Attention (And Money) Is All You Need: Why Universities Are Struggling to Keep AI Talent

In 2013, Nobel Laureate Geoffrey Hinton (popularly known as the godfather of artificial intelligence, AI), then a professor at the University of Toronto and working on the frontier of AI research, sold a startup built out of his academic lab to Google and moved to the firm together with two of his graduate students, marking the beginning of a broader reorganization of AI research away from universities and toward large tech firms. 

Since then, media reports have detailed some of the salaries available to “superstar” academics in the field, but many questions remain about this allocation of talent from the academy to the private sector: Who are the AI scientists? How large is the pay gap between academia and industry? Who leaves academia, at what stage of their career, and where do they go? And how does the move change what researchers produce in terms of papers, patents, and knowledge spillovers?

To address these questions, the authors employ new microdata that link academic publication records to administrative employer–employee data from the US Census Bureau. They track the employment, earnings, and research output of 42,000 AI researchers over two decades to reveal the following 10 facts: 

1. By 2019, 68% of AI researchers worked in industry, up from 48% in 2001.
2. The share of AI researchers born in the United States and working in industry has declined by 5.5 percentage points (pp); this decline is almost entirely accounted for by a rise in the share of researchers from China (+3.8 pp) and India (+2.0 pp).
3. The median age of AI researchers in industry has fallen by 2 years (39 to 37) but has remained flat in academia (42 to 42).
4. The female share of AI scientists in academia has risen by 13 pp (16% to 29%) whereas in industry it has remained relatively flat (+4 p.p., 19% to 23%). Academia now has greater female representation than industry.
5. In industry, the ratio of female to male average earnings has modestly widened (73% to 72%); in academia, it has narrowed (79% to 83%).
6. Since 2001, average academic real salaries declined, and AI researchers in academia were no exception.
7. Coinciding with the image recognition revolution kicked off by AlexNet (Krizhevsky et al., 2012), top 1% earnings in industry exploded from $595,000 in 2001 to $1.94 million (measured in 2015 dollars). Top academic salaries barely budged ($301,000 to $392,000). 
8. Coinciding with the publication of the paper, “Attention Is All You Need” (Vaswani et al., 2017), transitions from academia to industry accelerated, and a growing share of AI academics’ income began to come from secondary employment.
9. Rising job transitions from academia to industry are driven by young researchers leaving to incumbent firms (firms with greater than or equal to 1,000 employees and greater than or equal to 20 years old) and to the Professional Services and Information sectors.
10. After researchers permanently transition from academia to industry, on average their paper-writing declines (65% fewer papers per year, 30 pp less likely to publish a paper), patenting increases (530% more patents per year, 6 pp more likely to patent), and their earnings rise by 63% relative to similar job switchers within academia.

If top faculty increasingly face industry offers that universities cannot match, institutions must reconsider how they allocate resources, structure incentives, and manage research personnel. … Universities are not passive bystanders in this transition; they are active participants whose policies will shape the equilibrium allocation of talent.

These facts reveal a structural reorganization of the supply side of frontier innovation. AI research in the United States has shifted from a university-centered ecosystem to one increasingly dominated by large incumbent firms, with implications beyond labor reallocation to include issues relating to how knowledge is produced, diffused, and appropriated. The authors cite four key considerations: 

• The supply of frontier innovation is deeply intertwined with global talent flows. The growing presence of foreign-born researchers, particularly from China and India, reflects the central role of immigration in sustaining US technological leadership. At the same time, the decline in the share of US-born researchers in industry raises questions about the long-run domestic pipeline of scientific talent. Demographics are not peripheral to innovation; they shape its scale, composition, and resilience.
• The movement of young researchers toward large incumbent firms alters the structure of knowledge spillovers, from open knowledge platforms to patenting. This shift implies a reorientation from open science toward proprietary innovation. When frontier scientists concentrate in large firms, knowledge diffusion may become more mediated by organizational boundaries and intellectual property regimes. The resulting equilibrium may feature both faster commercialization and weaker knowledge spillovers.
• Questions about market concentration and competition are increasingly important. The superstar pay observed in AI is concentrated in large, compute-rich organizations, meaning that innovation may become increasingly tied to market power and access to proprietary data and infrastructure. This raises the possibility that competition in ideas is increasingly shaped by competition in compute, capital, and scale. The organization of innovation, in other words, is becoming inseparable from industrial structure.
• The infrastructure requirements of modern AI (e.g., massive datasets, specialized hardware, and large-scale cloud computing) have shifted the comparative advantage away from universities. The post-deep learning era is characterized by fixed costs that few academic institutions can match. As frontier research becomes more capital intensive, the boundary between science and industrial production blurs. This reorganization suggests that infrastructure policy, ranging from public compute provision to data access regimes, may be as important as traditional R&D subsidies in shaping the trajectory of innovation.

The authors speak directly about the implications for academia: Flat or declining real academic salaries, combined with rising outside options, place pressure on traditional university budget models built around teaching cross-subsidization and grant funding. If top faculty increasingly face industry offers that universities cannot match, institutions must reconsider how they allocate resources, structure incentives, and manage research personnel. The governance of faculty effort, intellectual property, and industry collaboration becomes central to the future of academic science. Universities are not passive bystanders in this transition; they are active participants whose policies will shape the equilibrium allocation of talent.

Written by David Fettig • Designed by Maia Rabenold