The brain drain nobody talks about

Here's a number that should be getting more attention: the number of AI researchers and developers moving to the United States has dropped 89% since 2017. And 80% of that decline happened in just the last year. While the tech world debates which model tops the latest benchmark, the people who actually build these models are quietly going somewhere else. This isn't a hypothetical risk. It's already happening, and the data from the Stanford HAI 2026 AI Index Report makes it hard to ignore.

The money is there, the people aren't

On paper, the US looks untouchable. Private AI investment hit $285.9 billion in 2025, more than 23 times China's $12.4 billion. The US also led in entrepreneurial activity, with 1,953 newly funded AI companies in 2025, more than 10 times the next closest country. Global corporate AI investment more than doubled, and generative AI alone grew over 200%. But capital without talent is just money. You can build all the data centers you want. You can pour billions into compute infrastructure. None of it matters if the researchers who push the frontier forward are choosing to work somewhere else. Investment buys capacity. Talent builds capability. The US is scaling one while losing the other.

Why researchers are leaving

The reasons are structural, not ideological. Several forces are converging at once. Visa friction remains a persistent bottleneck. The US immigration system was never designed for the speed at which AI talent moves. Multi-year processing times, lottery-based H-1B allocation, and restrictive per-country caps make it genuinely difficult for researchers to plan careers in the US. Other countries have noticed and are offering streamlined alternatives. The political climate has shifted. Funding cuts to federal research agencies, political pressure on universities, and a general atmosphere of uncertainty have made the US a less attractive destination for international scientists. A 2026 New York Times investigation found that countries across Europe are actively recruiting researchers who feel pushed out of the American system. France's Aix-Marseille University launched dedicated programs for what they call "refugee scientists." Remote-first labs changed the equation. You no longer need to be in San Francisco to do cutting-edge AI work. Distributed teams, open-source collaboration, and cloud compute mean that a researcher in Berlin, Singapore, or Bangalore can contribute to frontier research without relocating. The geographic premium the US once held has eroded. Better opportunities are emerging elsewhere. China, in particular, is offering high salaries, research funding, and housing support to attract AI talent. According to The Economist, China now has more active AI researchers than America, Britain, and Europe combined, and its cohort skews younger, with 47% being students compared to about 30% in the West.

China's open-source flywheel

China's rise in AI isn't just about government spending. It's about community. Since DeepSeek released its R1 reasoning model in January 2025, Chinese companies have repeatedly delivered AI models that match the performance of leading Western models at a fraction of the cost. Alibaba's Qwen family, Moonshot AI, and MiniMax now dominate worldwide usage rankings on platforms like HuggingFace and OpenRouter. The numbers tell a clear story. Chinese institutions produced roughly 23,700 AI publications in 2024, with citation share exceeding 40% in key datasets. China's share of the top-100 most-cited AI papers grew from 33 to 41 between 2021 and 2024. A US congressional advisory body warned in March 2026 that China's open-source dominance is creating a "self-reinforcing competitive advantage." This matters for talent because open-source ecosystems are talent magnets. When your models are open, researchers around the world can download them, build on them, and contribute back. It creates a virtuous cycle: better models attract more contributors, which produce better models. China's open-weight strategy isn't just a product decision, it's a talent acquisition strategy.

The export control irony

Here's the part that doesn't get discussed enough. US export controls on advanced AI chips were designed to slow China's progress. In practice, they may have accelerated exactly the kind of innovation they were meant to prevent. Faced with hardware constraints, Chinese labs got creative. DeepSeek's R1 achieved near-frontier performance on smaller compute budgets, proving that algorithmic efficiency could compensate for chip limitations. As Brookings noted, "starving China's supply of US-designed AI chips will push China to more effectively develop its own AI chip capacity." The restrictions also had a second-order effect on talent flows. By creating uncertainty about supply chains and cross-border collaboration, export controls made it riskier for Chinese researchers to build careers dependent on US infrastructure. The result: more talent staying in or returning to China, and more reason to build self-sufficient research ecosystems. This isn't to say export controls are inherently wrong. But the assumption that restricting hardware would create a durable advantage ignored how talent and innovation actually respond to constraints.

Distribution beats concentration

I've written before about how distribution beats product, the idea that getting something into more hands matters more than having the best version of it. The same logic applies to talent. The US approach has been concentration: attract the best people to a small number of elite institutions and companies. This worked when the US was the only game in town. But now the playing field is flattening. Remote work, open-source, cheaper compute, and proactive talent policies from other governments mean that AI capability is distributing globally. This isn't necessarily bad for the world. A more distributed AI ecosystem is probably more resilient, more diverse in its applications, and less dependent on any single country's political decisions. But it's a fundamental shift from the model that gave the US its dominance in the first place.

The Singapore play

Small countries that move fast on talent policy have a disproportionate opportunity here. Singapore is a case in point. The government committed S$1 billion over five years for public AI research. It's replacing its Tech.Pass with a new ONE Pass (AI and Tech) track, a five-year renewable work pass specifically targeting AI and tech leaders. It's building Kampong AI, a dedicated startup community modeled on the Block71 hub that produced companies like Carousell and Carro. Singapore's advantages are structural: English-speaking, politically stable, positioned between East and West, and small enough to update policy quickly. In a world where the major powers are creating friction for talent movement, a nimble, welcoming country can punch well above its weight. The question isn't whether Singapore can compete with the US or China on raw scale. It can't. The question is whether it can become the place where talent goes when the big players make it hard to stay.

What this means going forward

The 89% decline isn't a blip. It reflects a structural shift in how AI talent thinks about where to work and build careers. The forces driving it, visa friction, political uncertainty, remote-first norms, competitive offers from other countries, aren't going away. The US still has enormous advantages: the deepest capital markets, the most mature startup ecosystem, and many of the world's best universities. But advantages decay when you stop maintaining them. And right now, the US is spending heavily on the infrastructure of AI while neglecting the human infrastructure that makes it work. Capital follows talent, not the other way around. The countries and institutions that understand this will shape the next decade of AI. The ones that don't will wonder where all the researchers went.