The global race for artificial intelligence supremacy is often framed through chip supply chains and model benchmarks, but a new report from Carnegie China suggests the real bottleneck is human capital. Released in late September 2026, the study titled "Who’s Ahead in the Global AI Talent Race?" indicates that China has overtaken the United States as the leading work location for elite AI researchers. This shift marks a significant pivot from the narrative that dominated the post-ChatGPT launch era, where American labs held an undeniable advantage in frontier model development.

Methodology Meets Automation

What makes this report particularly relevant for developers and infrastructure builders is its production method. Carnegie China revived the defunct MacroPolo "Global AI Talent Tracker" using a hybrid workflow that leaned heavily on AI coding agents. The team employed Claude Code and OpenAI Codex to automate the scraping and analysis of data from OpenReview, the platform hosting NeurIPS submissions. By applying these tools to a sample of over 10,000 authors—roughly 40 percent of the total 2025 cohort—they achieved a scale of analysis previously impossible with manual coding, allowing for a more robust longitudinal study of researcher migration patterns.

The Data Behind the Shift

The analysis focuses on authors of accepted papers at the Neural Information Processing Systems (NeurIPS) conference, widely considered the "Navy Seals" of AI research. For the 2025 conference, NeurIPS accepted 5,823 papers with 25,677 unique authors. The study tracked the undergraduate origins, graduate training, and current workplaces of these researchers. The results show that while the US remains a massive magnet for talent, China is now producing and retaining more of the world’s top AI scientists. This trend is visible in the net gains and losses when comparing undergraduate origin countries to current workplace locations, with China showing strong retention and growth in domestic employment for elite researchers.

Quality Control in an AI-Generated World

The report’s use of AI for data processing highlights the growing role of LLMs in serious academic and policy research. However, the authors emphasize that human intervention remained critical for quality control. During the back-testing of the 2022 cohort, the automated workflow initially misidentified 1,705 authors (6.6 percent) as having no profile due to renamed or merged OpenReview accounts. This error was caught and corrected through a "stop, diagnose, correct, and rerun" process involving both human researchers and AI code reviewers. The final dataset passed eighteen automated consistency checks, though the authors note that confidence is higher in aggregate counts than in individual career profile accuracy.

Key Takeaways

  • China has overtaken the US as the leading work location for elite AI talent in the 2025 NeurIPS cohort.
  • Carnegie China used Claude Code and OpenAI Codex to automate data scraping and analysis for over 10,000 researchers.
  • Europe’s position is ambiguous, with conflicting signals in different sample subsets suggesting a potential faltering in talent retention.
  • The study highlights the increasing viability of AI-assisted research workflows, provided rigorous human oversight is maintained.

The Bottom Line

While the hardware and model benchmarks get the headlines, the talent pipeline is the true chokehold on AI dominance. Carnegie’s use of AI agents to track this pipeline proves that human-in-the-loop verification is not just best practice, but a statistical necessity for reliable data.