Technology diffusion: The role of venture capital, universities and China

You can listen to this podcast on Spotify, Apple Podcasts, or wherever else you get your podcasts. You can also watch this conversation on YouTube, and we have started a Substack for this podcast.

Even if AI progress ground to a halt today, we would still spend years uncovering the benefits, adopting these tools at work, and integrating them into our day-to-day lives.

The pace of innovation and progress in AI has been rapid. But there is a gap between this innovation and real world impact. In this episode of Ideas in Development, we were joined by Josh Lerner for a wide ranging conversation on technology diffusion, the role of venture capital, universities, and China in this process, and whether AI will follow historical patterns.

Technology diffusion: Lessons from hybrid corn to electricity

Josh begins by discussing a famous example of impact lagging behind innovation: hybrid corn. Zvi Griliches’s seminal work tracked adoption across US agricultural regions and found that even when an innovation is clearly better, its spread can still be slow and geographically uneven.

“even though this hybrid corn was much better than traditional corn, the process of its spread was really quite slow.”

A few places adopted quickly, others lagged for years, and the reasons weren’t simply ignorance. Diffusion ran into a range of local constraints: knowledge, habits, risk tolerance, complementary inputs, and social learning.

That lesson from Zvi’s work in a 1957 issue of Econometrica generalises. ANd Josh also highlights that once the electric motor arrived, large productivity gains still required factories to reorganise themselves. It wasn’t enough to buy the motor; as firms had to redesign production lines, restructure supervision, and retrain workers.

Venture capital and the geography of innovation: Why hubs form and persist

Diffusion itself is necessary because tech innovation is extremely geographically concentrated to start with. Research has identified a host of factors explaining why, but establishing which are most important is tough.

“There’s a whole confluence of things that came together and really untying that Gordian knot and really figuring out what really is the secret sauce remains pretty challenging.”

Josh discusses the classic Marshallian forces of agglomeration, including suppliers, labour pooling, knowledge spillovers, and the way proximity accelerates exchange. He also adds two more modern pillars. One is academia – certain universities become unusually good at commercialising ideas. The other is venture capital, which has been divided extremely unevenly around the world – where capital concentrates it tends to generate a flurry of entrepreneurial activity.

But there is no single driver. And because these forces are tangled together it’s hard to isolate causality, which should caution policymakers who are looking for a simple lever to pull.

Measuring technology diffusion through jobs

Josh’s research shows that the best jobs and the most valuable activities often stay clustered long after a technology becomes widespread.

High-end roles “remain super concentrated, not just for a few years, but for multiple decades afterwards.”

Even when consumers everywhere benefit, the highest-value design, R&D, and product work can remain anchored in a few places. This distinction between access and upside matters for AI. While AI tools will diffuse widely, the largest rents and best labour market outcomes, may remain concentrated in a small number of hubs unless other regions build the institutional scaffolding to compete.

Job postings are a valuable way of tracking diffusion because they’re granular and comparable; individual job announcements can be broken down by region, industry, and firm. This is also what policymakers often care about most.

But they do not paint a complete picture, and we need to understand not just job creation, but job transformation and job destruction, and with current data it’s harder to trace losses than postings. So to understand AI diffusion and inform AI policy we need to build better indicators of how tasks change inside occupations, and develop credible ways to track displacement geographically and sectorally, not just overall employment.

China, venture capital, and spillovers in technology diffusion

Venture capital used to be overwhelmingly US-centric, until China scaled rapidly and became a major hub of entrepreneurship and R&D.

After discussing Josh’s work tracing technological diffusion within the US, we move to his work on the rise of China as a global hub of innovation, which Josh’s work shows has stimulated entrepreneurship in other emerging markets across the world.

The mechanism is particularly interesting because it’s not a story of Chinese capital flooding into other regions. Instead, local entrepreneurs and investors in other countries began emulating business models forged under Chinese constraints, e.g. weak addressing systems, limited credit card penetration, and other different logistical realities.

This is diffusion through example and translation. Sometimes it’s a founder who spends time in China and comes home with an idea. Sometimes it’s a diaspora-linked investor who acts as a bridge.

This highlights the importance of South–South innovation linkages, because the constraints faced in one emerging economy are likely to be closer to those in another than to those in the US. That makes China’s role potentially relevant not just as a competitor to the US, but as a source of alternative business models and implementation strategies that travel.

How to build venture ecosystems that last

If venture capital matters, what can policymakers actually do? While governments often want a shiny programme, the real bottlenecks are frequently regulatory and institutional. Josh argues that if DFIs and governments want to accelerate technology diffusion, they should focus on (1) fixing the enabling environment, and (2) crowding in experienced investors, rather than trying to become investors of first resort.

“just throwing money at the problem is unlikely to be successful”

When labour laws, business registration, technology transfer rules, or stock option treatment make entrepreneurship structurally difficult, money alone is insufficient and unsustainable. Japan’s experience illustrates the risk of building a venture sector on subsidies rather than on fundamentals, as when government funding receded, activity collapsed because the wider environment still discouraged risk-taking.

So what does work better? Josh points to the logic of matching funds, using private investors to help reveal where opportunities actually are, and he highlights Israel’s Yozma model as a canonical case of public capital catalysing a durable private market rather than replacing it.

This is also where DFIs enter. In many emerging markets, DFIs have been crucial limited partners, backing funds that private institutional investors won’t touch yet. Josh is supportive, but he’s also clear about the design challenge.

“Absolutely, DFIs have been essential and will continue to be. But on the other hand, I think there are some right ways to do it and some not so right ways to proceed.

His concern is a mission drift towards direct investing, where incentives and bureaucracy can undermine the judgement-heavy, high-variance nature of venture decisions.

Universities and technology diffusion: Making research commercially relevant

Universities are often central to the diffusion story, not just as machines churning out human capital, but as institutions that shape whether ideas become companies, and whether regions become hubs.

Universities have been central to industries like the internet and biotech, but commercialisation capacity varies enormously. And importantly, that capacity appears to be partly place-based, not just about individual talent.

His most striking evidence comes from faculty moves between universities. Josh’s work finds “place matters a lot, that a professor moving from Duke to Stanford becomes much more commercially oriented”, and the reverse move reduces commercial orientation. Roughly 20–25% of the effect is attributable to the environment itself.

So it’s not simply a case of more research funding = more startups. Funding matters, but it’s not sufficient, as commercial relevance depends on technology transfer systems, norms around entrepreneurship, networks with investors, support for spin-outs, and the surrounding ecosystem that makes risk-taking feel feasible.

Three takeaways on AI diffusion

So where does all this leave us in terms of AI and development.

First, AI will not ‘arrive’ because models improve, as the binding constraints will often be organisational, e.g. data access, workflow redesign, training, incentives, procurement, and regulation.

Second, invest in the ecosystem. Regions that want high-value AI activity need credible risk capital and commercially engaged universities, plus policies that reduce the downside of experimentation (for example, around labour market flexibility, business formation, and stock options).

Third, get serious about measurement. We need better indicators of task transformation and displacement, and we need them with the same granularity we bring to job postings.