Why leading on AI innovation means thinking beyond 'market freedom' versus 'state funding'

A common debate – should governments fund breakthrough innovation or get out of the way – misses the key ingredient that separates high-functioning innovation economies from those that falter.

The decisive ingredient isn’t state spending or market freedom. It’s the strength of the institutional layer between them. This is the powerful engine that turns discoveries into industries.

Two announcements frame this debate: China is putting $138 billion into a state-backed venture fund to direct strategic technology investments. In contrast, the US president has proposed cutting its National Science Foundation funding by 55% – from $8.75 billion to $4 billion – while at the same time launching the $293 million Genesis Mission to apply artificial intelligence (AI) to more than 20 national challenges.

Together, these moves bookend the common debate. Beijing is betting that state-directed capital can pick the winners of the next technology wave. Washington is moving in the opposite direction by retreating from the broad-based science funding that has underpinned US innovation for 75 years, even as it places a smaller, mission-driven wager on AI.

Both wagers, however, run through institutions neither government is investing in. Neither side of the debate is complete without the institutional layer that helps make research commercially viable.

In a high-functioning innovation economy, the institutional layer is the critical engine between basic research and commerce. Made up of tech transfer offices (organizations within universities that aim to commercialize research), university-industry programmes, applied-research institutes, accelerators and founder-mentor networks, the institutional layer has a long, proven history of transforming discoveries into industries.

But this critical institutional layer is often left out of the discussion about whether government funding or market freedom is the best way to support breakthrough innovation.

Two strands of research address this gap. Josh Lerner’s 2009 Boulevard of Broken Dreams documents that most government attempts to engineer innovation underperform. Mariana Mazzucato’s The Entrepreneurial State from 2011 shows that public investment has nonetheless been central to developing the internet, biotechnology, mRNA vaccines and today’s AI.

Both explain that a strong institutional layer compounds investment – without it, investment dissipates.

This layer doesn’t build itself. Private actors underinvest in basic research at scale, a problem described by Nobel Laureate Kenneth Arrow in 1962. Universities, venture firms and industry operate on different time horizons and cannot coordinate on shared infrastructure that benefits all, but that no single actor will build alone. Returns are also spatially concentrated – only locally rooted actors can justify investing in the connective institutions that a region needs.

Neither state spending nor markets alone can solve these problems.

Silicon Valley is the case everyone knows, and usually misreads. The standard story is either that the market built it or that the US Defense Advanced Research Projects Agency (DARPA) built it. In fact, as Margaret O’Mara documents in 2019's The Code, Silicon Valley grew from institutional construction across four decades.

In the late 1940s, Fred Terman, Stanford University’s dean of engineering, set out to turn federal research investment into industrial activity as deliberate design, not happenstance. The Stanford Industrial (now Research) Park, established in 1951, put industry next door to academic research.

The Office of Technology Licensing, founded by Niels Reimers in 1970, designed the faculty-equity model later partially codified in the 1980 Bayh-Dole Act, which enabled ownership, patenting and commercialization of inventions developed with federal funding.

William Miller, later Stanford provost and president of Silicon Valley R&D firm SRI International, restructured the organization as an off-campus intermediary and co-founded the Stanford Program on Regions of Innovation and Entrepreneurship.

Federal research dollars flowed through this layer for 40 years before Stanley Cohen and Herbert Boyer seeded the biotechnology industry, and networking research at SRI led to the internet. Without the institutional layer, the same state investment would have produced papers. With it, state investment produced industries.

Silicon Valley was not a market triumph or a state-funded one, it was an intermediary-institutions triumph.

This architecture repeats across other, similar cases.

MIT’s parallel history to Silicon Valley was anchored by Ed Roberts, whose alumni-impact methodology recorded more than 30,000 alumni-founded companies employing 4.6 million people. A Stanford study into the university’s economic impact based on its involvement in entrepreneurship that I coauthored with Bill Miller identified around 40,000 companies and 5.4 million jobs that could trace their routes to the institution's activities in this area.

Asian systems show the same architecture at earlier stages. Taiwan, China’s Industrial Technology Research Institute (ITRI) did the transfer work universities and firms could not, spinning out the Taiwan Semiconductor Manufacturing Company (TSMC) in 1987. Tsinghua University’s TusPark network now spans more than 30 Chinese cities. Shenzhen has built a city-level institutional layer through municipal innovation funds, hardware supply chains and accelerators.

And 20 years of visits to Beijing have shown me how recognizable the architecture is. Whether at TusPark or open-source research community X-lab, the institutional fingerprints differ, but the bridging function does not. Academic research on China’s Project 985 reforms, which funded university research centres, finds the institutional changes – not the headline state spending – drove the high-tech entrepreneurship effects.

A recent wave of public and philanthropic instruments is building the intermediary infrastructure that forms this vital institutional layer.

The US National Science Foundation Regional Innovation Engines programme has unlocked tenfold leverage at the award stage, more than $1 billion in matching commitments against $135 million across nine engines spanning chips, AI, quantum and clean energy. Focused research organizations can target the bottlenecks universities and venture capital funds avoid, as does the UK’s Advanced Research and Invention Agency.

The private sector side has evolved in parallel through deep-tech venture capital, longer horizon vehicles, sovereign and corporate strategic capital. The question is whether the two layers will fit together or substitute each other.

AI is now stress-testing every innovation system. Its bottlenecks – compute access, evaluation infrastructure, safety research, frontier-model auditing – have no settled intermediary institutions yet. The Genesis Mission and National AI Research Resource are early attempts to build that layer in the US. AI itself may now reshape the layer as mentor matching, scouting and due diligence move increasingly within the reach of these models.

But the risk is that the tool is mistaken for the institution. Academic research on government efforts to spur new firms shows that opening entry does little by itself. Rather, the largest gains come when public action also provides resources and intermediaries. Effects compound over years, not quarters.

The real debate about fuelling AI innovation economies should not be about “more state” or “more market”, it should be about how to strengthen the institutional layer between them.

Any country that wants to lead on innovation must answer two questions: Who is building these connective institutions and sustaining them across political cycles? And is success measured in decades, as the historical cases have required, or in budget cycles?

Countries that get both answers right will lead the AI era. Those that continue arguing about state versus market will be left behind.