Beijing has put a number on its answer to America's chip controls: 9,800 exaflops.
China's Ministry of Industry and Information Technology published its 15th Five-Year Plan for the information and communications industry on September 7, setting a 2030 target of 9,800 eflops of intelligent computing capacity — up from a 2025 baseline of 1,590 eflops — and calling for 3.8 trillion yuan (about US$532 billion) in cumulative information infrastructure investment across the 2026-2030 period. The document, dated August 12 and released through the ministry's Information and Communications Development Department, lays out 13 headline indicators and 26 key tasks across six priority areas.
The most consequential line in the plan is not the headline figure but a single instruction buried in the deployment section: the plan calls for greater efforts to adapt infrastructure to home-grown computing chips. That phrasing reframes the entire buildout. This is a compute roadmap sized for hardware made in Zhengzhou, Shenzhen and Shanghai — not for anything shipping out of Santa Clara.
The Numbers Behind the Target
The 9,800 eflops goal extends an expansion already running hot. At a State Council Information Office briefing on July 20, Xie Cun, director of MIIT's Information and Communications Management Bureau, said China's intelligent computing capacity had reached 2,185 eflops at FP16 precision by the end of June 2026 — up 177 percent year over year — with a nationwide rack utilization rate of 71.4 percent. China had by then built 52 intelligent computing facilities each equipped with more than 10,000 accelerator cards. By the end of July, the National Data Administration put capacity at roughly 2,450 eflops.
Hitting 9,800 eflops therefore means more than quadrupling the June level in four and a half years. The plan directs "orderly deployment" of intelligent computing clusters at the 10,000-card scale and at 100,000 accelerator cards or more, with inference facilities provisioned separately for specific applications. Advanced storage capacity is to rise from 540 exabytes in 2025 to 1,700 exabytes by 2030. Newly built large and hyperscale facilities must run at a power usage effectiveness below 1.2, down from 1.25 in 2025 — a meaningful constraint when the target implies gigawatts of new load.
The broader industry targets are equally aggressive. Sector revenue is to reach 4.1 trillion yuan by 2030, up from 3.8 trillion in 2025. On the network side: 50 5G and 5G-Advanced base stations per 10,000 people, 95 percent 5G user penetration, more than 320 million gigabit broadband subscribers, and commercial 6G services "at an appropriate time." The plan also orders exploratory research into space-based computing.
Chinese analysts have framed the plan less as a spending program than as an architectural shift. He Baohong, chief engineer at the China Academy of Information and Communications Technology, told China Daily: "China's computing strategy is shifting from building individual computing centers to weaving resources into a nationally connected computing network. Such a system would use hub, regional and edge nodes, communications networks, power networks and unified dispatching platforms to make computing resources more accessible and efficient."
Wang Zhiqin, CAICT's vice-president, put it in competitive terms, saying the 2026-30 window is when China's information and communications industry moves from scale leadership to system leadership — and that the sector must coordinate networks, computing power, data and AI capabilities to support breakthroughs in future industries.
Why It Matters
This is compute nationalism written into statute. The plan does not merely fund data centers; it specifies that the national computing fabric should be adaptable to domestic silicon, which converts an export-control workaround into a procurement mandate for Huawei, Cambricon and their peers. For Chinese chipmakers, a state five-year plan naming their category as the substrate for a $532 billion buildout is about as strong a demand signal as industrial policy can generate.
The timing sharpens the point. The document went up on a Sunday. On Monday, the NSA, FBI and CISA issued a joint advisory accusing Chinese AI firms — DeepSeek and Moonshot AI among them — of industrial-scale distillation of American frontier models. Beijing offered no direct rebuttal. It did not have to: the infrastructure plan published the day before is a more durable argument than a denial, because it describes a compute base large enough that access to American models and American clouds matters progressively less.
The skepticism is warranted, though, and it is mostly methodological. MIIT does not specify how "intelligent computing capacity" will be measured against the 9,800 target. Xie Cun's 2,185 eflops figure is quoted at FP16, a precision that inflates headline throughput relative to the FP64 numbers that anchor traditional supercomputing benchmarks — and China's compute accounting has drawn scrutiny before, including an April 2026 claim of 1,882 exaflops in a quantum circuit simulation that independent analysts said was not comparable to standard floating-point measures. Aggregate national eflops is also a poor proxy for the thing that actually matters: contiguous, high-bandwidth clusters capable of training a frontier model. Fifty-two facilities with 10,000-plus cards is real capacity, but it is distributed capacity, and interconnect is where domestic accelerators still trail Nvidia's rack-scale systems most sharply.
The gap that remains, then, is less about raw FLOPs than about density, memory bandwidth and the software stack. China is closing the first faster than Washington's controls anticipated. The other two are harder.
What to Watch
Three markers will show whether the plan is tracking. First, whether the Zhengzhou node — where state broadcaster CGTN reported in August that China activated its first 100,000-chip AI cluster built entirely from domestic silicon — produces a frontier-scale training run rather than an inference farm. Second, the mid-term evaluation MIIT has written into the implementation section, which will be the first official checkpoint against the 9,800 figure. Third, whether the ministry ever publishes a measurement methodology. Until it does, the target is a political commitment rendered in scientific notation — and the distance between Chinese planning numbers and delivered capability has historically been a feature of five-year plans, not an exception to them.
“China's computing strategy is shifting from building individual computing centers to weaving resources into a nationally connected computing network.”— He Baohong, Chief Engineer, China Academy of Information and Communications Technology