stop caring. Because the institutional accountability horizon ends there. Mandate periods close. Reporting obligations conclude. Teams move on to the next cohort.

The ventures continue. For years. Sometimes for decades.

And what happens to them in those years — the stalled Series A rounds, the failed scale-up attempts, the pivots away from the core science, the quiet acquisitions at below-potential valuations, the technologies that simply never reach the market — none of it feeds back into the programme’s record.

The programme’s record shows what happened during the programme.

Which means it shows: activity.

In one of my published works — Do Startup Programmes Actually Commercialise Science? — I mapped this gap systematically for the first time. The finding is not that programmes are failing. The finding is more uncomfortable than that:

This is not a measurement error. It is a measurement domain mismatch. The instrument is accurate. It is being applied outside the range where it is valid.

A thermometer is not broken because it cannot measure distance.

Why this matters differently depending on where you sit

The measurement gap feels different depending on which side of it you are standing on.

If you run an initiative, the gap is a defensibility problem you cannot fully solve from inside the system. You report what you can see. What you can see is activity. Activity looks like progress. Until the renewal discussion arrives and someone — a funder, a ministry, an oversight body — asks for evidence of actual commercialisation impact. And the honest answer is that you do not have it. Not because you did not create it. Because the tracking ended before it could exist.

If you are a founder inside one of these initiatives, the gap operates differently — and more dangerously. The programme optimises for what it can measure within its window. Which means it rewards what looks like progress on a two-to-four year horizon. Investor readiness. Pitch quality. Follow-on funding. Narrative clarity.

These are not bad things to develop. But they are not the same as the things that actually determine whether your science reaches the market. Technical validation. Industrial integration. Regulatory groundwork. The slow, expensive, non-linear work of turning a laboratory result into something that functions at scale in the real world.

When a programme rewards the former and cannot see the latter, it creates a subtle but consequential pressure: to optimise for fundability rather than deployability. To build the story before the science is ready to support it. To raise the round before the evidence gate is cleared.

Most founders feel this pressure without being able to name it. It feels like the programme is pushing you toward something slightly wrong. Because it is. Not intentionally. Structurally.

The compounding consequence

Here is what makes the measurement gap more than a theoretical problem.

It does not stay static. It compounds.

Each reporting cycle under a measurement system that cannot see real outcomes produces institutional adaptations — in how cohorts are selected, in how escalation is handled, in how renewal cases are built — that make the gap harder to close in the next cycle.

Programmes begin to select for ventures that will look good within the programme window. Ventures that can show investor traction within twelve months. Ventures that can produce a compelling narrative before the technical validation is complete. Ventures that are, in short, optimised for the measurement system rather than for the market.

The technically strongest deep tech ventures — the ones working on problems that genuinely require seven, ten, fifteen years to solve — are progressively disadvantaged. They do not fit the window. They do not produce the signals. They do not get selected, or they get selected and then pressured into a timeline that does not fit their physics.

And the ecosystem wonders why European deep tech keeps stalling between TRL 4 and TRL 7.

It is not stalling because the science is weak.

It is stalling because the measurement system rewards the wrong signals at precisely the moment when the right signals are most expensive and most invisible.

What valid measurement would actually require

If the current system measures too early and too narrowly, what would a valid alternative look like?

Three things would need to change — not as ideals, but as structural requirements.

First: the definition of commercialisation needs to become stricter. Spinout formation is not commercialisation. A patent filing is not commercialisation. A seed round is not commercialisation. These are legitimate intermediate steps. They are not the end state. Any evaluation framework that treats them as outcomes is measuring the wrong thing. Hard thresholds — TRL 8/9, recurring revenues above a meaningful threshold, IPO, M&A, documented industrial deployment — are the only indicators that actually tell you whether science reached the market.

Second: the tracking horizon needs to extend beyond the programme. This is institutionally difficult. It requires cooperation from ventures after graduation, from later-stage investors, from industrial partners. It requires accepting that the programme’s contribution — if it exists — may not be visible for five to ten years. That is a hard position to hold when mandate periods are three years long. But without it, programme evaluation is structurally incapable of measuring what it claims to measure.

Third: the distinction between venture-level and institutionally-induced outcomes needs to become part of standard evaluation. When a venture fails after programme completion, the default attribution is founder execution or market conditions. Almost never is the question asked: did the programme’s own incentive architecture contribute to this outcome? Did the pressure to show investor traction before technical readiness set the venture up for the stall it hit two years later? This question is uncomfortable. It is also necessary. Because without it, the system cannot learn from its own failures — only from the ones it can see, which are not the ones that matter most.

The gap is not neutral

The measurement gap is not simply an absence of data. It is an active force.

It shapes which ventures get selected. It shapes what founders are incentivised to optimise for. It shapes how institutional resources are deployed. It shapes the renewal cases that funders receive and the policy decisions that follow from them.

A system that cannot see what it is producing does not remain neutral. It drifts — toward whatever it can see, which is activity, visibility, and short-horizon signals.

Europe is investing heavily in deep tech infrastructure. New programmes are being created. Existing ones are being scaled. Capital is being added.

But capital added to a system with a measurement gap does not close the gap. It scales it.

The question is not whether European deep tech has enough programmes.

The question is whether those programmes are capable of measuring the thing they were built to produce.

Right now, structurally, most of them are not.

FOR INSTITUTIONS

You have read the structural diagnosis. The next question is the one that actually matters for your organisation:

Does your current measurement architecture have this gap — and if so, where exactly is it building pressure?

The answer is not the same for every institution. It depends on your specific KPI configuration, your mandate horizon, and the interaction between your reporting logic and the development timescales of the ventures you support.

What follows is the operational layer: the concrete indicators that show where the measurement gap is producing structural over-demand in your programme right now — before it becomes visible in outcomes, and before it becomes a problem in your renewal discussion.

FOR FOUNDERS

You have felt the pressure this gap creates — even if you could not name it. The programme was pushing you toward something that felt slightly wrong.

What follows is the operational layer: how to recognise when a programme’s incentive architecture is creating pressure that misaligns with your actual development needs — and what that means for how you navigate the system without being reshaped by it.

Where the gap shows up operationally

The measurement gap does not announce itself. It produces specific, recognisable patterns — in institutional behaviour, in founder experience, and in portfolio outcomes. The following are the most consistent for founders and for institutions.

Read more

A programme cycle ends. The numbers look good. Follow-on funding secured. Investor meetings held. Spinouts created. Pitches delivered. Cohort completed.

And programme managers sit in the renewal discussion justifying that something is still off.

Not because the programme was poorly managed. Not because the team didn’t work hard. Not because the ventures were weak.

But because the numbers you can present don’t actually demonstrate what you were supposed to demonstrate.

You measured activity. But you were commissioned to produce commercialisation.

It is a structural problem. And almost no one has noticed — because the system that produces it is simultaneously the system that makes it invisible.

What programmes actually measure

If you look at what startup programmes supporting deep tech and science-based ventures actually report, the list looks roughly the same everywhere:

• Number of cohort participants

• Follow-on funding secured

• Investor meetings held

• Spinout entity creation

• Patent filings

• Pilot projects and letters of intent

• Media visibility

• Pitch event participation

These are legitimate indicators. They are measurable, documentable, and they demonstrate that something is happening.

What they do not demonstrate is whether what is happening leads to what the programme was actually built to produce.

A paper I recently published — Do Startup Programmes Actually Commercialise Science? A Preliminary European Analysis — examined exactly this.

Download here

The question was simple, but had rarely been asked this directly:

The answer was sobering.

Of 20 European deep tech companies meeting strict commercialisation thresholds, only 2 had documented accelerator participation. 13 had spinout or university-linked origins. The dominant commercialisation paths were strategic-partner-first (40%) and revenue-first (30%).

Not demo day. Not seed round. Not VC-first scaling.

The companies that actually reached the market largely did not get there through the logic that most programmes are built around.

Why this is not a coincidence

The confusion between activity and outcome is not the result of bad intentions. It is the result of a very specific structural logic — one that is worth understanding precisely, because understanding it is the only way to change it.

Programmes operate within fixed institutional parameters. Mandate periods of two to four years. Reporting cycles of three to twelve months. KPI frameworks that were, in most cases, inherited from consumer startup ecosystems or public grant reporting requirements — neither of which was designed for deep tech development timescales.

Here is the critical problem: deep tech does not mature on reporting cycles.

Technology readiness levels 1 through 4 routinely require three to seven years under optimal conditions. Regulatory validation in pharmaceuticals, medtech, and advanced materials adds two to five years on top. Industrial scale-up in capital-intensive sectors adds another four to ten years. The median time from founding to a hard commercialisation outcome in my pilot sample was 9.5 years. Several companies took fifteen to twenty-two years.

A programme with a three-year mandate and quarterly reporting cannot capture these outcomes. It ends before they exist.

So it measures what it can see. Which is activity. Which is not outcome.

This is not a failure of measurement discipline. It is a failure of measurement domain. The instruments are being applied outside the range where they are valid.

And here is the part that makes it structurally self-reinforcing: the same reporting system that cannot see real commercialisation outcomes can see activity very clearly. So activity gets rewarded. Activity gets optimised for. And over time, the programme becomes very good at producing exactly what it is measuring — while the outcomes it was supposedly built to produce remain unmeasured, and therefore invisible.

A venture can produce every positive signal on that list — funding, spinout, patents, pilots, investor meetings — and still fail to reach the market. The current measurement system cannot detect this divergence. It registers success while structural failure accumulates silently underneath.

The false positive problem

This produces something I call the systematic false positive.

A programme appears successful because its metrics are green. Reports go to oversight bodies showing activity within expected parameters. Renewal discussions proceed on the basis of those reports. Funding continues.

Meanwhile, the ventures that looked most promising in-programme stall at Series A two years later. The pilots never convert to deployment. The spinouts enter grant-dependent survival. The industrially strongest teams — the ones who needed long validation cycles, not pitch coaching — quietly leave the ecosystem.

None of this shows up in the programme’s own reporting. Because the reporting was never designed to capture it.

The false positive is not a lie. It is what happens when you apply a measurement system to a domain it was not calibrated for. The metrics are accurate. They are just measuring the wrong thing.

And because they look accurate, no one questions them.

Why it was rational

Before moving forward, something important needs to be said.

The institutions running these programmes are not naive. The people designing these KPI frameworks are not incompetent. In many cases, they are genuinely excellent at what they do.

The activity-outcome confusion persists because, within the constraints of a mandate period and a quarterly reporting cycle, measuring activity is rational. It is the only thing that is visible. It is the only thing that maps onto the accountability horizon of the programme.

Measuring hard commercialisation outcomes would require tracking ventures for ten to fifteen years after programme completion. It would require defining commercialisation more strictly than most evaluation frameworks currently do. It would require admitting that the programme’s contribution might not be visible within the programme’s own lifetime (more will be published soon).

That is a difficult institutional position to hold. Especially at renewal.

So the system optimises for what it can see. Which is activity. Which looks like progress. Until it doesn’t.

The question this leaves open

If activity metrics are structurally incapable of capturing commercialisation outcomes in deep tech — then what would valid measurement actually look like?

What are the indicators that predict long-horizon industrial deployment, not just near-term venture formation?

How do you distinguish between a programme that is genuinely building towards hard outcomes and one that is optimising for visible signals while structural failure compounds underneath?

And critically: how do you know which one you are running?

That is what the next section examines. Upgrade to keep reading for the solution:

Is your institution unknowingly part of this system?

Read more

There is a particular kind of institutional fear that does not look like fear from the outside.

It looks like confidence. It looks like programme updates, strategy papers, pilot announcements, partnership slides, funding dashboards, ecosystem maps, and carefully worded statements about momentum.

It sounds like: “We are well positioned.”

It sounds like: “The market is moving.”

It sounds like: “We are building capabilities.”

It sounds like: “The window is now.”

And for a while, all of that may be true.

The problem is that deep tech failure rarely announces itself at the moment when the structure begins to fail. In the early phase, the opposite often happens. Activity increases. Metrics improve. More people attend the meetings. More start-ups are counted. More pilots are launched. More capital is attracted. More strategic language appears around the programme. The institution looks active, credible, and aligned with the future.

Only later does the tone change.

A few years pass, and the questions become harder. Why did the pilots not become deployment? Why did follow-on funding not become industrial capability? Why did adoption not become sovereignty? Why did so many promising ventures stall after looking strong inside the programme? Why did the reporting show progress while the long-term outcomes remained weak?

That is the moment institutions should fear most: not the moment of visible failure, but the moment when they have to explain why failure was not visible earlier.

Quantum computing is now entering exactly that zone.

According to McKinsey’s Quantum Technology Monitor 2026, investment in quantum technology start-ups (yeah sure, still calling deep tech ‘startups’, but one problem at a time…) reached $12.6 billion in 2025, a 6.3-fold increase from the previous year. Quantum computing revenues exceeded $1 billion for the first time. The financing structure changed even more dramatically: public funding, which still represented roughly one-third of start-up investment in 2024, fell to 3% in 2025, while capital markets, including IPO proceeds, accounted for 44%.

The headline interpretation is obvious: quantum computing is crossing from research into market formation. The sector is no longer only a scientific promise. It is becoming a strategic management question.

But that is not the part that should worry institutions.

The more important shift is that quantum computing is moving from scientific uncertainty into institutional accountability. Once a field becomes commercially and geopolitically urgent, it is no longer enough to fund research, publish strategies, or announce partnerships. Institutions are expected to show results. Corporations are expected to demonstrate readiness. Public programmes are expected to prove impact. Funders are expected to justify allocation. Ecosystems are expected to show that years of investment created something durable.

That is where the real risk begins.

Because the moment a deep tech field becomes measurable, it also becomes distortable.

The Commercial Tipping Point Is Also a Measurement Tipping Point

Market inflection points create excitement. But inside institutions, they also create pressure.

Once quantum computing is described as a billion-dollar market, everyone around it must start proving relevance. Corporations need use cases. Governments need sovereignty narratives. Public-private programmes need impact evidence. Accelerators need ventures. Universities need spin-outs. Investors need exposure. Procurement bodies need pilots. Strategy teams need roadmaps. Ministries need defensible progress before the next review cycle.

None of this is irrational. In fact, this is precisely how institutional systems behave under accountability pressure. People do not wake up and decide to distort a strategic technology field. They respond to the signals they are judged on.

The danger is not that these signals are false. The danger is that they may be incomplete.

A pilot is not deployment. A spin-out is not commercialization. A funding round is not industrial readiness. A use case is not sovereignty. Adoption is not control. Visibility is not capability.

In software markets, the distance between activity and outcome can sometimes be short enough that these proxies are tolerable. In deep tech, that distance is often enormous. Quantum computing does not mature on reporting cycles. Semiconductor capacity does not appear because a policy cycle needs proof. Clean tech deployment does not become bankable because a pilot was announced. Biotech evidence does not accelerate because an investor narrative requires it.

The technology follows its own developmental physics. Institutions follow their own accountability clocks. When those clocks are misaligned, the institution begins to compress the technology into whatever can be measured in time.

That is how a strategic field can become very active before it becomes structurally capable.

Europe’s Quantum Position: Strength or Warning?

McKinsey’s regional picture is revealing. The United States leads quantum ventures investment, with 64% of total start-up investment in 2025. China remains strong in patents and scientific publications. Europe is described as leading in operational adoption, with European companies advancing a significant share of quantum computing use cases.

At first glance, this sounds encouraging. The US finances, China publishes and patents, Europe adopts. Each region appears to occupy a distinct strategic role.

But from a European perspective, this picture should create unease.

Adoption is not the same as sovereignty. A European company can become an excellent user of quantum systems without Europe controlling the hardware stack, the enabling intellectual property, the industrial supply chain, or the capital structures that decide where the next layer of value is captured. A public programme can support quantum readiness without creating durable European leverage. An ecosystem can become fluent in quantum language while remaining dependent on platforms, chips, infrastructure, and financing architectures shaped elsewhere.

This is not an argument against adoption. Adoption matters. It builds literacy, demand, operational familiarity, and early capability. But adoption becomes dangerous when it is mistaken for structural position.

A country can participate in a strategic technology domain without controlling the strategic layer of value. A company can use a technology without owning the infrastructure that determines its future dependence. A programme can report strong engagement while quietly building capability in the wrong part of the stack.

That is why Europe’s quantum position should not only be celebrated. It should be interrogated.

The European Commission’s Quantum Europe Strategy recognizes quantum as a strategic domain spanning quantum chips, computing, simulation, sensing, communication, skills, infrastructure, ventures, industry, and procurement. That breadth is exactly right. Quantum is not one market. It is an institutional system.

And institutional systems do not fail only because the science is weak. They fail when the structures used to govern them reward signals that are easier to report than the outcomes they claim to support.

The question for Europe is therefore not simply: “Are we active in quantum?”

The question is: “What kind of quantum position is our current activity actually building?”

That question is more uncomfortable, because it cannot be answered by counting pilots, partnerships, or funding announcements alone (Beyond Company Law: What EU Inc. Solves, And What Deep Tech Still Needs).

The Private Capital Shift Changes the Risk

For years, the standard critique of deep tech was that public funding supported early research, but private markets were too slow to engage. Quantum now appears to be moving beyond that phase. Private capital is entering. Capital markets are involved. IPO proceeds are part of the financing picture. The centre of gravity is shifting.

Many people will read this as validation. In one sense, it is. But capital does not merely fund a system. It changes the system.

A grant has one rhythm. A venture fund has another. A corporate balance sheet has another. A public market has another. Each capital form brings its own expectations about time, evidence, governance, reporting, risk, and progress. Public research funding may tolerate uncertainty for longer. Venture capital may seek milestone acceleration. Corporate capital may demand strategic fit. Public markets introduce visibility, valuation pressure, quarterly narratives, and comparability.

This matters because quantum technologies are still developing under deep scientific and engineering uncertainty. If the capital rhythm becomes faster than the technology rhythm, the sector may begin optimizing toward valuation narratives before the underlying evidence base and industrial deployment pathways are mature enough to carry them.

That does not make private capital bad. It makes capital timing non-neutral.

In deep tech, the order of actions changes the outcome (as outline in the Deep Tech Playbook). Evidence before scale is not the same as scale before evidence. Industrial validation before capital-market exposure is not the same as capital-market exposure before industrial validation. Adoption before capability formation is not the same as capability formation before adoption.

The same activities, sequenced differently, can produce radically different futures.

This is the point most market reports cannot capture. They can tell us that money is moving. They cannot tell us whether the structure receiving that money is capable of turning it into durable industrial outcomes.

The Real Problem Is Not Failure. It Is False Confidence Before Failure.

The most dangerous institutional condition is not visible underperformance. Visible underperformance can be addressed. It can be discussed, explained, corrected, sometimes even forgiven.

The more dangerous condition is false confidence.

False confidence appears when the dashboard is populated, the programme is busy, the metrics are defensible, and yet the system is quietly optimizing for signals that do not determine long-term survival. It appears when institutions can prove activity but not structural progression. It appears when reporting creates comfort precisely because it cannot see the layer where risk is accumulating.

That is why this problem is so psychologically difficult for institutional leaders. They are not afraid of doing nothing. They should become afraid of doing many things correctly, only to discover later that the structure itself was producing the wrong outcome.

Stakeholders should aim for standing in front of a board, ministry, funder, or oversight body and being capable of answering the question: “Why did no one see this earlier?”

That question is brutal because it is rarely answered well by pointing to effort. Effort is not the issue. Competence is not always the issue. The issue is whether the programme had a way to evaluate what its own measurement architecture was incentivizing over time.

This is the missing layer in most deep tech discussions.

We ask whether founders are strong enough, whether capital is available enough, whether markets are ready enough, whether technologies are mature enough. But we ask far less often whether the institutions around those technologies are structurally capable of distinguishing real progress from progress-shaped activity.

Quantum computing now forces that question into the open.

If deep tech ecosystems are structurally optimized for visible short-term activity rather than long-term deployment survivability, then the real question is no longer whether individual ventures fail. The real question becomes:

What kinds of commercialization failures are current systems statistically incentivized to produce, even when everyone inside them is acting rationally?

That is where market commentary ends and structural diagnosis begins.

Before We Continue

The public argument is this: quantum computing’s commercial inflection point is also an institutional measurement test. The sector is attracting capital, policy attention, corporate adoption, and strategic urgency. But unless institutions can distinguish activity from structural position, they may reproduce the same failure patterns that have weakened deep tech commercialization in other domains.

The next part is where this becomes operational.

I look at what happens when strategic urgency collides with industrial reality, why Northvolt is relevant to the emerging quantum hype, how the same structural pattern appears in semiconductors, AI infrastructure, biotech, clean tech, and advanced manufacturing, and which diagnostic questions institutions need to ask before scaling programmes, pilots, and funding vehicles.

If you are responsible for a programme, portfolio, policy instrument, innovation unit, TTO, or strategic deep tech initiative, this is the part where the question stops being abstract.

It becomes: what is your current structure setting you up to defend two cycles from now?

When Strategic Urgency Becomes Institutional Exposure

The danger becomes easier to understand when we step outside quantum for a moment.

Read more

AI labs are starting to buy the digital remains of failed startups. That should make science ventures ask a better question: can data become a legitimate income stream before VC, before distress, and before insolvency?

A strange new market is becoming increasingly visible.

According to a recent Golem report, insolvent startups are beginning to sell internal company data — Slack messages, Jira tickets, emails, code repositories, Google Drive folders — as training material for AI systems. The example that caught my attention was Cielo24, a transcription startup that reportedly generated a six-figure amount by selling its internal digital archive.

At first glance, this sounds like another late-stage symptom of the AI data hunger: companies fail, their assets are liquidated, and now even their internal work noise becomes monetizable.

But I think the more important question is not what happens after insolvency.

The more important question is this:

My take is: yes, but…

Not by selling Slack archives.

Not by dumping emails.

Not by treating employee communication, lab notes, customer files, or operational mess as an emergency asset.

The real opportunity is much more strategic: science ventures can start treating their evidence, failure modes, process data, experimental histories, validation traces, and technical decision pathways as structured data assets from day one.

Not as exhaust.

As architecture.

The AI data market has changed

For years, AI scaling was powered by the public internet. Crawlable websites, books, forums, code repositories, academic papers, social media, image archives — the internet became the great data substrate of modern machine learning.

But that phase is no longer infinite.

Ilya Sutskever, former chief scientist of OpenAI, warned in late 2024 that the industry had effectively reached “peak data” in the sense that the high-quality public internet data available for training was becoming exhausted. As reported by Observer, the implication was simple: if models are to continue improving, especially beyond text prediction, they need new forms of data.

And this is where the market becomes interesting.

The next generation of AI is not only about chatbots producing fluent language. It is about agentic systems: AI agents that can reason through workflows, operate tools, debug code, manage processes, assist in laboratories, navigate compliance, support R&D, and coordinate complex work.

For that, the public internet is insufficient.

The public internet contains descriptions of work. It does not contain the full anatomy of work.

It does not usually contain the Jira ticket, the Slack debate, the pull request, the failed deployment, the customer complaint, the root-cause analysis, the next sprint decision, and the internal reasoning that connected all of it.

That is why the digital remains of startups suddenly matter. They contain something the internet does not: real operational sequences.

SimpleClosure, which has launched an Asset Hub to help startups monetize assets left behind during shutdowns, is part of this emerging market. The premise is clear: a company may die, but its code, data, domains, IP, internal archives, and operational history may still have value.

For AI labs, this value lies in the realism of the data. Internal workspaces contain exactly the messy, contextual, multi-step traces that AI agents need in order to learn how work actually happens.

That is the surface story.

But for science ventures, there is a much deeper implication.

Science ventures produce better data than startups

Startups produce operational data: customer support tickets, product analytics, sales calls, project management traces, marketing experiments, code decisions.

Science ventures produce something more valuable.

They produce evidence.

They produce failed experiments. They produce process windows. They produce instrument readings, lab protocols, quality deviations, formulation iterations, stability measurements, scale-up failures, regulatory interpretations, experimental dead ends, and the tacit logic by which scientists decide what to try next.

This kind of data is rare because it is expensive to generate. It is usually not public. It is often not published. And in many cases, it is exactly the kind of information that prevents other teams from repeating the same mistakes.

In deep tech, the most valuable data is often not the triumphant result. It is the negative result that explains why something did not work.

A materials company may know which synthesis parameters repeatedly destabilize a compound. A biotech company may have hundreds of failed assay conditions. A medtech startup may have detailed traces of why a prototype behaved differently under clinical workflow conditions than in the lab. A battery company may have degradation patterns that never make it into a publication. A robotics company may have edge-case failure logs from real industrial environments.

These are not simply “data points.”

They are compressed technical experience.

And compressed technical experience is exactly what AI systems, industrial partners, simulation providers, and vertical AI companies increasingly need.

The question is not whether science ventures have valuable data.

The question is whether they have structured it, governed it, and retained the rights to monetize it.

Most have not.

The wrong version: selling digital exhaust

The most dangerous interpretation of the current market would be this:

“If failed startups can sell Slack and Jira data, maybe science ventures should sell their internal data too.”

That is the wrong lesson.

Selling raw internal communication is legally, ethically, and strategically risky. Employee messages may contain personal data, confidential opinions, health information, trade secrets, customer details, partner information, and sensitive IP. Even if a company technically owns its systems, that does not automatically mean it has a clean right to sell private workplace communication to third parties for AI training.

The risks are not theoretical.

The European Data Protection Board has made clear in its Opinion 28/2024 on AI models and GDPR that questions of anonymization, lawful basis, legitimate interest, and downstream use remain central when personal data is used in AI development. The UK’s ICO has similarly emphasized that using data for generative AI training requires a real legal basis and cannot simply be justified by technological ambition, as explained in its work on web scraping and lawful basis for AI training.

The EU AI Act adds another layer. For high-risk AI systems, Article 10 requires appropriate data governance and data management practices for training, validation, and testing data. The EU has also introduced transparency obligations for general-purpose AI models, including summaries of training content and copyright-related requirements, as outlined by the European Commission in its GPAI documentation.

There is also a technical problem: training data can sometimes be memorized and extracted. Carlini et al. demonstrated in Extracting Training Data from Large Language Models that large language models can reproduce parts of their training data, including sensitive or unique sequences. Later work has shown similar risks for production models.

So the naive version — “sell the archive” — is not a strategy.

It is a liability event waiting to happen.

The right version: build data assets by design

The real opportunity for science ventures is not to sell raw internal data.

The opportunity is to build data products.

That means designing the venture from the beginning so that its development process generates reusable, licensable, legally clean data assets alongside its core technology.

This is a very different mindset.

A science venture does not only ask:

“What product are we building?”

It also asks:

• “What evidence are we generating?”

• “What failure modes are we documenting?”

• “What process data are we creating?”

• “Which parts of this data can be abstracted without exposing trade secrets?”

• “Which rights do we need in our customer, research, employee, contractor, and collaboration agreements?”

• “Could this become a benchmark dataset, an evaluation environment, a synthetic data product, or a licensed technical data feed?”

This is where the strategic shift happens.

The data is no longer a byproduct.

It becomes part of the capital architecture.

Why this matters before VC

Deep tech and science ventures have a structural financing problem. They often need years of technical development before the core product is ready for scalable revenue. The valley between scientific promise and industrial deployment remains one of the most persistent failure zones in the innovation economy.

The European Deep Tech Report by Dealroom has shown that deep tech ventures require substantially more time and capital than conventional software startups to reach comparable revenue milestones. The Royal Academy of Engineering’s State of UK Deep Tech similarly emphasizes the difficulty of scaling deep tech due to capital intensity, infrastructure needs, regulatory complexity, and long development cycles.

This creates the familiar trap.

A science venture has valuable technology, but not enough evidence for strategic buyers. It has promising data, but not enough revenue for conventional investors. It has technical depth, but not enough market proof. It has long-term industrial value, but short-term financing pressure.

The usual answer is: raise venture capital.

But venture capital is often badly timed for science ventures at early TRLs, as explained in the Beyond Venture Capital – Industry Report. Equity may enter before the evidence base is strong enough to support valuation, before the industrial path is clear, and before the company knows whether it is building a venture-scale company, a licensing engine, a strategic acquisition candidate, or an industrial technology supplier.

This is why generating revenue matters (among others…).

Grants matter. R&D-as-a-service matters. Paid pilots matter. Strategic co-development matters.

And now, potentially, data licensing matters too.

Not as a replacement for all other capital.

But as an additional layer.

A science venture that can generate €25k, €50k, €100k, or more through licensed datasets, benchmark environments, or technical evaluation products may buy itself something extremely valuable: time without dilution.

And in deep tech, time without dilution is often the difference between premature collapse and real de-risking.

Training data is not the only product

One mistake would be to assume that the only monetizable form is “training data.”

For science ventures, the more defensible and often more valuable product may be evaluation.

AI companies do not only need data to train models. They also need ways to test whether models work in specific domains.

• Can an AI agent follow a laboratory SOP without making unsafe assumptions?

• Can it interpret a noisy instrument trace?

• Can it identify when a material characterization result is inconsistent?

• Can it distinguish between a regulatory documentation gap and a technical failure?

• Can it reason through a design-of-experiments workflow?

• Can it assist in quality deviation analysis?

• Can it support a scientist without hallucinating a protocol?

These questions require evaluation environments, not just training corpora.

A science venture could therefore build a benchmark dataset or private sandbox that tests whether AI systems can operate in a specific technical domain. That sandbox might be more valuable than the underlying raw data because it preserves control, reduces IP leakage, and turns the venture’s domain expertise into a recurring infrastructure product.

Instead of selling “our lab data,” the company sells:

1. access to a validated materials failure benchmark,

2. a synthetic but scientifically grounded assay-design environment,

3. a QA deviation reasoning dataset,

4. a regulatory workflow evaluation suite,

5. a process-optimization simulation dataset,

6. or a private clean-room API that lets AI companies test models against real-world technical constraints without downloading sensitive data.

This is a better business.

It is also a safer one.

The data rights problem must be solved early

The biggest blocker will not be technical.

It will be contractual.

Most science ventures do not structure data rights properly in the beginning. They sign research collaborations, customer pilots, grant agreements, university IP arrangements, CRO contracts, lab service contracts, and employment agreements without a clear view of future data monetization.

Later, when the data becomes valuable, nobody knows who has the right to use what.

The customer owns some of it. The university claims some of it. The employee generated some of it. The CRO processed some of it. The grant agreement restricts some of it. The partner considers some of it confidential. The dataset contains personal data. The lab notebook contains trade secrets. The metadata reveals process know-how.

At that point, monetization becomes almost impossible without legal risk.

So if science ventures want data to become a real revenue stream, they need data rights by design.

This means that early agreements must distinguish between raw data, derived data, aggregated data, anonymized data, synthetic data, customer-specific confidential data, generalizable process knowledge, and reusable benchmarking structures.

A customer may own its project-specific result.

– But the venture may retain the right to use generalized, anonymized, non-identifying process patterns.

A partner may restrict disclosure of its proprietary use case.

– But the venture may retain the right to generate synthetic training examples based on abstracted technical classes.

An employee may generate internal notes.

– But the company must still handle personal communication responsibly and should not treat private workplace messages as a commercial dataset unless this was clearly and lawfully structured from the beginning.

This is not administrative detail.

What makes science data valuable?

Not all data is valuable. Most data is noise. Valuable data has structure, scarcity, provenance, and a buyer use case.

In science ventures, the strongest data assets usually have four characteristics.

First, they are hard to generate. If a dataset requires specialized equipment, long-duration experiments, regulated environments, rare failure cases, or expensive technical expertise, it may have real market value.

Second, they are linked across the workflow. A measurement is useful. But a measurement connected to the experimental setup, the parameter history, the interpretation, the failure, and the next decision is far more valuable.

Third, they are documented. AI buyers do not want mystery data. They want provenance, metadata, quality labels, limitations, versioning, and traceability.

Fourth, they are legally usable. A messy dataset with unclear rights is not an asset. It is a risk.

This is why the insolvency data market is only the crude early signal.

The mature market will not reward companies for having a chaotic archive. It will reward companies for having high-integrity data assets.

A new revenue layer for science ventures

If we translate this into a practical financing architecture, data monetization sits between grants, R&D services, pilots, and strategic partnerships.

It is not identical to R&D-as-a-service because the output is not just a service result. It is a reusable asset.

It is not identical to licensing core IP because the venture does not necessarily give away its invention. It licenses structured evidence, evaluation infrastructure, or abstracted technical learning.

It is not identical to venture capital because it does not require equity.

It is not identical to customer revenue because the buyer may not be the future product customer. The buyer may be an AI lab, an industrial AI provider, a simulation company, a digital lab platform, a regulatory technology company, or a sector-specific model developer.

This gives science ventures a new question to ask before raising capital:

What data asset are we unintentionally building — and could it be intentionally monetized without damaging the core company?

That question should become part of early venture architecture.

But it will not work for everyone

This is not a universal hack.

Some science ventures should not monetize data. If the data exposes core trade secrets, contains sensitive personal information, depends on customer confidentiality, lacks clean rights, or would weaken future IP positioning, then selling or licensing it may destroy more value than it creates.

Some ventures also simply do not generate data that others would pay for. Data volume alone is not enough. The data must be rare, structured, interpretable, and attached to a real use case.

And some ventures are better off using their data internally to strengthen defensibility, improve technical performance, support regulatory submissions, or increase valuation later.

Data monetization is not automatically good.

It is only good if it strengthens the venture’s trajectory.

In operator terms: data monetization must be sequenced correctly. If it enters too early, before rights and sensitivity are understood, it creates risk. If it enters after the evidence base is structured and the reusable layer is separated from the core IP, it can create non-dilutive leverage.

The new question for founders

The founder question should no longer be only:

“How do we finance the next milestone?”

It should also be:

“What evidence are we generating on the way there, who else needs it, and can we monetize it without selling the company’s future?”

That is a very different question from the default startup script.

It pushes science ventures to think architecturally.

• A materials startup may realize that its degradation data is useful to AI companies working on predictive materials design.

• A lab automation venture may realize that its workflow traces could become an evaluation suite for scientific agents.

• A medtech company may realize that its documentation workflows could train compliance-support tools — if properly abstracted and anonymized.

• A biotech platform may realize that its negative assay results are commercially meaningful because they help others avoid expensive dead ends.

• A robotics company may realize that its failure logs from industrial edge cases are worth more as a benchmark than as a forgotten internal archive.

This does not mean every science venture becomes a data company.

It means every science venture should understand whether it is accidentally creating one.

The deeper strategic shift

The Golem article describes distressed monetization: companies selling internal data after they fail.

But the strategic version is pre-distress monetization.

Not: “What can we sell when we die?”

But: “What assets are we building while we live?”

That distinction matters.

A failed startup sells digital remains. A well-designed science venture licenses structured learning.

One is liquidation. The other is architecture.

The first happens when the company runs out of options.

The second creates options.

And this is precisely why the topic matters for deep tech. Science ventures are constantly told they need to raise more capital earlier. But often times the more intelligent move is not to raise earlier — or at all. It is to identify which non-dilutive assets the venture is already generating and turn them into structured, defensible revenue.

In AI times, evidence itself becomes an asset class.

Not all evidence. Not raw chaos. Not private communication. Not confidential customer data.

But curated, rights-cleared, domain-specific, technically meaningful evidence.

That may become one of the most important new financing layers for science ventures.

Final thought

The current market for startup data after insolvency is a warning signal and an opportunity signal at the same time.

It warns us that internal company data is becoming monetizable in ways most teams never anticipated.

It warns us that privacy, employment contracts, IP boundaries, and data governance will become strategic issues, not legal afterthoughts.

But it also reveals something powerful:

Science ventures may already be generating valuable AI-era assets long before their products reach the market.

The question is whether they treat those assets as accidental residue — or design them into the company from the beginning.

My thesis is simple:

The value is not in the digital leftovers. The value is in the structured evidence of how science actually becomes technology.

Europe is mobilising historically large amounts of public and publicly catalysed capital for venture capital and growth investing.

The reflex is understandable.

Europe has too few large technology companies, too few scaleups, too few unicorns, and too little late-stage capital compared with the United States.

But the decisive question is not:

Does Europe need more capital?

The better question is:

For which type of company, at which maturity point, against which bottleneck — and through which market logic?

Because if the problem is miscategorised, more capital does not scale the solution.

It scales the mismatch.

Europe’s New Capital Offensive

A recent article by The Next Web (TNW) summarises the scale of the current push well1: Europe is moving more public and publicly mobilised capital into VC and growth structures. The programmes include:

- EIF / ETCI 2.0: a major fund-of-funds initiative intended to mobilise scaleup capital. The EIF describes ETCI 2.0 as the second generation of the European Tech Champions Initiative, designed to mobilise a multiple of the first ETCI volume.2 3

- Germany / WIN Initiative: an expected total impact of around €12 billion by 2030 through KfW and private partners.4

- France / Tibi Initiative: institutional capital mobilisation for French and European technology funds, with several billion euros in commitments.5

- Scaleup Europe Fund: a new European growth fund for strategic technology fields such as AI, quantum, biotech, and semiconductors.6 7

- European Innovation Council (EIC): a €10 billion programme through 2027 with a clear deep tech orientation.8

The rationale is obvious. Europe has a growth funding problem. The European Commission itself describes the gap between Europe and the US in later-stage financing very clearly.9

So the issue is not that this capital offensive comes out of nowhere.

It responds to a real weakness.

But it primarily addresses the most visible symptom:

capital scarcity.

Not necessarily the structural causes.

The Category Error: A Startup Is Not Just a Startup

The TNW article discusses startups, scaleups, VC, growth capital, and strategic technologies. In several places, it clearly points toward deep tech — but without cleanly naming the category.

That may sound semantic.

It is not.

They follow a different development logic.

Software startups can often grow through rapid iteration, early customer signals, scalable distribution, and relatively reversible product decisions.

Deep tech companies, by contrast, often first need to prove that their technology is:

• scientifically reproducible,

• industrially scalable,

• regulatorily viable,

• compatible with real value chains,

• functional under real market conditions,

• and adoptable by institutional or industrial customers.

These are different bottlenecks.

A SaaS company can often use growth capital to move faster into the market.

A deep tech company can be pushed into the wrong sequence by premature growth capital: more visibility, more fundraising, more scaling pressure — before evidence, industrialisation, and adoption are mature enough.

I have argued elsewhere that science innovation needs to be defined as a distinct venture category, because otherwise it is systematically evaluated, financed, and supported through instruments designed for a different class of companies.10

That is the core point here:

More VC Does Not Automatically Produce Deep Tech Scale

This is not an anti-VC argument.

Venture capital is a useful instrument for certain companies, at certain points in time, with certain scaling profiles.

But VC only solves the problems VC is built to solve.

VC is particularly effective for companies with high growth potential, clear scalability, large markets, and a plausible exit path within a fund lifecycle.

Deep tech often violates these assumptions by default:

• development cycles are longer,

• technical risks remain open for longer,

• industrialisation is capital intensive,

• regulation cannot be skipped,

• markets often need to be formed rather than entered,

• customers are frequently industrial or public institutions, not fast-moving consumer markets.

Official and institutional deep tech analyses repeatedly emphasise these differences: deep tech is typically characterised by longer development phases, higher capital intensity, and higher technological risk.11 12

So the decisive question is not:

How do we mobilise more capital?

It is:

When does which type of capital become effective — and when does it simply reinforce structural misalignment?

Because capital is not neutral.

Capital changes behaviour.

Research & Frameworks

When European deep tech companies are pushed too early into VC logic, they optimise for what investors and programmes can see quickly:

• pitchability,

• follow-on funding,

• visibility,

• market narrative,

• investor readiness,

• demo-readiness.

But deep tech does not scale on reporting cycles.

Deep tech scales on scientific, industrial, and regulatory maturity cycles.

The Real Problem: Europe Has Not Only a Funding Gap, but a Translation Gap

The dominant frame is:

Europe has too little growth capital.

That is partially true.

But beneath it sits a deeper bottleneck:

Or more simply:

Europe produces research, prototypes, and programmes.

But it is weak at translating new technologies into validated demand under real market and industrial conditions.

This is especially visible in deep tech.

The EIC reported in 2024 that its supported companies had reached an overall portfolio value of nearly €70 billion across hundreds of deep tech companies. That is impressive.13

But portfolio value is not the same thing as broad industrial scale.

The decisive question is:

How many of these companies achieve real industrial adoption, repeatable revenues, regulatory viability, and robust customer integration?

That is a different question from: How much capital is available?

Capital can accelerate growth but it never replaces market validation.

→ It doesn’t replace procurement.

→ It doesn’t replace industrial integration.

→ It doesn’t replace regulatory clarity.

→ It doesn’t replace demand.

The Political Temptation: Financing What Is Measurable Instead of Building What Is Difficult

A large fund is politically attractive.

It creates numbers that are easy to communicate: €15 billion in fund size, €80 billion in potential mobilisation, 100 funds, strategic technologies, European champions.

It is visible. It is legible. It is institutionally clean.

But the harder work sits elsewhere.

It lies in building the less glamorous transition layer between scientific possibility and industrial adoption: innovation procurement, industrial testbeds, regulatory sandboxes, SME-driven venture clienting, co-development structures, real first-customer programmes, pan-European validation infrastructure, and harmonised market access.

These mechanisms do not produce the same headline as a new billion-euro fund.

They are slower, more operational, and harder to package politically.

But they are often where the fate of deep tech is decided.

The EU already recognises this lever. Pre-commercial procurement and innovation procurement are explicitly designed to help develop innovative solutions by involving public demand earlier and more structurally.14

That matters because it means the alternative to a pure fund logic already exists institutionally. It is just not the dominant narrative.

What If Part of the Money Were Used Differently?

A better research question is not:

Does Europe need more VC?

It is:

What share of public money should go into capital supply — and what share should go into demand, validation, and adoption infrastructure?

Instead of only capitalising funds, Europe could more aggressively finance the conditions under which:

- SMEs and Mittelstand companies test new technologies under real conditions,

- industrial customers do not have to bear first-pilot risk alone,

- public procurement validates innovation earlier,

- deep tech companies generate real deployment data,

- venture clienting becomes a systematic market bridge,

- co-development between industry and young technology companies is funded structurally.

This would not be less market-oriented.

Because it would not only ask investors whether a company is fundable but also ask customers, integrators, and procurers whether a technology is useful, integrable, and worth paying for under real conditions.

That is the difference between fundability and deployability.

Europe too often optimises for fundability.

Deep tech needs deployability.

Venture Clienting Instead of Only Venture Capital

Venture clienting is particularly relevant in this context because it shifts the logic.

Traditional VC asks:

Who invests in the company?

Venture clienting asks:

Who buys, tests, integrates, or validates the solution?

A venture client is not primarily an investor, but an early customer. The company receives not symbolic support, but real application, real requirements, and real market signals.15 16

For Europe, this is especially relevant because Europe’s structural strength does not lie only in financial markets.

It lies in:

• industry,

• Mittelstand,

• specialised engineering,

• regulated value chains,

• technical procurement,

• long-term B2B relationships,

• application-oriented engineering competence.

These strengths are underused in a VC-centric debate.

If Europe tries to copy the US merely by mobilising more VC, it ignores its own industrial advantages.

Europe should not only ask:

How do we build more European VCs?

It should also ask:

How do we make European industry, Mittelstand, and public procurement better first customers for new technology?

Crowding Out: The Wrong Question

The debate around public capital often focuses on crowding out:

Does public capital displace private capital?

That is a legitimate question. The Jacques Delors Centre, in its analysis of the Scaleup Europe Fund, explicitly notes that design matters and that market distortions need to be avoided.17

But for deep tech, there is a second and more important crowding-out question:

Does a VC-centred funding logic crowd out alternative commercialisation architectures?

Because not every deep tech company should grow through the same path.

Some need VC.

Majority needs:

- licensing models,

- industrial co-development partners,

- venture clienting,

- public first procurement,

- project finance,

- strategic minority investments,

- R&D-as-a-service,

- long-term integration into value chains.

If funding programmes, KPIs, and public narratives treat VC and scaleup funding as the dominant success signal, other viable paths become invisible.

Then capital is not the only thing being misallocated.

The Real Risk: Europe Scales the Wrong Metric

If follow-on funding, fund size, and company valuation become the dominant signals of success, the system will optimise for exactly those signals.

In the short term, that looks attractive. Capital is visible. Fundraising rounds are easy to report. Rising valuations create momentum. Larger funds signal confidence.

But in the long term, this can become dangerous, especially for deep tech.

Deep tech requires different forms of proof. It needs evidence, reproducibility, technical validation, regulatory viability, industrial scalability, real customer integration, buyer-side budget readiness, and deployment under real operating conditions.

These are harder to measure than capital raised. They take longer to produce. They are less convenient for headlines and programme reporting.

But they are what determine whether a technology can actually move from research into industrial use.

If Europe primarily measures capital volume, Europe will produce capital volume.

But capital volume is not the same as technological sovereignty.

And it is certainly not the same as industrial deployment capability.

What a Better European Approach Would Look Like

A better approach would not abolish VC.

It would contextualise VC.

Europe does need more growth capital.

But in parallel, Europe needs:

1. Clean venture categories

Software, high tech, and deep tech should not be forced into the same startup logic.

2. Deep-tech-specific maturity and validation logic

Capital should not be tied only to growth signals, but to evidence, industrial, and adoption maturity.

3. Venture clienting and industrial first-customer programmes

Mittelstand, large industry, and public institutions need to be systematically enabled to test and integrate new technologies early.

4. Innovation procurement

Public demand should not only appear at the end of the commercialisation pathway. It should become part of the validation infrastructure earlier.

5. Regulatory and market harmonisation

A fragmented internal market cannot be repaired by fund volume alone.

6. Programmes that measure deployability, not only fundability

The central question should not only be:

Who raised capital?

It should be:

Who proved, under real conditions, that the technology works and is needed?

The Uncomfortable Conclusion

Europe is putting a lot of money behind venture capital and scaleups.

That is not wrong. Europe does need more growth capital, larger funds, and stronger financing capacity for companies that are ready to scale.

But it is incomplete.

If Europe defines the problem primarily as capital scarcity, the answer will be to mobilise more capital. That is exactly what is happening now. But if Europe defines the problem as a deep tech translation problem, the answer looks very different.

Then the missing infrastructure is not only larger VC funds. It is validation infrastructure, demand creation, procurement pathways, industrial integration, regulatory viability, and access to real market conditions.

That distinction matters.

Europe’s deep tech problem does not begin only at the point where no growth fund is available. It begins much earlier: when technologies are not categorised correctly; when programmes reward the wrong signals; when fundraising is treated as proof of progress; when industrial reality only enters after the capital round; and when demand is not built, only capital supply.

More money can accelerate a system. But if the system is miscalibrated, more money also accelerates the misdirection.

So the real question is not whether Europe can mobilise €80 billion for venture capital.

The real question is whether Europe can prevent those €80 billion from scaling the same structural mismatch that has kept deep tech stuck between research, funding, and industrial deployment.

That is the real bet.

Subscribe now

One end of the system is tightening. The other is leaking.

That is not a coincidence.

In April 2026, ERC President Maria Leptin explained why the European Research Council will tighten application rules from the 2027 calls onward1: review pressure has grown sharply, while the capacity of the panels that evaluate applications remains structurally limited. The ERC now runs 91 peer-review panels, each meeting twice a year in Brussels. Where panels once handled roughly 50 to 150 applications, some now face more than 250. According to Leptin, panel size and meeting duration cannot simply be expanded indefinitely, and prior optimisation efforts have already been exhausted. The result: the ERC will extend resubmission restrictions by one year for unsuccessful applicants in the 2027 calls, and similar restrictions may also apply in 2028 and 2029. Her message to applicants was explicit: do not submit automatically. Consider whether your proposal is truly ready, because an unsuccessful attempt may now carry a longer penalty.

At first glance, this looks like a classic story of overload: too many proposals, not enough review capacity, stricter access rules.

But another story is unfolding at the opposite end of the system.

A recent t3n piece highlighted a different problem2: publicly funded research projects often launch with polished communication, dedicated project websites, and high-visibility announcements, yet after completion their results become surprisingly difficult to locate. The article points to projects such as Solifly and Matisse, both related to batteries for electric aviation. In the Solifly case, the article reports €1.35 million in public funding yet says no publicly accessible final document could be found; in the Matisse case, outputs appeared scattered across conference PDFs without a clear named contact point. The article’s broader critique is not just poor communication. It is that publicly funded knowledge is too often inadequately consolidated, exposed, or maintained once the funded project period ends.

These two phenomena should not be treated as separate failures.

They are, I would argue, two expressions of the same structural asymmetry.

The underlying structural question

The hypothesis

The answer is not simply “too little money” or “too much bureaucracy.”

The deeper problem is this:

Public research funding systems are becoming highly optimised for application selection and allocation under pressure, but remain comparatively weak at ensuring result retention, public visibility, and long-term knowledge yield.

In other words: the system is growing stricter where access is concerned, but not equally strict where public knowledge preservation is concerned.

That asymmetry matters. Because once access becomes more competitive and reputationally loaded, actors rationally adapt. They become better at navigating calls, formatting proposals, timing applications, and aligning with selection logics. But those capabilities are not identical to robust post-project dissemination, durable public knowledge storage, or long-term research usability.

This is where the concept of grantpreneurship becomes relevant.

What the ERC case really shows

The ERC’s 2027 rule change is not merely an administrative adjustment. It is a window into the internal stress of the system.

Maria Leptin’s explanation is revealing precisely because it is so direct. The problem is not that peer review suddenly became philosophically stricter. The problem is that demand has risen to a level that the existing review architecture can no longer comfortably absorb. Panels are overloaded. Capacity is finite. Further operational optimisation has, by the ERC’s own account, largely run its course. So the system is shifting from simply evaluating proposals to actively managing review load by restricting repeated access.

That matters for two reasons.

First, it increases the strategic cost of a weak or premature submission. If unsuccessful applications now trigger longer re-entry restrictions, proposal timing becomes more consequential. Applicants are encouraged to wait until their ideas are more mature. That may improve average proposal quality. But it also means the burden of system overload is being redistributed: what was once mostly absorbed by the review machinery is now more heavily pushed back onto researchers.

Second, the change reveals how valuable access itself has become. In a highly competitive funding system, the ability to produce an eligible, well-timed, review-robust proposal becomes a specialised skill. Once that happens, “fundability” starts to solidify as a competence in its own right.

This is not a moral criticism of applicants. It is a structural observation.

The stricter the front end becomes, the stronger the incentive to optimise for it.

What the disappearing-results problem really shows

Now turn to the back end.

The t3n article is useful because it names something many researchers, journalists, and programme observers already recognise intuitively: public research communication is often strongest at project launch, weaker during implementation, and weakest after the money stops. The pattern it describes is familiar:

1. Project start with polished messaging and institutional visibility

2. A few announcements or event notices during the funding period

3. After project completion, little sustained visibility of findings, reports, lessons learned, or even basic traceability of outputs.

This is not just frustrating. It is a structural waste.

And it becomes more serious when placed next to the EU’s own formal expectations. The European Commission states plainly that beneficiaries of the EU’s research and innovation programmes are legally obliged to disseminate and exploit results. It frames dissemination and exploitation not as optional PR, but as part of increasing the impact of funded research.3 The European Research Executive Agency says the same in operational terms: Horizon Europe beneficiaries have a legal obligation to carry out activities that increase the impact of project results through dissemination and exploitation.4

So the issue is not that Europe lacks a normative framework for public visibility of results.

The issue is that, in practice, the system appears more forceful and disciplined in deciding **who gets funded** than in ensuring that the resulting knowledge remains publicly visible, reusable, and economically or scientifically meaningful afterwards.

That is the asymmetry.

Grantpreneurship as the connecting mechanism

This is where grantpreneurship should be understood properly.

The term is often used dismissively, as if it referred to individuals gaming the system. But that is too shallow. In this context, grantpreneurship is better understood as a rational adaptation to an environment where funding access is scarce, prestigious, procedurally demanding, and institutionally consequential.

In such an environment, actors become highly skilled at:

• reading calls

• designing proposals around evaluator expectations

• structuring consortia

• producing “impact” language

• aligning narrative with programme logic

• timing submissions strategically

None of this is irrational. None of it is even necessarily bad.

It becomes problematic only when the system rewards those front-end capabilities more consistently than it rewards or enforces back-end knowledge retention, result visibility, and public learning.

That is exactly the pattern we are beginning to see:

• front-end pressure rises

• application competence professionalises

• entry becomes more selective

• but result exposure remains patchy, inconsistent, or weakly institutionalised

In that environment, grantpreneurship is not an anomaly. It is one of the system’s predictable products.

The deeper asymmetry

The real issue, then, is not simply review overload or weak dissemination. It is the architecture connecting them.

At the front end, the system is:

• highly structured

• competitive

• reputationally loaded

• resource-intensive

• increasingly punitive toward mistimed or weak applications

At the back end, the system is too often:

• weakly standardised in public output retention

• under-resourced after project close

• tolerant of fragmented visibility

• inconsistent in how final public knowledge is consolidated

• much less visibly enforced than application discipline or grant compliance

This means public research funding is becoming increasingly strict where entry is concerned, while remaining comparatively weak where public knowledge retention is concerned.

Put more sharply:

We are becoming better at filtering proposals than at retaining the public knowledge those proposals were funded to generate.

That should worry us.

Because a system can be both highly competitive and surprisingly weak at learning.

And broader system can be effective in activity but limited in making progress:

Research & Frameworks

Why this matters

There are at least three consequences.

1. Knowledge loss

When results become hard to find, later researchers cannot build on them efficiently. Negative or partial outcomes disappear instead of reducing duplication. Lessons that could have improved future projects remain locked inside temporary project infrastructures, scattered PDFs, or internal reporting channels.

2. Lower public knowledge yield

Public research funding is not just about spending money on good intentions. It is meant to generate durable scientific, technical, and societal value. If outputs are not made persistently visible and reusable, the effective return on public funding is lower than it could be. This is especially serious in fields where learning cycles are expensive, slow, and cumulative.

3. A legitimacy problem

If access to funding becomes stricter and stricter, but the publicly visible downstream knowledge remains fragmented or disappears from view, the system risks a legitimacy gap. Why should researchers and the public accept intensifying front-end discipline if the back end does not show equivalent seriousness about preserving publicly funded results? A system that grows harsher at entry while remaining weak at output retention risks losing not only knowledge, but trust.

What follows from this

The main conclusion is not that grant review should become looser. Nor is it that all output failures reflect bad intent.

The real conclusion is more structural:

The key policy problem is no longer only who gets funded. It is what the funding architecture makes durable — and what it quietly allows to disappear.

If that diagnosis is correct, several next steps follow.

First, result visibility at project closure must be treated as infrastructure, not as optional aftercare.

If final reports, public summaries, lessons learned, or negative findings already exist, they should not depend on post-project goodwill, leftover communications capacity, or whether a website remains maintained. Persistent public traceability should be designed into the funding architecture itself.

Second, negative and partial results must count as outputs.

A project that did not achieve its hoped-for breakthrough may still generate high-value knowledge. In many scientific and engineering domains, knowing what failed, under which conditions, is socially and economically useful. Treating only success narratives as communicable outputs weakens collective learning.

Third, proposal systems and output systems should be analysed together.

The front end and the back end are usually discussed separately: review pressure here, dissemination there. That separation obscures the system effect. The more competitive and professionalised proposal access becomes, the more important it is to ensure equally robust structures for post-award public knowledge retention. Otherwise the asymmetry deepens.

Fourth, funding policy should monitor knowledge retention, not only allocation and launch.

It is not enough to know how many calls were run, how many projects were funded, or how much money was disbursed. We should also ask: two years after project completion, how much of the produced knowledge remains publicly findable, attributable, and reusable? If the answer is weak, then the system has an output problem even if its selection procedures look excellent.

Final thought

The ERC is responding to overload by making access harder. That may be understandable. But it also reveals a broader truth: public research funding systems are under pressure, and pressure changes behaviour. As access becomes more selective, actors adapt ever more professionally to the logic of selection.

If, however, the same system remains comparatively weak at preserving and exposing what funded projects actually produce, then we are drifting toward an unhealthy equilibrium:

That is not a sustainable model for public knowledge production.

A funding system that penalises unready proposals but tolerates vanishing public results is not yet optimised for knowledge yield. It is optimised for selection under load.

That is a different thing.

Subscribe now

Interesting Findings:

- The ERC’s 2027 change is not just about scientific quality; by the ERC’s own explanation, it is also a response to review-capacity overload, with some panels now handling 250+ proposals.

- EU institutions already state that beneficiaries are legally obliged to disseminate and exploit results, which makes the persistence of weak public result visibility more striking.

- The strongest structural insight is the asymmetry itself: the system is becoming more disciplined in controlling access than in retaining publicly funded knowledge.

THE PATTERN

There is a specific type of professional that every woman in STEM and deep tech knows but few people name in public.

He has never founded anything. Never shipped a product that had to survive in the real world. Never put personal capital — financial, reputational, or otherwise — behind a bet that could fail. His career has unfolded within institutions that do not require this of him: tenured universities, publicly funded research bodies, committees and boards where authority is granted by proximity to power rather than demonstrated by results.

He is, in Nassim Taleb’s precise formulation, a man without skin in the game.1

And he is, in ways that are neither accidental nor inevitable, frequently the most dangerous person in the room — particularly if you are a woman who has built something real.

The pattern, once you have seen it, is unmistakable.

A woman publishes. Speaks. Leads. Becomes visible. Her work is specific and testable — it exists in the world and can be evaluated against it. And then, reliably, he appears. On a panel. In a review. On LinkedIn. In a hallway conversation that somehow reaches the right ears. The critique is rarely substantive. It is strategic. It is, underneath the borrowed language of rigor and standards, an act of territory-marking by someone whose territory was never earned.

I have watched this across over almost two decades, in two worlds. The academic and the industrial. The pattern does not change with the context. Only the vocabulary does.

This is not about bad men. It is about a very specific institutional logic — one that protects certain people from consequence, rewards authority untethered from accountability, and then, predictably, turns that protected authority against the people who threaten it most.

This is not a personality problem. It is a structural one.

And it is time to name the structure.

Subscribe now

THE INSTITUTION AS INCUBATOR

How the System Produces, Protects, and Rewards the Man Just Described

Let’s be precise about something.

The figure described in the opening is not a statistical outlier. He is not the result of a hiring mistake or a cultural accident. He is, in many of the institutions that govern knowledge production and technology development, a predictable output.

The system did not fail to filter him out. The system built him. Tenured him. Funded him. Handed him committees, editorial boards, advisory roles, and conference keynotes. And it did so through mechanisms that are well-documented, structurally coherent, and remarkably stable across decades.

This chapter is about those mechanisms.

Tenure, Civil Servant Status, and the Architecture of Immunity

There is a thought experiment worth running.

Imagine a professional environment in which you cannot be fired for poor judgment. In which your income is guaranteed regardless of whether your decisions produce results. In which the people most affected by your choices — students, junior researchers, grant applicants, early-career scientists — have no formal mechanism to hold you accountable. In which your position, once secured, is yours until you choose to leave it.

You do not have to imagine it. It exists. It is called tenure. And in various forms — civil servant protections, lifetime appointments, publicly funded chairs — it structures large portions of academia and publicly funded research across the world.

To be clear: the original argument for tenure is not unreasonable. Academic freedom requires protection from political and economic pressure. Researchers need the security to pursue questions that are unpopular, unfundable, or inconvenient. These are real concerns.

But the structural consequence of tenure — particularly when combined with weak performance evaluation, opaque promotion criteria, and the slow accumulation of institutional power — is something sociologist Joan Acker identified with clinical precision: what she called inequality regimes.

In her foundational 2006 paper ”Inequality Regimes: Gender, Class, and Race in Organizations,” Acker argued that organizations do not produce inequality as a side effect of otherwise neutral processes. They produce it routinely, as part of their normal functioning. The practices that appear neutral — hiring committees, peer review, informal mentorship, conference invitations — are embedded in assumptions about who counts as serious, who counts as authoritative, who counts as a safe investment of institutional resources.2

The result, in Acker’s framework, is not a collection of individual biases. It is a regime — a self-reinforcing system in which certain profiles accumulate advantage and certain profiles accumulate friction, and in which the system presents this outcome as meritocratic.

Tenure, in this light, is not just job security. It is consequence insulation. It removes the feedback loop between decision quality and professional outcome. And when consequence insulation is distributed unequally — more available to those who already hold structural advantage — it compounds over time into something that looks, from the outside, like natural authority.

It is not natural. It is engineered. And it is, in most STEM institutions, engineered in ways that systematically favor men who entered the system early, accumulated positional power, and now operate from positions that are, for practical purposes, unassailable.

The Glass Escalator: Accumulating Authority Without Earning It

In 1992, sociologist Christine L. Williams published a study that should have been more disruptive than it was.

Studying men working in female-dominated professions — nursing, teaching, social work, librarianship — Williams found something counterintuitive: rather than facing the penalties typically associated with entering a “wrong-gender” field, men in these professions were systematically elevated. Promoted faster. Moved into leadership roles more quickly. Treated as natural authorities in domains where women had more experience and, often, more competence.

She called it the glass escalator — the invisible upward pressure that carries men toward leadership positions regardless of whether they sought them, deserved them, or were particularly suited to them.3

Williams revisited her findings in 2013, noting that the escalator effect was most powerful in high-status, male-coded domains — precisely the fields where you would expect meritocracy to be most rigorously enforced. In STEM and deep tech, where expertise is supposedly measurable and credentials are supposedly clear, the glass escalator does not disappear. It accelerates.4

The mechanism is not primarily overt discrimination. It is institutional proximity to power. Men in STEM institutions are more likely to be taken seriously in informal conversations that precede formal decisions. More likely to be invited into the networks where funding opportunities are discussed before they are announced. More likely to be seen as “the future of the field” by the senior men who currently define what the field is. More likely to receive the benefit of the doubt when their work is incomplete, their ideas are half-formed, or their track record is, to put it generously, developing.

The result, accumulated over a career, is a cohort of men who hold significant institutional authority — over hiring, over funding, over publication, over reputation — that was built less on demonstrated competence than on structural momentum.

This matters enormously when we ask why these men feel entitled to judge the work of women who have actually built things.

The answer is not that they have earned that judgment through superior expertise. The answer is that the institution handed them the gavel before anyone thought to ask whether they had read the case.

No Skin in the Game: The Cost of Judgment Without Consequence

Nassim Nicholas Taleb’s central argument maps onto gender dynamics in STEM institutions with uncomfortable precision.

Taleb’s thesis, stripped to its core: systems that separate decision-making from consequence-bearing are systems that will make bad decisions, reliably and systematically. When the person who makes the call does not bear the cost of being wrong, the incentive to be right weakens. Judgment becomes performative. Authority becomes positional. And the people who actually carry the risk — who have put something real on the line — are systematically underweighted in the decisions that affect them most.

In deep tech and STEM, this dynamic is not hypothetical. It is operational.

Consider the architecture of consequential decisions in these ecosystems: research funding allocation, faculty hiring, grant peer review, startup investment (in its more institutional forms), technology strategy in large organizations. These decisions are frequently made by people who have never founded a company that could fail, never shipped a product that had to survive in a competitive market, never put their own capital — financial or reputational — behind a technical bet with real stakes.

They have, instead, accumulated the institutional credentials that grant access to these decisions: professorships, committee memberships, board seats, advisory roles. Credentials that were built, as described above, in systems that do not require a market test, do not punish sustained mediocrity, and do not correct for the gap between claimed expertise and demonstrated results.

The consequences of this, in innovation ecosystems, are not trivial.

Funding decisions made by people without operational experience tend to systematically undervalue the kind of knowledge that cannot be credentialed: the knowledge of someone who has actually tried to build the thing and learned where it breaks. They tend to overvalue the kind of knowledge that can be performed in committee rooms: articulate positioning, institutional affiliation, familiarity with the right vocabulary.

Women in deep tech — particularly founders, operators, and applied researchers — disproportionately hold the former kind of knowledge. They are, disproportionately, the people who have skin in the game. And they are evaluated, disproportionately, by people who do not.

This is not incidental to the pattern described in this article. It is central to it.

When the man with institutional authority and no operational experience attacks the woman with operational experience and insufficient institutional authority, he is not engaging in an act of intellectual critique. He is engaging in an act of system maintenance — reasserting the primacy of the credential over the result, the position over the performance, the committee over the market.

The institution trained him to do this. The institution protects him when he does. And the institution, in the absence of structural intervention, will continue to produce the conditions in which this exchange is not only possible but predictable.

THE THREAT RESPONSE

Why Competent, Visible Women Are Not Random Targets

There is a question that took me longer than it should have to ask correctly.

For years, I framed it as: why do certain men behave this way? It felt like the right question. It directed attention at the behavior, at the individual, at the specific act of professional hostility I was trying to understand.

It was the wrong question. Or rather, it was an incomplete one.

The more useful question is: why these women, specifically?

Because the targets are not random. They never are. The women who draw the most sustained, most public, most institutionally consequential hostility from the kind of men described in this article share a common profile. They are visible. They are specific. Their work exists in the world in ways that can be evaluated. They have, in short, done the thing that their attackers have not: they have built something real, and it has been noticed.

This is not coincidence. It is the mechanism.

Status Threat, Backlash, and the Sociology of Defensive Authority

The research on this is, at this point, substantial enough that “backlash” has become a term of art in the literature on gender and organizations.

The core finding, replicated across contexts: when men whose identity and authority are rooted in male-coded domains perceive those domains as threatened by the increasing presence or visibility of women, the likelihood of hostile, demeaning, or undermining behavior increases. This is not a fringe result. It appears in studies of hiring decisions, performance evaluations, peer review, and public professional discourse.5

A 2022 study examining resistance to gender diversity in male-dominated fields found that the strength of resistance was most reliably predicted not by explicit sexist attitudes but by perceived status threat — the sense that one’s position, identity, or authority was being encroached upon.6 The man who attacks is often not, by his own account, hostile to women in general. He is hostile to this woman, now, in this field, at this level of visibility.

The distinction matters. It tells us something important about the structure of the behavior.

Kate Manne, in her widely cited work “Down Girl: The Logic of Misogyny” (2018), offers a framework that clarifies this considerably. Manne argues that we misunderstand misogyny when we treat it primarily as an attitude — a set of negative beliefs about women held by individual men. This framing, she suggests, leads us to look for the wrong thing: explicit prejudice, conscious hostility, stated contempt.

What Manne identifies instead is misogyny as a social enforcement mechanism — a system of penalties and rewards that operates to maintain a particular social order. Sexism, in her framework, provides the ideology that justifies the hierarchy. Misogyny provides the enforcement — the reactions that are triggered when the hierarchy is violated, when women occupy positions or claim authority that the system has not sanctioned for them.7

In this light, the attacks are not expressions of personal feeling. They are expressions of system logic. The man who publicly undermines a woman who has built something visible and real is not, primarily, acting on hatred. He is acting on threat detection. His position in the hierarchy — an institution built on credentials rather than competence, on exclusion rather than selection — depends, structurally, on the devaluation of the kind of authority she represents.

This is not a metaphor. It is a description of a zero-sum calculation that certain institutional architectures make structurally inevitable.

The Specific Pattern: What It Looks Like, and Where

The attacks do not come from strength. This is the first thing to understand about them, and it is the thing most obscured by the form they take.

They are performed with confidence. They deploy the vocabulary of rigor, standards, concern for the field. They arrive in formats that carry institutional authority: the peer review, the panel comment, the published critique, the conference hallway conversation that reaches the right people. They are, on the surface, professional.

Underneath, they are a fear response dressed in the language of expertise.

The specific choreography varies by context, but the logic is consistent.

In academia, the instruments are well-known to anyone who has spent time there. Peer review — anonymous, consequential, structurally protected from accountability — is the most powerful. A hostile reviewer does not need to make a compelling argument. He needs only to raise sufficient doubt, introduce sufficient friction, consume sufficient time. Careers are built and destroyed on the cumulative weight of these interventions, and their authors bear no cost for being wrong, unfair, or strategically motivated.

Conference dynamics operate similarly. Who gets invited to speak. Who gets asked the hard questions from the floor. Whose work gets cited in the literature review and whose gets omitted. Whose graduate students get taken seriously by the senior figures whose networks will determine their careers. These are not trivial decisions. They are, collectively, the infrastructure through which scientific reputation is constructed — and they are, in fields where the figures described in this article hold significant institutional power, available as instruments of competitive suppression.

Informal networks complete the picture. The conversation after the conference dinner. The email to a mutual colleague. The offhand remark in a hiring committee meeting that, formally, never happened. These are the mechanisms through which reputations are managed at the margins — softened, complicated, made *controversial enough* that the path of least resistance is to look elsewhere.

In deep tech, the instruments are different but the logic is identical.

Funding decisions — particularly at the institutional and later-stage level — are shaped by networks of advisors, board members, and evaluators whose composition tends to reproduce the same demographic patterns described in Chapter II. A woman founder whose technical credibility is quietly questioned in the right conversation may never know why the term sheet didn’t come. A female researcher whose work is framed as “interesting but not quite ready” by the wrong reviewer may find her visibility in the ecosystem quietly managed downward.

Media and public narrative are particularly powerful here. Deep tech ecosystems are small enough that a few authoritative voices can significantly shape who is treated as a credible expert and who is not. The man with the tenured position and the conference keynotes and the LinkedIn following built on years of institutional proximity to power has, structurally, more capacity to shape this narrative than the woman who has spent those same years actually building things.

This asymmetry is not accidental. It is, again, a feature of the architecture.

A Personal Register: What Two Decades of Pattern Recognition Looks Like

I want to be precise here, because precision is what this section requires.

I am not describing a single incident. I am not describing a particularly bad year, or a particularly hostile institution, or a particularly egregious individual. I am describing a pattern — one that I have observed, experienced, and documented across two decades, in two distinct professional worlds, on multiple continents.

The specific forms it has taken have varied. The academic context has its own vocabulary; the deep tech and startup ecosystem has another. The men involved have been different people with different credentials and different stated motivations. Some were explicit. Most were not. Most, I suspect, would be genuinely surprised to hear their behavior described in these terms.

That is, in a sense, the point.

The pattern does not require malice to function. It requires only that certain men have been placed in positions of authority that their actual competence does not justify, that they have been protected from the feedback that would normally correct this, and that they encounter, in visible and competent women, a form of evidence they are not structurally equipped to process without discomfort.

The discomfort is real. The response to that discomfort is what I am describing.

What I can say, with the specificity that two decades of observation permits: the attacks have been most acute at the moments of greatest visibility. A significant publication. A successful raise. A public recognition. A keynote. The moments, in other words, when the evidence of competence becomes hardest to ignore — these are the moments that tend to trigger the most deliberate, most institutionally mediated responses.

I kept building anyway.

I also kept noticing. And what I noticed, over time, was that the pattern was too consistent, too structurally coherent, too reliably concentrated in specific institutional contexts to be explained by individual psychology.

It was not about me. It was not about the specific men involved. It was about what the institution needed from both of us — and what it could not tolerate me becoming.

The noticing, it turns out, is the part that matters. Because you cannot name what you cannot see, and you cannot change what you cannot name.

THE COST

This Is Not a Diversity Problem. It Is an Innovation Problem.

Let’s set aside, for a moment, the language of equity.

Not because equity doesn’t matter. It does. But because framing this exclusively as a fairness problem has a well-documented failure mode: it allows the people most responsible for the dysfunction to treat it as someone else’s concern. An HR matter. A pipeline issue. A question of representation that will resolve itself, eventually, if we run enough workshops and publish enough commitments.

It will not resolve itself. And the cost of not resolving it is not primarily a moral cost — though it is that too. It is an operational cost. A competitive cost. A cost measured in the specific currency that deep tech and STEM institutions claim to care about most: breakthroughs that don’t happen, problems that don’t get solved, and companies that don’t get built. Companies, that fairly are the only vehicle to fund those tenure positions — without which none of this would be happening.

This chapter is about that cost. Specifically, and without abstraction.

The Systemic Loss of Operational Knowledge

There is a particular kind of knowledge that cannot be credentialed.

It is the knowledge of someone who has tried to build the thing and learned, through direct encounter with physical, computational, or market reality, where it breaks. Who has managed the gap between what the model predicts and what the prototype does. Who has made a resource allocation decision under genuine uncertainty and lived with the consequences. Who has, in short, been wrong in ways that cost something and updated accordingly.

This is not the same as the knowledge produced by studying systems from a distance, however rigorously. It is not better or worse in some absolute sense — both forms of knowledge are necessary in any functioning innovation ecosystem. But they are different, and the difference matters enormously when the question is: who gets to make consequential decisions?

The institutional dynamics described in Chapters II and III produce a systematic bias in how these two forms of knowledge are weighted. The man with the tenured position, the publication record, and the committee seat holds knowledge of the second kind — produced at a distance, protected from consequence, evaluated by peers operating within the same insulated system. The woman who has founded a deep tech company, shipped a product, or navigated the specific brutal feedback loop of applied research in a competitive environment holds knowledge of the first kind.

The institution, for the structural reasons already described, weights the former over the latter. The committee trusts the professor over the founder. The peer review trusts the credential over the result. The funding decision trusts the network over the track record.

What the ecosystem loses, in this exchange, is not representation. It is reality contact.

The decisions that shape the direction of deep tech — which research directions get funded, which companies get backed, which technical bets get made — become progressively less grounded in the knowledge of what actually works. They become more grounded in the knowledge of what sounds credible to people who have never had to find out.

This is not a theoretical risk. It is a description of how significant portions of deep tech funding and research prioritization currently operate.

The Data: Performance Without Power

The empirical picture is, at this point, consistent enough to be embarrassing.

Women-founded companies, when they receive venture funding, outperform their male-founded counterparts by meaningful margins. A 2018 analysis by Boston Consulting Group found that startups founded or co-founded by women generated 10% more cumulative revenue over a five-year period than those founded by men — despite receiving, on average, significantly less funding.8

In deep tech specifically, the gap between performance and access is particularly acute. Women represent a growing share of STEM graduates across most developed economies — approaching or exceeding parity in fields including biology, chemistry, and mathematics — but remain dramatically underrepresented in the leadership structures of deep tech companies, research institutions, and the funding bodies that shape both.9

The European Institute for Gender Equality has documented that women account for fewer than 20% of science and technology board members across EU institutions, and that the gap widens significantly at the level of research program leadership and funding committee membership.10

What this means, concretely, is that the people making the decisions that shape the ecosystem are, disproportionately, not the people with the deepest operational knowledge of it. And the people with the deepest operational knowledge are, disproportionately, making their contributions in contexts where their authority to shape those decisions is structurally limited.

This is the cost, rendered in numbers.

But numbers, in this context, are the floor of the analysis — not the ceiling.

What Doesn’t Get Built

The harder cost to quantify is the counterfactual: the things that don’t exist because the people who would have built them were managed out, crowded out, or worn out by the dynamics described in this article.

Consider what systematic exclusion actually removes from an innovation ecosystem.

It removes perspectives that are not redundant. The value of diverse teams in innovation contexts is not primarily symbolic — it is informational. Teams that bring genuinely different mental models, different relationships to risk, different understandings of user need, produce better solutions to hard problems. This is not a controversial finding in the organizational literature.11 It is, however, a finding that is systematically ignored by institutions whose hiring and advancement practices reproduce the same profile repeatedly and call the result a meritocracy.

It removes specific forms of risk tolerance. Women founders in deep tech — operating, as described, with less institutional protection and less access to the networks through which opportunities are distributed — tend to make more conservative capital allocation decisions and build more robust technical foundations before seeking scale. This is frequently misread, in ecosystems that fetishize the blitzscale narrative, as a lack of ambition. It is more accurately read as the rationality of someone who has skin in the game and cannot afford to be wrong.

It removes the feedback that corrects systems. When the people with the most direct knowledge of where a technology or a research direction is failing are also the people with the least institutional power to communicate that failure upward, systems lose their error-correction capacity. Bad bets persist longer. Wrong assumptions get embedded in roadmaps and funding priorities. The gap between institutional consensus and ground truth widens — slowly, invisibly, expensively.

The Specific Irony

There is something worth naming directly about the particular configuration this article has been describing.

The men who, through the institutional mechanisms of Chapter II, have accumulated authority over deep tech and STEM ecosystems without the operational experience those ecosystems require — these men are now the primary managers of systems whose entire value proposition depends on doing things that have never been done before.

Innovation, in any meaningful sense, requires tolerance for being wrong. It requires the institutional capacity to recognize when a bet has failed and update accordingly. It requires the judgment that comes from having made real decisions with real consequences and learned from the results.

It requires, in other words, exactly the qualities that institutions built on consequence-insulation and credential-accumulation systematically select against — and exactly the qualities that the women being pushed out of these ecosystems disproportionately possess.

The ecosystem is protecting the people least equipped to lead it. It is discarding the people most equipped to do so. And it is doing this not despite its stated commitment to innovation, but in direct structural contradiction to it.

This is not a diversity problem that the innovation ecosystem can afford to treat as peripheral to its core concerns. It is a dysfunction at the center of how the ecosystem allocates its most consequential resource: the authority to decide what gets built next.

Naming It Is Not Nothing

I am not going to end this with a list of recommendations.

Not because recommendations don’t exist — they do, and some of them are even good. Blind review processes. Structured funding evaluation criteria. Accountability mechanisms for institutional gatekeepers. Mandatory disclosure of reviewer conflicts of interest. These are real interventions with real evidence behind them, and they matter.

But policy lists have a particular failure mode in pieces like this one: they allow the reader to metabolize a structural critique as a management problem. To convert something that is, at its core, a description of how power works into a checklist that can be delegated, completed, and filed.

That is not what this is.

What this is — what it has been, from the opening scene forward — is an attempt to do something more primary and, I would argue, more durable: to name a pattern precisely enough that it becomes harder to unsee.

On the Act of Naming

The pattern described in this article is not new. The women who have navigated it know it with the specific, embodied knowledge of people who have encountered it repeatedly, in multiple institutions, across the span of careers. They know it the way you know a road you have driven in the dark — not from a map, but from the turns.

What has been missing, in many of these encounters, is not the experience. It is the language.

This matters more than it sounds.

When a pattern has no name, each instance of it appears as an isolated event. A difficult colleague. An unusually competitive peer. An unfortunate dynamic in a particular institution. The woman who experiences it is left to process it as something specific to her — her field, her institution, her particular bad luck. She adapts. She develops workarounds. She builds, as I have, a kind of private cartography of where the hazards are.

What she cannot easily do, without shared language, is connect her experience to the structural mechanism that produced it — and therefore cannot easily connect it to the experiences of the women beside her, ahead of her, and coming after her.

Naming changes this. Not immediately. Not completely. But in ways that are, over time, genuinely consequential.

When you can say: this is not a personality conflict, it is a status threat response — you have done something. When you can say: this is not rigorous criticism, it is institutional discipline executed through the vocabulary of expertise — you have done something. When you can say: this man does not have authority because he has earned it, he has authority because a system that was never designed to evaluate him gave it to him and then protected it — you have done something that individual resilience, however admirable, cannot do alone.

You have made the invisible visible. And visible things can be responded to.

What This Does Not Resolve

Let me be honest about the limits of this.

Naming a pattern does not dismantle the institutions that produce it. It does not restore the careers that were quietly managed downward, the funding that didn’t come, the research directions that were abandoned because the friction cost became too high. It does not give back the years spent navigating systems that were, beneath their stated commitments to merit and excellence, running a different and unacknowledged algorithm.

It does not protect the next woman who encounters the next version of this man in the next institution that will shelter him.

These are real limits, and I will not paper over them with optimism I don’t feel.

What naming does — what it has always done, in every context where structural power has eventually been successfully challenged — is create the conditions for collective recognition. The moment when a person reads a description of something they have experienced privately and thinks: so it wasn’t just me. The moment when the diffuse, hard-to-articulate sense that something is wrong becomes a shared framework with enough precision to be argued, challenged, and built upon.

That moment is not the end of anything. It is, at most, a beginning.

But beginnings require a first step. And the first step is always, without exception, the same: someone has to say it out loud.

A Note to the People Who Recognize This

If you have read this far and found yourself nodding — not with the comfortable nod of someone encountering an abstract argument, but with the specific, slightly uncomfortable nod of someone recognizing their own experience in a description they didn’t expect to find written down — then this closing is, in part, addressed to you.

You already knew this. You knew it from the first time it happened, even if you didn’t yet have the framework to explain it. You have been carrying the knowledge of it, in the form of accumulated caution and recalibrated expectations and the very specific energy expenditure of operating in systems that require you to be, simultaneously, better than the standard and more careful about being seen to be better.

That knowledge is not nothing. It is, in fact, data. It is the most reliable data available about how these systems actually function, as opposed to how they describe themselves.

What I am suggesting — the only thing I am suggesting, in the absence of a policy list — is this:

Trust the pattern you have been observing. Name it when you can. Document it when it’s safe to do so. Connect it to the structural mechanisms that produce it, so that it stops presenting itself as personal and starts presenting itself as what it is: a predictable output of a specific institutional architecture that can, in principle, be redesigned.

And keep building.

Not because the system deserves your continued contribution to it. Not because persistence in the face of structural hostility is a virtue that should be demanded of anyone. But because the work is real, and the problems are real, and the alternative — ceding the field to the people who have never had skin in the game — is one the ecosystem genuinely cannot afford.

The Last Thing

There is a man, somewhere, who will read this article and recognize himself in it.

Not the caricature — not the pantomime villain with explicit malice and conscious contempt. The more ordinary version. The man who has spent a career in institutions that rewarded him in ways he mostly experienced as natural. Who has, on occasion, felt genuinely threatened by someone whose competence illuminated the gap between his authority and his actual capabilities. Who has, in those moments, done something — in a review, in a conversation, in a small consequential act of institutional gatekeeper-ship — that he would not, if pressed, be proud of.

To that man, I do not have a condemnation. I have a question.

What would it cost you to stop?

Not to become an activist. Not to dismantle your institution. Not to perform a public reckoning. Simply to stop — to decline the next opportunity to use your institutional position to manage downward someone whose competence makes you uncomfortable.

The system will not immediately notice. Your career will not be affected. The committee will find another way to reach its preferred conclusion.

But the woman on the other side of that decision will notice. And she will keep building. And eventually — not quickly, not without cost, not without the kind of sustained effort that institutions structured the way ours are make genuinely difficult — the work will speak for itself.

It always does.

The pattern has a name now.

That is not nothing.

It is, in fact, exactly where everything that comes next begins.

Finish Note: I am not stopping here.

We keep diagnosing the wrong disease.

Every few months, a new report lands: European deep tech is underfunded. Founders are burning out. Capital is fleeing to the US. The (policy) response is predictable — more accelerators, more pitch coaching, more investor matchmaking events. The ecosystem response is equally predictable — founders learn to tell better stories about technology that isn’t ready to be a story yet.

And nothing structurally changes.

This article argues something that will be uncomfortable for a lot of people building programs right now:

Deep tech doesn’t need a better version of the program designed for software startups. It needs a program designed for what deep tech actually is.

The Transplant Problem

The modern accelerator was built for a specific kind of innovation: software, platforms, consumer apps. The logic is elegant and internally consistent. Build fast. Ship early. Get user feedback. Iterate. The truth of your product is revealed quickly — through retention, through conversion, through the brutal clarity of whether people actually use the thing.

This logic works beautifully when the core uncertainty is market uncertainty. You don’t know if people want it. So you test fast, fail cheap, and pivot.

Deep tech has a fundamentally different uncertainty structure. The core uncertainty isn’t whether the market wants it. The core uncertainty is whether the technology actually works at scale, under real conditions, within regulatory frameworks, across supply chains that don’t yet exist. That’s not market uncertainty. That’s technical and scientific uncertainty — and it cannot be resolved through pitch iteration or investor signal.

When you apply accelerator logic to this kind of uncertainty, something quietly catastrophic happens. Founders learn to perform market readiness before achieving technical readiness. They build decks that describe a company that doesn’t exist yet — not because they’re dishonest, but because the program rewards that behavior. The KPI is investor interest. The investor is looking for traction signals. The founder produces traction signals.

Meanwhile, the technology stays at TRL 4.

What We’re Actually Losing

Let’s be precise about the damage, because it’s distributed across the entire ecosystem and most of it is invisible until it’s too late.

Founders burn out not just from workload, but from cognitive dissonance. They are scientists and engineers who built something real, and they are now spending 60% of their time on activities that have nothing to do with making it more real. The burnout isn’t about hours. It’s about the sustained pressure to perform a narrative that is ahead of the substance.

Investors and funding bodies take capital losses that are systematically misattributed. When a deep tech company fails after Series A, the post-mortem usually says “team issues” or “market timing.” Rarely does it say: “the technology was not actually ready, and no program along the way was designed to verify that.” The capital loss isn’t bad luck. It’s a structural outcome of the program design.

Public programs and TTOs face mandate erosion. They’re measured on portfolio size, deal flow, and startups created — not on whether those startups produce durable technological value. When the portfolio underperforms five years later, the accountability is diffuse. Nobody designed bad metrics on purpose. But the metrics were wrong, and the programs followed the metrics.

Societies and governments are perhaps the largest losers. Europe’s technological sovereignty agenda is real and urgent. But technological sovereignty is not created by funding companies that can pitch. It is created by technologies that actually reach market readiness — materials that can be manufactured, therapeutics that can be approved, hardware that can be produced at scale. Deck polish does not produce technological sovereignty. Technical depth does.

The Diagnostic Question Nobody Asks

Before a deep tech venture enters any program, there is one question that should be asked — and almost never is:

What is the actual source of uncertainty in this venture right now, and is this program designed to reduce that specific uncertainty?

For most deep tech, the answer at early stage is: the uncertainty is technical. The molecule might not bind. The material might not be manufacturable at volume. The device might not pass regulatory stress testing. The supply chain for a critical input might not exist.

None of these uncertainties are resolved by Demo Day.

A program designed for deep tech would ask this question at intake, and would route the venture accordingly. Not every company needs the same intervention. Some need extended laboratory access. Some need regulatory affairs expertise embedded in the program. Some need manufacturing partners before investors. Some need two more years of technical development before any commercial activity is appropriate at all.

The accelerator model collapses all of these into a single 12-week sprint toward investor readiness. It is not malicious. It is a category error.

What a Deep Tech-Specific Program Actually Looks Like

This is where I want to be concrete, because the critique without the alternative is just noise.

A program designed around deep tech commercialization — not startup acceleration — would contain the following structural elements:

1. TRL-gated progression, not cohort-based timing

The accelerator model runs everyone through the same 12-week program regardless of where they are. A deep tech program gates progress on technical milestones, not calendar time. You move to the next phase when you’ve achieved reproducible results at scale, or a validated manufacturing pathway, or a completed regulatory pre-submission — not because 12 weeks have elapsed.

This sounds obvious. It is almost never done.

2. Embedded scientific and regulatory expertise, not mentor networks

The value in a software accelerator is often the network — introductions to investors, to potential customers, to other founders. Deep tech founders need something different: access to people who can evaluate whether the science holds up, whether the regulatory pathway is realistic, whether the manufacturing assumptions are founded. This is not a mentor. This is a scientific and technical review board, embedded in the program, with actual accountability for the assessments they make.

3. Failure taxonomy that distinguishes technical from commercial failure

When a venture exits a program without a deal, the current ecosystem records that as failure. A deep tech program would disaggregate this. A company that exits having definitively answered a critical technical question — even a negative answer — has produced value. A company that exits having produced a polished deck and an unanswered technical question has produced a liability. These are not the same outcome and should not be recorded as the same outcome.

4. Time horizons that match the technology, not the fund cycle

A 10-year fund cycle is genuinely incompatible with many deep tech timelines. A deep tech program should be honest about this at intake and should not pretend otherwise. This means building bridge structures, phased instruments, and long-horizon public co-investment that isn’t trying to exit in seven years. It means being willing to say: this technology needs a 15-year development arc, and the right capital structure for that is not a VC round.

5. KPI architecture that measures technical maturation, not commercial performance proxies

This is perhaps the most important and least visible element. Programs are measured on what they track, and what they track shapes what founders optimize for. If the program KPI is “number of investor meetings,” founders will produce investor meetings. If the KPI is “TRL advancement per program cohort” or “regulatory milestone achieved” or “manufacturing pilot completed,” founders will produce those things instead.

The KPI architecture is not an administrative detail. It is the program’s actual theory of change, made visible.

A Note on Speed

I want to address the objection I hear most often when I raise these arguments: “But we can’t slow down. The competitive pressure is real. China is moving fast. The US is moving fast. We need speed.”

This confuses two different kinds of speed.

There is velocity — the appearance of movement, the production of signals, the rate at which things happen in the near term. And there is momentum — the actual accumulation of technical substance that creates durable competitive position.

Accelerators are very good at producing velocity. They are structurally poor at producing momentum.

Technological sovereignty is built on momentum. A technology that reaches genuine market readiness — that can be manufactured, deployed, regulated, and scaled — creates durable competitive position. A company that raised a Series A on a premature technology creates a future liability.

Speed toward the wrong milestone is not a competitive advantage. It is a future cost, distributed across founders, investors, and public programs that will eventually absorb the losses.

The argument for deep tech-specific programs is not an argument for slowness. It is an argument for speed toward the right milestones.

What This Requires from the Ecosystem

I’m not naive about what it would take to actually build this.

It requires funders — both public and private — to accept that their measurement frameworks need to change. You cannot fund a TRL-gated program and measure it on deal flow within 18 months. The metrics and the program logic have to be aligned.

It requires accelerator operators to be honest about their scope. There is nothing wrong with building an excellent program for software startups. But marketing that program to deep tech founders is a disservice to the founders and to the ecosystem. The category distinction matters.

It requires policy-makers to hold two things simultaneously: urgency about technological competitiveness, and patience about technological timelines. These are not contradictory. They are complementary. The urgency should be directed at building the right infrastructure now, not at forcing the wrong infrastructure to produce faster results.

And it requires the deep tech community itself to stop accepting the terms of the wrong conversation. Every time we ask “how do we help deep tech founders pitch better,” we are accepting a frame that locates the problem in the founder. The problem is not in the founder. The problem is in the program design.

The Shift That’s Actually Needed

We have spent a decade trying to make deep tech fit into the accelerator model. We have coached scientists to be pitchers, compressed multi-year development arcs into 12-week sprints, and measured success by investor attention rather than technical progress.

The results are in. The burnout is real. The capital losses are real. The gap between deep tech’s potential and its actual delivery to market is real.

It is time to stop optimizing the wrong program and start designing the right one.

Deep tech doesn’t need better accelerators.

It needs programs built around what deep tech actually is: a category of innovation with structural technical uncertainty, long development timelines, specific regulatory and manufacturing requirements, and a value creation logic that is simply not the same as software.

That program doesn’t fully exist yet. Building it is one of the most important things the European innovation ecosystem could do in the next five years.

The question is whether we’re willing to admit that what we’ve been building so far isn’t it.

Here’s something habit researchers figured out a long time ago that most people in innovation policy have never connected to their work:

You don’t change behavior by wanting to change it.

You change behavior by changing the context that triggers it.

That’s not a motivational reframe. That’s neuroscience. The dorsolateral striatum — the part of the brain that runs habitual behavior — operates on a simple rule: same context, same response. Automatically. Without deliberation. The prefrontal cortex, the part that reasons and plans and commits to New Year’s resolutions, is simply slower. It gets overruled.

This is why habits are so hard to break at the individual level. Willpower is structurally outgunned.

Now — and this is the jump I want you to make with me — what happens when you scale that up to an institution?

Institutions Don’t Have Neurons. But They Have Something Equivalent.

Call them KPI architectures. Reporting cycles. Funding structures. The procedural rhythms that define what gets measured, what gets rewarded, what gets flagged as success.

These are the context cues of institutional behavior.

And just like the dorsolateral striatum, they operate automatically. A programme manager at a TTO doesn’t sit down every quarter and consciously decide to optimize for output metrics over outcome quality. The system does it for them. The grant cycle is 18 months. The interim report is due in nine. The KPIs were written when the programme was designed, by people who had a mental model of what “innovation support” looks like — a mental model built, mostly, on software startup logic.

That’s the cue. It fires every quarter. The behavior follows.

This is not a criticism of programme managers. It’s a description of how context-dependent automaticity works at organizational scale.

Now Add Deep Tech. And Watch the Mismatch Compound.

Deep tech is not a derived version of a software startup. It is a different category of venture entirely — one defined by core technical uncertainty, not market uncertainty. The risk profile is different. The timeline is different. The failure modes are different.

Deep tech ventures need, on average, 35% more time and 48% more capital than traditional tech startups before generating any revenue (McKinsey). The average time to Series A runs approximately 18 months longer than for software startups. These aren’t edge cases. They’re structural features of the category.

And the Valley of Death — the phase between early-stage research and commercial viability where funding consistently dries up — is precisely the phase that European public programmes are designed to bridge.

So the programme that’s supposed to help a materials scientist get from validated synthesis to pilot-scale production is operating on a 12-month output logic, measuring the number of workshops attended, the number of coaching sessions delivered, the number of startups “in portfolio.”

The scientist is at month 14 of a technical problem that has no known solution yet. The report says: on track.

That’s not a funding gap. That’s a structural mismatch between the temporal logic of the programme and the temporal logic of the venture it’s supposed to support.

Why This Doesn’t Fix Itself

Here’s where the habit analogy gets genuinely uncomfortable.

Research on habit change — at the individual level — identifies one of the most reliable conditions for new pattern formation: disruption of the old context. Moving house. Changing jobs. A health event. Something that removes the environmental cues that were automatically triggering the old behavior, and opens a window where new patterns can form before the old ones reassert themselves.

Organizational habit research puts the stabilization window at 6 to 24 months of sustained behavioral change — and that’s under conditions where the change is actively supported.

Now ask: what disrupts the context of a publicly funded innovation programme?

Budget cycles don’t change. Reporting requirements don’t change. The KPI framework was written four years ago and governs the next three. The institutional logic is, by design, stable. Stable enough to maintain continuity of public investment. Stable enough to prevent arbitrary reallocation.

Stable enough to be completely unresponsive to evidence that the programme architecture is misaligned with what it’s actually trying to produce.

This is why deep tech problem persists at the structural level despite genuine awareness of the problem. It’s not that the people running these programmes don’t understand the challenge. Many of them understand it in sophisticated detail. But understanding a problem from inside the context that generates it does not give you the leverage to change that context.

The system knows. The system continues.

The European Amplifiers

On top of this baseline institutional habit problem, Europe layers several structural conditions that make the dynamic worse.

Market fragmentation as a cue stabilizer. A US deep tech startup operates in a single regulatory context, a single legal framework, a single language market. European scale means navigating 27 distinct regulatory environments. This doesn’t just create operational friction — it makes the institutional context for support programmes heterogeneous in ways that resist standardization. Every national programme has its own logic, its own cues, its own deeply embedded behavioral patterns. There’s no single context to disrupt.

Risk aversion encoded as culture. European institutional culture — across both public programmes and the investor community — has historically coded failure as reputational damage rather than informational signal. This is not irrationality. It’s a rational adaptation to an incentive environment that penalizes visible failure. But it means the reward loop for programme managers doesn’t run through venture outcomes. It runs through compliance. Programmes that don’t produce obvious failures get renewed. The behavioral implication is exact: optimize for the absence of visible failure, not the presence of commercial success.

No accountability loop connecting programme architecture to venture outcomes. Habit formation requires feedback. The behavior needs to produce a signal — positive or negative — that reinforces or disrupts the pattern. In European innovation programmes, the signal chain between programme design decisions and eventual venture outcomes is so long, so diffuse, and so contaminated by other variables that meaningful feedback rarely reaches the level of programme architecture. When a portfolio company fails to scale, the cause is attributed to the founder, the market, the timing — rarely to the structural mismatch between programme logic and venture requirements. No feedback, no learning, no disruption of the pattern.

Bureaucracy as a cue machine. Application formats, interim reports, compliance audits, deliverable taxonomies — each one is a context cue that triggers documentation-optimization behavior rather than outcome-oriented behavior. This isn’t accidental. It’s the accumulated result of decades of accountability requirements, each individually reasonable, collectively producing a system whose primary behavioral output is the generation of compliant paperwork.

What the Numbers Look Like When Structural Mismatch Accumulates

This isn’t abstract.

In 2023, US deep tech startups raised $22 billion in Series B+ funding. European counterparts raised $7 billion. European tech startups are half as likely as their US peers to raise growth-stage rounds above $15 million — a gap that, measured across a decade, represents roughly $375 billion in missing growth capital.

And while deep tech attracted €15 billion in European VC in 2024 — a record — half of Europe’s growth capital still comes from outside the continent. The scientific talent is here. The research output is here. The structural translation machinery is misaligned.

These are not funding gaps in the sense that more money would solve them. They are outcome gaps produced by programme architectures that were never designed for the venture category they’re now being asked to support.

The Actual Mechanism

Let me be precise about what I’m claiming, because the argument gets misread easily.

I’m not saying European programme managers are doing their jobs badly. I’m not saying more training or better intentions would fix this.

I’m saying that the KPI structures and reporting cycles of European innovation programmes have created institutional habits — automatic, context-triggered behavioral patterns — that are structurally mismatched with the commercialization timeline and risk architecture of deep tech ventures.

And I’m saying that you cannot fix a context problem with a motivation solution.

What’s needed is structural diagnosis first: a rigorous analysis of where a programme’s measurement logic is creating behavioral incentives that diverge from its stated outcomes. Where the reporting cycle is shortening the cognitive horizon of the people making decisions. Where the milestone taxonomy was designed for a different category of venture and is producing systematic misallocation — not through malice, but through mechanism.

The same way you’d diagnose an individual’s environment before you redesign their habit loop.

Change the context. The behavior follows.

Why This Has to Come From Outside

Institutions that need to diagnose their own programme architecture face the same problem as individuals trying to identify their own unconscious habits.

The context that generates the behavior is the same context you’re using to evaluate whether the behavior needs to change.

That’s not a failure of intelligence. It’s a structural feature of self-referential systems.

This is why the diagnostic function needs to be external. Not as criticism — but as a structural mirror. Something that can read the KPI architecture, the reporting logic, the milestone sequence, and map where the temporal mismatch sits. Where the programme’s internal logic and the venture’s actual development trajectory diverge.

That gap is where the budget goes. Quietly. Consistently. Quarter after quarter.

And it’s measurable. Which means it’s fixable.

Here's how it can start

Innovation policy and venture capital are built on a hidden assumption: that all high-growth ventures follow the logic of the startup. That assumption is no longer just outdated. It is actively distorting the future of science-based innovation.

As economies pivot toward climate technologies, advanced materials, quantum systems, and next-generation therapeutics, a growing share of transformative ventures does not begin with market uncertainty but with scientific uncertainty. Their primary risk is technical proof, not customer acquisition. Their bottleneck is validation, not velocity. Yet they are still measured, funded, and accelerated as if they were software companies searching for product–market fit.

The result is structural misallocation of capital, premature kill decisions, performative milestones, and a widening gap between scientific potential and industrial deployment.

This article advances a clear thesis: science innovation is not an early-stage startup problem. It is a distinct venture category with different timelines, risk structures, capital sequencing requirements, and success metrics. Treating it as a variant of startup logic does not merely create inefficiency. It reshapes behavior in ways that systematically undermine technical breakthroughs.

This article reveals:

• how startup logic became the default lens for all innovation

• what structurally differentiates science innovation from software and applied engineering ventures

• the economic consequences of misclassification for investors, policymakers, and venture builders

• a shift toward category-specific evaluation models that align capital deployment with technical risk reduction rather than superficial growth signals.

The question is not whether science innovation deserves more funding. The question is whether we are using the right architecture to recognize and scale it at all.

The Challenging Status Quo

Across European policy documents, the language is calm, technical, almost bureaucratic. Words like monitoring framework, key impact pathways, leverage ratios, valorisation principles appear without drama. There are no grand announcements about revolutionizing innovation. No declarations that the old model has failed.

For decades, European innovation ecosystems optimized around visibility and activity. Programs were evaluated by number of supported ventures, demo days hosted, projects signed, logos acquired, follow-on funding secured. None of this was irrational. These were the variables embedded in the monitoring architecture. When input and output indicators dominate reporting systems, rational actors optimize for inputs and outputs.

The system produced what it was designed to measure.

What is changing now is not the rhetoric of innovation. It is the optimization variable. Check-mate.

From Inputs to Impact

Horizon Europe’s monitoring framework1 is no longer centered on counting activities alone. It is structured around Key Impact Pathways that explicitly track scientific, societal, and economic effects over time. The language has moved from project completion to downstream transformation.

InvestEU has refined its performance architecture as well.23 Its indicators increasingly focus on mobilized investment, private capital leverage, and sustained deployment rather than simply the number of agreements signed.

Parallel to this, the European Commission and Council have formalized a valorisation mandate: research and innovation results must be transformed into sustainable products and services. Not published. Not showcased. Deployed.

Individually, each of these adjustments appears incremental. Collectively, they reweight the system.

The monitoring architecture is moving from “activity performed” to “economic and industrial consequence generated.”

That distinction sounds semantic. It is not.

Measurement Systems Shape Behavior

Every measurement system shapes the behavior of the system it observes. When evaluation frameworks emphasize visibility, ecosystems evolve toward visibility. When they emphasize throughput, ecosystems evolve toward throughput.

For years, Europe’s innovation management layer operated under a metric regime that rewarded:

• Participation rates

• Event density

• Project counts

• Funding secured

• Network expansion

These variables are legible within short funding cycles. They create dashboards that look active. They generate political reassurance.

But deep tech does not operate on short feedback loops. Its time constant is closer to a decade than a quarter. The central uncertainty is not customer acquisition velocity. It is scientific reproducibility, industrial scalability, and regulatory integration.

If you measure short-term motion inside a long-term technological regime, you produce structural distortion.

The distortion remains invisible as long as funding renewals depend on short-term optics.

That condition is beginning to erode.

What Is Actually Being Optimized Now?

The public narrative suggests accountability and efficiency. The structural shift is more specific.

The new monitoring logic increasingly optimizes for:

• Market uptake

• Deployment in real economic contexts

• Private capital mobilization

• Sustainable industrial scaling

• Long-term impact trajectories

This alters incentive alignment across the ecosystem.

Accelerators designed around demo-day theater must now demonstrate industrial throughput. Venture builders that optimized for early valuation spikes must show credible progression toward deployment. Innovation agencies that previously reported participation metrics must track downstream effect.

The operator that once rewarded visibility is losing weight. The operator that rewards deployability is gaining it.

In system terms, the Hamiltonian has changed.

The Compounding Problem of Short-Term KPIs

In isolation, most legacy KPIs are not inherently wrong. Attendance numbers, follow-on funding rates, signed projects, press coverage — these can all signal short-term ecosystem health.

The problem emerges when we model their interaction over time.

Deep tech ventures rarely deliver meaningful industrial or societal outcomes within one funding cycle. A science-based technology may require seven to ten years before deployment scale reveals whether early-stage support created durable value.

If ecosystem actors optimize for short-term satisfaction metrics under a ten-year technological horizon, three second-order effects accumulate:

1. Capital is allocated toward ventures that can signal traction quickly, not those reducing deep technical risk.

2. Programs prioritize portfolio breadth over technological depth.

3. Institutional incentives favor optics over industrial integration.

This creates what can be described as structural debt.

On paper, the ecosystem appears vibrant. In material terms, industrial transformation lags.

The new policy architecture implicitly acknowledges this lag. By embedding impact pathways and leverage metrics into monitoring systems, Europe is attempting to align measurement with the actual time horizon of scientific innovation.

Irreversibility and Path Dependency

Metric architecture does not merely describe performance. It locks in trajectories.

When funding renewal depends on demonstrable long-term market outcomes, organizations that cannot track industrial progression will gradually lose mandate legitimacy. Not because of moral failure. Because of misalignment with the new optimization function.

This introduces path dependency.

Ecosystems that redesign their internal KPIs to model:

• TRL progression over time

• Evidence gates

• Industrial validation milestones

• Capital sequencing logic

will adapt.

Those that continue optimizing for activity density will encounter increasing friction when renewal cycles demand proof of materialization rather than motion.

The shift will not be dramatic. It will be cumulative.

This Is Not About Punishment

It is tempting to frame this as a warning: “innovation managers are next.” That framing misunderstands the dynamic.

This is not punitive. It is architectural.

Europe is attempting to reconcile its industrial ambition with its measurement logic. If deployment and strategic autonomy are genuine goals, then the evaluation system must weight deployment and industrialization.

The uncomfortable reality is that many ecosystem actors were rationally optimizing under the previous regime. They did what the metrics rewarded.

Now the metrics are evolving.

The question is not who will lose their position. The question is who understands that metrics define system behavior — and is willing to redesign their internal architecture accordingly.

The Structural Imperative

Deep tech breakthroughs do not conform to five-year political cycles. They rarely align with quarterly reporting logic. They compound slowly, across regulatory layers, infrastructure build-out, capital stacks, and industrial integration.

When measurement systems ignore this temporal structure, the outcome is predictable: ecosystems rich in visible activity and poor in durable industrial consequence.

Europe’s recent policy adjustments suggest a growing recognition of this mismatch. The language is shifting. Impact pathways are becoming explicit. Leverage ratios are scrutinized. Valorisation is no longer an afterthought but a condition embedded in funding logic.

If this shift holds, it signals more than administrative refinement. It marks a recalibration of the innovation system’s gravitational field. What is funded, what is scaled, and what survives will increasingly be determined by whether outcomes materialize in the real economy rather than circulate within reporting cycles.

Visibility will no longer be sufficient.45

Materialization will determine legitimacy.

The transition will not be loud. It will unfold through evaluation criteria, capital sequencing rules, and portfolio construction decisions. But structural changes rarely announce themselves. They alter incentives quietly, and the system reorganizes around them.

Those who understand that optimization variables define outcomes will adapt early. They will redesign their funding architectures, governance models, and venture-building practices before external pressure makes that redesign unavoidable.

The question is no longer whether Europe supports innovation. It is whether its measurement architecture is aligned with the physics of deep tech deployment.

The answer depends on whether institutions are prepared to treat sequencing, capital design, and temporal alignment not as operational details, but as strategic determinants. Or if they opt-out to become forgotten soon…

Future editions will move from diagnosis to architecture, unpacking the evaluation models, capital logic, and deployment frameworks shaping science-based innovation.

Full framework analyses and applied system designs are available to members.

Selected References:

Every year, the Financial Times publishes its Europe’s Leading Start-up Hubs list, an exercise in comparison that names 180 accelerators and incubators across the continent. In the 2026 edition, hubs like UnternehmerTUM in Munich, Start2 Group, and BayStartUp occu…

Read more

A new European Life Sciences Coalition has been launched to strengthen investment coordination and competitiveness across the region (Invest Europe).1

The visible narrative is clear.

• Europe needs more capital.

• More alignment.

• More …

Read more

The study Wachstumspfade für Deutschland1 conducted jointly by BCG and UnternehmerTUM argues that Germany can access approximately €1.7 trillion in serviceable deep hightech (yes, that’s actually deep tech we’re talking about here) market potential by 2030 across…

Read more

Europe keeps repeating the same diagnosis: we need more patient capital. When still nothing changes, that’s because it’s bullshit.

The implication is psychological. Investors are too short-term. Funds are too impatient. If only capital waited longer, deep tech would scale.

This framing is convenient for the comfort zone. Nothing good ever arose from there.

Europe has financed capital-intensive, science-based industrial transformation for decades without venture capital as the coordinating instrument. Engineering SMEs, advanced materials firms, photonics companies, industrial biotech manufacturers, and precision instrumentation leaders were built long before VC became ecosystem default (35+ years, 50+ real case studies, and analysis for be found inside the Beyond Venture Capital - Industry Report, Witte, 2026, compressed results below)1.

They did not rely on “patience.”

They relied on capital architecture.

They stacked:

• early revenue from institutional customers
• R&D contracts and co-development fees
• licensing income
• strategic corporate partnerships
• bank debt and export finance
• public grants as validation capital
• retained earnings as scale capital

The problem today is not impatience.
It is mis-sequencing.

Venture capital is optimized for power-law outcomes within a finite fund lifecycle. Deep tech is optimized for technical validation, institutional adoption, and long integration cycles (Witte, 2026).

When VC logic is applied indiscriminately, two failure modes appear:

• premature commercialization
• mid-TRL stagnation when equity dries up

Calling for patient capital externalizes responsibility. It assumes the investor temperament is wrong.

The deeper issue is that many ventures, institutions, and funds apply a single financing script to technologies that do not mature in venture-shaped curves.

In the full analysis, I will map:

• why “patient capital” is a moral framing of a structural design error
• how European science SMEs historically sequenced capital without VC
• what capital stacking actually means in practice
• why architecture precedes fundraising
• and why repeating the patience narrative delays structural reform

If you operate in deep tech, this is not theoretical. It determines whether your organization supports deployable infrastructure or a stranded pitch deck.

Read more

Europe and the United States do not simply fund science differently. They encode risk differently.

In the US, knowledge is externalized early. Founders are expected to articulate ambition before certainty. Visibility is treated as leverage. Narrative becomes a capital instrument.

In much of Europe, knowledge is stabilized before exposure. Legitimacy is accumulated institutionally before public amplification. Visibility without validation is often penalized.

This difference is not stylistic. It is structural.

It determines:

• how quickly capital concentrates around emerging technologies
• how early ventures are allowed to scale probabilistic science
• who is permitted to define viability
• how mid-stage stagnation emerges
• why certain regions repeatedly lose industrial leadership despite scientific excellence

Recent data shows that Europe produces world-class research output and channels nearly half of venture investment into deep tech sectors (McKinsey, 20241; Dealroom, 20242). Yet scale-up gaps persist. European ventures are statistically more likely to stall between TRL 4 and TRL 6 compared to US counterparts (BCG, 20233).

This is not primarily a funding gap.

It is a sequencing gap.

And sequencing is not an accident. It is enforced by institutions.

This analysis does not ask whether Europe is “less bold” or whether the US is “reckless.” It examines how risk is socially signaled, financially priced, and institutionally gated.

In the full analysis, you will see:

• how signaling theory translates into capital velocity
• why public gatekeeping compresses optionality
• how private capital acts as early legitimacy proxy
• where European mid-TRL stagnation is structurally manufactured
• why sovereignty debates ignore upstream capital sequencing
• and how your own decision-making reinforces one of these architectures

If you operate in deep tech, you are not observing this system from outside. You are participating in it.

The question is whether you recognize which risk architecture you are (re)producing. This reveals below:

Read more

Irreversible Baseline Change

Deep tech is best understood not by listing specific technologies (AI, biotech, quantum, etc.), but by the nature of its impact on fundamental systems. It refers to innovations that upend the foundational state of what is possible in a domain, rather than incremental improvements to existing products. In other words, there is…

Read more

A pattern founders notice before it becomes visible

Over the past years, several deep tech founders have told me, often casually and without much emphasis, that they deliberately avoid university startup centers.

They do not frame this as opposition to academia or public support. It usually comes up as a practical adjustment, mentioned alongside other dec…

Read more