On September 7, China's Ministry of Industry and Information Technology (MIIT) issued the "15th Five-Year Plan for the Development of the Information and Communications Industry," which for the first time writes intelligent computing power into its headline targets: by 2030, the country's intelligent computing capacity is to reach 9,800 EFLOPS, and cumulative investment in information infrastructure between 2026 and 2030 is to total 3.8 trillion yuan.
The starting point looks like this. As of the end of June, China's intelligent computing capacity was 2,185 EFLOPS, up 177% year over year; by the end of July the figure was roughly 2,450 EFLOPS. In other words, the plan calls for more than quadrupling that total within four and a half years.
What's on the list of targets
The plan lays out 13 headline indicators in total, including 4.1 trillion yuan in information and communications industry revenue, 7% average annual growth in total telecom business volume, 50 5G base stations per 10,000 people, and 95% 5G user penetration. Placing intelligent computing power alongside these established metrics is itself a shift — in this kind of five-year plan over the past five years, compute has typically played a supporting role, with base-station counts and subscriber numbers occupying the headline slots.
Two supporting measures are spelled out in more concrete terms: the orderly deployment of intelligent computing clusters at the 10,000-GPU and 100,000-plus-GPU scale, and strengthening the compatibility between infrastructure and domestically made AI chips. According to public information, China has already built roughly 52 intelligent computing centers at the 10,000-GPU scale or above.
The second point carries real weight. Making infrastructure compatible with domestic chips means rebuilding scheduling, compilation, networking and storage around non-Nvidia hardware — a task whose difficulty has shown up repeatedly over the past two years in the rollout pace of multiple domestic clusters.
The milestones that came before this one
Read on its own, this plan can look like a single statement of intent. Strung together with what came before, it reads differently.
In June, a compute-network plan led by China's National Development and Reform Commission (NDRC) was reported by foreign media — roughly 2 trillion yuan in scale, aiming to build a nationwide, interconnected compute network within five years, with a domestic-chip usage requirement mentioned as part of it. At the NDRC's monthly briefing on July 31, officials said intelligent computing power had grown 2.8x year over year in the first half, that cumulative global downloads of domestic large language models had passed 10 billion, and that China would "accelerate the legislative process for an AI law." In July, Sugon 8000, a fully domestic 100,000-GPU cluster, went into operation.
These come from different agencies and different documents, and the investment figures don't overlap — the NDRC's compute-network plan is priced separately from MIIT's 3.8 trillion yuan, which covers information infrastructure as a whole, including 5G, fiber and satellites. Adding the two numbers together doesn't work. What they share is a direction that keeps compounding: first funding and networks, then chip self-sufficiency, and now a total compute target written into a five-year plan's headline indicators.
How to read the target
The 9,800 EFLOPS figure needs some care in how it's compared. Intelligent computing power is typically tallied using low-precision conversions, and the baselines different organizations disclose aren't fully consistent, which limits the value of cross-country comparisons. More telling is the shape of its own growth curve: getting from 2,185 to 9,800 requires a compound annual growth rate of roughly 35% to 40% (a rough estimate, counting from mid-2026 to the end of 2030), a notch below the 177% year-over-year figure of the past year.
That doesn't make the target unambitious — it's that the base has already climbed. The real constraints sit elsewhere: how much power is available, whether advanced-node manufacturing capacity gets allocated to this, and whether actual utilization rates on domestic chips can keep up with their rated specs. The plan text itself doesn't answer these three questions, and the accompanying implementation rules haven't been made public yet.
Sources: Ministry of Industry and Information Technology, "15th Five-Year Plan for the Development of the Information and Communications Industry," South China Morning Post, CocoLoop, National Business Daily; intelligent-computing stock and growth figures verified against MIIT's public data, with investment totals and the 13 indicators per the plan text.