The conventional framing of US-China AI competition focuses on compute (GPU access), data (training corpus size), and capital (research funding). These inputs matter. But they miss the decisive variable: the small number of elite researchers who drive fundamental breakthroughs.
Transformers were conceived by eight researchers at Google. GPT was developed by a small team at OpenAI. ResNet came from four researchers at Microsoft Research Asia. AlphaFold emerged from a focused group at DeepMind. The history of modern AI is a history of small teams with exceptional talent, not large organizations with massive resources.
Breakthrough AI: Small Teams, Big Impact
The Talent Pipeline
Elite AI talent doesn't appear spontaneously. It emerges from a pipeline: foundational education in mathematics and computer science, advanced training in machine learning, and integration into research ecosystems where breakthrough work happens.
Our Global AI Talent Tracker data reveals the structure of this pipeline with unusual clarity. Chinese universities produce the largest number of students who eventually become top AI researchers. American institutions host the largest number of top AI researchers during their productive years. The pipeline flows from one to the other.
"China trains them. America employs them. Both countries benefit from this arrangement — until one decides to end it."
The Policy Paradox
US policymakers face a paradox. The American AI advantage depends on attracting foreign-born talent, particularly from China. But security concerns about technology transfer motivate restrictions on that same talent pool.
The tension is real: some Chinese-born researchers have indeed transferred knowledge to Chinese institutions in ways that concern national security officials. But broad restrictions that deter the entire talent pool may cost more than they protect. The marginal security risk from one researcher must be weighed against the marginal capability loss from deterring hundreds.
China's Retention Challenge
China faces the inverse problem: it produces exceptional foundational talent but struggles to retain it through the most productive research years. The combination of higher salaries, better computing resources, more open research cultures, and deeper venture capital ecosystems makes American institutions the destination of choice for ambitious researchers.
Chinese policy has responded with retention incentives: the Thousand Talents Program, competitive research funding, and efforts to build world-class research environments domestically. These programs have had mixed success. Some elite researchers have returned; many have not.
The Third Path
A growing fraction of top AI talent is choosing neither the US nor China. Canada, the UK, Singapore, and the UAE have all invested heavily in AI research environments designed to attract global talent. For researchers wary of US-China geopolitical tensions, these alternatives offer compelling options.
The fragmentation of the global talent pool may ultimately harm both superpowers. If the US and China both implement policies that drive talent to third countries, neither achieves an advantage — they simply shrink the pie they're competing over.
Strategic Recommendations
The talent-centric view of AI competition suggests several strategic priorities:
- Streamline high-skill immigration: The US visa system is optimized for neither security nor talent attraction. Reforms that enable faster processing for vetted researchers while maintaining security review would serve both goals.
- Invest in domestic STEM education: Reducing dependence on foreign talent begins with producing more domestic talent. This requires sustained investment in K-12 mathematics, undergraduate computer science, and graduate AI programs.
- Make American institutions attractive: Competitive salaries, computing resources, and research freedom are more effective talent attractors than restrictive policies are deterrents.
- Build allied talent networks: If talent flows to third countries, ensure those countries are allies whose research ecosystems integrate with American ones.
The US-China AI race will not be won by export controls, GPU bans, or research restrictions. It will be won by whichever country becomes the most attractive destination for the small number of researchers who drive breakthrough innovations. In that race, welcoming beats restricting.