China Targets 9,800 EFLOPS of AI Computing Capacity by 2030
China's Ministry of Industry and Information Technology has published a five-year plan targeting 9,800 EFLOPS of intelligent computing capacity by 2030, up from 1,590 EFLOPS in 2025, supported by 3.8 trillion yuan of cumulative information infrastructure investment. The plan arrives the same week the US and China prepare their first dedicated AI safety dialogue.
On this page
The targets in the plan
China's Ministry of Industry and Information Technology has released the information and communication industry plan for the 15th Five-Year period, and its AI numbers are the largest committed anywhere. The plan targets 9,800 EFLOPS of intelligent computing capacity by 2030, up from 1,590 EFLOPS in 2025, a roughly sixfold increase in five years. English-language coverage, including the South China Morning Post, frames the target as a fourfold boost from mid-2025 levels, since China's intelligent computing capacity had already reached 2,185 EFLOPS by the end of June, up 177 percent year on year. [1] [3]
The plan lists 13 key indicators overall, including information and communication industry revenue of 4.1 trillion yuan by 2030 and average annual telecom business growth of 7 percent. [2]
How big the buildout really is
Behind the computing target sits an infrastructure commitment: 3.8 trillion yuan in cumulative information infrastructure investment by 2030, roughly 530 billion dollars at current exchange rates. That figure covers the networks, data centers, and power infrastructure that AI computing runs on. Bloomberg reporting adds that China is preparing around 2 trillion yuan, about 295 billion dollars, over five years specifically for a nationwide data center buildout. [2] [3]
For scale, the intelligent computing figure counts the specialized accelerator capacity that trains and serves AI models. Growing it sixfold in five years means building at a pace no other country has attempted, using a domestic stack that increasingly relies on Huawei Ascend silicon after US export controls cut off top-end Nvidia parts. [1]
Why compute is the strategic resource
Compute capacity has become the measuring stick of national AI capability because it is the input no lab can improvise. Model weights can be copied; training runs cannot happen without clusters. China's leading open-weight labs, DeepSeek, Alibaba's Qwen, and Z.ai, have spent 2026 shipping models that rival anything from US labs, and this plan is the state committing the electricity, land, and silicon to keep that pipeline fed. [1] [3]
The timing is also diplomatic. The plan dropped days before the first dedicated US-China AI safety dialogue is expected in Beijing, giving the Chinese delegation a concrete answer to any question about whether its AI buildout is slowing. It was not slowing; it is now the largest publicly stated computing target in the world. [1]
What it means for the model race
For people who run local models, the downstream effect is straightforward: the labs shipping the open weights you download are about to get proportionally more training compute, which historically converts into better models at every size class, from frontier systems down to the phone-capable releases that dominate on-device use. [3]
It also sharpens the contrast in governance approaches. The United States is pitching voluntary lab self-policing, Europe is regulating through the AI Act, and China is answering with state-directed capacity. Which strategy produces the best models, and under whose rules, is the question the rest of this decade will answer. [1]