Beijing unveils 2026–2028 AI plan to build 'AI scientists'

Beijing’s new AI‑for‑science plan is a clear bid to hard‑wire frontier AI into the city’s basic research apparatus. Instead of just funding generic models, the plan calls for autonomous “AI laboratories,” scientific‑domain models, and shared high‑quality datasets across materials, biomedicine, quantum, breeding and more. That is essentially a blueprint for building AI co‑pilots—and in some cases AI agents—that sit inside the experimental loop from hypothesis to lab execution.

From an AGI perspective, this is a quiet but important shift. A lot of cutting‑edge capability will now be justified as “AI to accelerate science,” which is politically easier to fund and shield from export controls than overtly general‑purpose AGI projects. If successful, Beijing will grow a generation of systems that are extremely capable in narrow scientific domains, but whose architectures, tooling and data pipelines are highly reusable for more general reasoning tasks. It’s a way to advance the frontier while wearing a lab coat instead of an AGI badge.

The competitive implication is that scientific AI is becoming an explicit policy race. Other hubs—Boston, the Bay Area, Paris, Cambridge—are doing similar things through grants and institutes, but Beijing is now coordinating this at a municipal level with a three‑year roadmap. That will push both Chinese and foreign labs to ask how much of their AGI stack they can justify as “AI for science” to secure funding and political air cover.