A talent race can be measured at three different moments: where researchers were trained, where they work now and which countries gain or lose people between those points. China’s recent rise looks overwhelming under the first measure, striking under the second and more complicated under the third. The apparent contradiction is the useful part.[1][2]
The 2025 NeurIPS talent sample
| Measure | Value | Context |
|---|---|---|
| Undergraduate education in China | 57% | Up from 46% in the 2022 cohort. Undergraduate institution is used as a proxy for origin, not nationality.Source 1Source 2 |
| Workplace in China | 41% | Up from 27% in 2022 and ahead of the United States at 34% in this sample.Source 1 |
| Retention of Chinese-trained researchers | 69% | Up from 57% in 2022.Source 1Source 2 |
| Retention of US-trained researchers | 89% | Still higher than China’s retention rate.Source 1 |
| Net US talent gain | 2,145 researchers | The United States continued to attract more sampled researchers than it lost.Source 2 |
The base of the pyramid is domestic training. More than half of the tracked elite researchers had completed their undergraduate education in China, compared with 13% in the United States. That does not tell us where the most consequential model will be built, but it says that China has created a very large pool from which universities, start-ups and established technology firms can recruit.[1][2]
Retention changes the meaning of production
A country can educate researchers and still lose much of the value if the strongest people leave. China’s retention rate rose from 57% to 69% between the two cohorts. Nature reports that better domestic career options are part of the explanation, while restrictions and uncertainty around mobility to the United States may also matter. Yet the United States remains a magnet: it still attracted a net inflow, and roughly one fifth of the Chinese-trained cohort moved there for graduate study.[1]
Three different kinds of advantage
| Capability | China’s current strength | US current strength | Why the distinction matters |
|---|---|---|---|
| Training pipeline | Very large undergraduate-origin share | Smaller domestic origin share | Sets the long-run supply of researchersSource 1Source 2 |
| Retention and workplace concentration | Rapidly rising retention; largest workplace share in the sample | Higher retention of US-trained researchers and positive net inflow | Determines where teams and institutions accumulateSource 1Source 2 |
| Frontier compute and private capital | Large domestic ecosystem but constrained access to some advanced chips | Stronger access to frontier hardware and much larger private investment | Shapes how quickly talent can turn ideas into the largest training runsSource 1 |
Talent does not erase the hardware stack
Nature’s broader account places the talent shift beside a different balance of power. The United States retains advantages in advanced computing hardware, frontier-model infrastructure and private investment. A researcher can design an excellent method with modest resources; training and serving the largest systems require chips, networking, electricity, data centres and capital at a very different scale. China’s export-control constraints therefore sit near the top of a pyramid whose human base is expanding quickly.[1]
The most durable advantage may be institutional density rather than one leaderboard position. A large training pipeline creates more laboratories, mentors, start-ups and specialised teams. Compute and capital decide which projects can scale furthest; talent decides how many credible projects exist to scale. The global competition is therefore not a simple count of researchers against chips. It is a race to join the whole pyramid together.
Sources & further reading
- China’s AI research is booming: what it is like to be on the front line Nature · 14 Oct 2026 · Accessed 16 Oct 2026
- China now produces 57 percent of the world’s elite AI talent Information Technology and Innovation Foundation · 13 Oct 2026 · Accessed 16 Oct 2026