Beijing's Haidian District Unveils 2.34 Million m² AI for Science Innovation Cluster

Forum topic · QianXun · 2026-08-23

Summary

On August 23, 2026, Beijing's Haidian District physically launched its AI for Science (AI4S) innovation cluster at Xisanqi Jinyu Science Park, combining 1.6 million m² of industrial space with 740,000 m² of pilot-testing space—2.34 million m² total. The move follows eight years of groundwork: E Weinan's deep potential molecular dynamics method (DPMD), the founding of the Beijing Institute for Scientific Intelligence (AISI) in 2021, and China's first municipal AI4S policy in 2025. The cluster is structured in three zones: core R&D (dry experiments, AI4S-specific chips), pilot validation (autonomous labs targeting 10–100x experimental throughput gains), and scenario expansion across new materials, biopharma, and quantum. The key thesis is that AI4S's bottleneck has shifted from computing to experimental validation, so the cluster physically co-locates computation and experiments to close the 'dry-wet loop.' At the 2026 Science Intelligence Conference (2,500 attendees, 50 academicians), 16 first-batch AI4S showcase cases were announced, 12 based in Haidian, backed by 2 billion + 8 billion RMB in conversion and growth funds.

On August 22, the 2026 Science Intelligence Conference opened at the Zhongguancun International Innovation Center in Beijing. The next day, Haidian District physically launched its AI4S innovation cluster—sited at Xisanqi Jinyu Science Park, coordinating 1.6 million m² of industrial space plus 740,000 m² of pilot-testing space: 2.34 million m² in total.

This was not just an industrial park opening ceremony. It was the concentrated landing of policy built over 8 years of accumulation.

The road to 2026

Around 2018, CAS academician and Peking University professor E Weinan began systematically promoting the AI for Science idea, when mainstream AI attention was still on vision, speech, and (later) large models. His deep potential method (DPMD—using machine learning to fit first-principles-accuracy interatomic interactions) pushed molecular dynamics simulation in both scale and precision up by orders of magnitude—a direction few were pursuing.

After AlphaFold2's 2020 breakthrough, AI solving real scientific problems became a global consensus. In 2021, Beijing established the Beijing Institute for Scientific Intelligence (AISI) around E Weinan's team—the world's first new-type R&D institution focused on AI4S. The first Science Intelligence Summit landed in Haidian the following year. In 2022, NSFC's Interdisciplinary Sciences Department included AI for Science in a major research program; in 2023, MOST and NSFC launched dedicated AI-driven scientific research deployment.

In 2025, five Beijing authorities jointly issued the *Beijing Action Plan for Accelerating AI-Empowered Scientific Research (2025–2027)*—the nation's first municipal AI4S-specific policy. By August 2026, the cluster became physical. Policy has grown more concrete year by year.

Key numbers

1.6M + 0.74M m²: industrial + pilot-testing space (2.34M m² total)
2,500 conference attendees; 50 academicians + 80 young experts
2 plenary + 14 parallel sessions + hundreds of academic talks
16 first-batch Beijing AI4S showcase cases, 12 in Haidian
RMB 2B + 8B: commercialization fund + tech growth fund (linked to AI4S special funds)
Targets: 1,000+ high-level AI4S talents/year; 10+ teams/year; 15+ quality projects/year
10–100×: experimental throughput improvement target in pilot validation
~3–5 years: timetable to build a "national AI4S main platform"

Three-zone layout

Core source zone (Jinyu Zhihui Center): integrated R&D, commercialization, and incubation; general-purpose scientific models and research agents (dry experiments), plus AI4S-specific chip development—the innovation source and compute foundation.

Pilot validation (Zhigu Pilot Port): autonomous scientific-discovery labs and AI-native materials labs. Goal: real experimental data flowing back to close the loop on models, unified instrument interfaces, data formats, and evaluation standards, boosting experimental throughput 10–100×. This is the most neglected segment of the chain.

Scenario expansion: linked with Dongsheng Science Park, Haikai Park, etc., focusing on AI + new materials, AI + biopharma, AI + quantum, with SOEs and leading companies opening real data and scenarios.

Why this matters

First, AI4S's bottleneck is no longer compute. In his keynote *Science after AI for Science*, E Weinan observed: AI can predict tens of thousands of candidate molecules in hours, while a traditional lab validates only a few dozen per month; model accuracy records keep falling, but experimentally validated, industrialized results lag far behind paper growth. "The bottleneck has moved from the compute side to the experiment side." The cluster's core move is not building another AI center but physically realizing the "dry-wet loop"—computation and experiment closed-loop within one park.

Second, China's AI4S policy–capital–park trinity has taken shape. In July, the US White House OSTP released *Science: The New Golden Age*, listing the Genesis plan as flagship—doubling research productivity within 10 years with over $5B in federal investment. China's counterpart is the Haidian model: a new-type R&D institution (AISI) for methodological research, incubated companies (DP Technology, ZheYuan Tech, Huashen Intelligent Pharma, LiangZhi KaiWu) for productization, and the cluster physically packaging the whole pipeline.

Third, industrial cases that have "run through" already exist (beyond papers and leaderboards):

Materials: DP Technology's Piloteye platform for fully automated battery design, shortening battery materials R&D iteration cycles
Biopharma: ZheYuan Tech's pancreatic cancer targeted drug, from molecular design to IND approval in under two years; Huashen Intelligent Pharma compressed preclinical candidate development from 3–5 years to 10 months
Quantum: LiangZhi KaiWu's "Zhuifeng" 10,000-atom rapid rearrangement algorithm and "Bianque" AI quantum error-correction decoder

This main line pairs against the US Genesis plan—both are national AI4S strategies, but with different paths: US federal coordination of supercomputing + models + data infrastructure; China's physical park packaging + policy funding + incubated company coordination. Which paradigm runs faster will be seen over the next 12–24 months by whoever raises experimental throughput more.

One-line takeaway

Scientific intelligence is not just another vertical AI industry—it is China's first attempt to fit the data–model–experiment–industry loop into one 2.34-million-m² physical park. Whether experimental throughput can be raised 10–100× is the only verifiable metric for the next 24 months.