9,800 EFLOPS by 2030: China's ¥3.8 Trillion Compute Plan Turns Ambition Into a Procurement Curve

China’s Ministry of Industry and Information Technology (MIIT) has published a five-year development plan that commits the country to reaching 9,800 EFLOPS of intelligent computing capacity by 2030 — more than a quadrupling of its installed base — backed by 3.8 trillion yuan (roughly US$532–566 billion) in cumulative information infrastructure investment over the 2026–2030 window.

The document, formally the 15th Five-Year Plan for the information and communications industry, was posted publicly on September 7, 2026 through the ministry’s Information and Communications Development Department. It is dated August 12, 2026 internally, and it converts what has long been rhetorical ambition — “compute is the new electricity” — into a hard, measurable procurement curve with named milestones, budget lines, and a mid-term evaluation clause.

The headline numbers

The plan’s target table is unusually specific for a central-government planning document:

• Intelligent computing capacity: 9,800 EFLOPS by 2030, up from a 2025 baseline of 1,590 EFLOPS — a 6.2× increase over the plan window.
• Advanced storage capacity: 1,700 exabytes by 2030, up from 540 EB in 2025.
• Power usage effectiveness (PUE): below 1.2 for new large and hyperscale computing facilities, tightened from 1.25 in 2025.
• Cumulative information infrastructure investment: ¥3.8 trillion across 2026–2030, against ¥3.7 trillion in the prior five-year period.
• ICT industry revenue: ¥4.1 trillion by 2030 (about US$604.8 billion), with 7% average annual growth in total telecommunications business volume.

An EFLOPS is one quintillion (10^18) floating-point operations per second. For context, the fastest single supercomputers on the planet today measure in the low hundreds of PFLOPS — China is planning aggregate national AI compute thousands of times larger, spread across dozens of intelligent computing centres.

Where the capacity actually stands

The current-state figures behind the growth target predate the plan, and they show just how fast the base is already moving. At a State Council Information Office press conference on July 20, 2026, 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% from a year earlier — with an overall rack utilization rate of 71.4% across national computing facilities. The National Data Administration put the figure near 2,450 EFLOPS a month later.

In other words, measured against the June base, the 2030 target is a 4.5× expansion. Measured against the plan’s official 2025 baseline of 1,590 EFLOPS, it is a 6.2× expansion. Either way, the required compound growth rate — roughly 30–45% annually depending on the starting point — is aggressive but not fanceful for a country that just grew the same metric by 177% year-over-year.

10,000-card clusters as the base unit

On deployment, the plan directs the “orderly deployment” of intelligent computing clusters at two scales: the 10,000-accelerator-card tier, and 100,000-card-plus megacentres, with inference computing facilities deployed as needed for specific applications. Fifty-two intelligent computing centres of the 10,000-card tier are already built, according to figures cited with the plan.

The programme extends the “East Data, West Computing” project launched in 2022, which routes workloads from eastern demand centres to western regions with cheaper land and energy across eight national hub nodes and ten data-centre clusters. Xie said MIIT had coordinated more than 70 high-capacity computing corridors linking national hub nodes over the past two years, with a roughly 10% improvement in network performance between hubs.

Two directives in the infrastructure section carry particular weight for the semiconductor side. The document calls for greater efforts to adapt infrastructure to domestically produced computing chips — a quiet but consequential line, given U.S. export controls that continue to restrict top-tier Nvidia hardware — and for exploratory research into space-based computing, hinting at orbital data centres as a long-horizon item rather than a funded near-term build.

Beyond compute: 6G, satellites, and 50G-PON

The compute targets sit inside a broader ICT blueprint with 13 headline indicators and 26 key tasks across six priority areas. The non-AI infrastructure goals are substantial in their own right:

• 50 5G and 5G-Advanced base stations per 10,000 people by 2030, with 95% adoption of 5G services and 500,000 new 5G-Advanced stations.
• 320 million gigabit-and-above broadband subscribers, up from 240 million in 2025.
• One million 50G-PON optical access ports, up from 20,000 in 2025 — a 50× jump that signals next-generation fixed broadband is moving from trial to rollout.
• 3.8 billion mobile IoT terminal connections.
• Commercial 6G services “launched at an appropriate time”, with continued research under the national IMT-2030 promotion group covering network-native AI, integrated sensing and communication, and satellite-terrestrial spectrum sharing.
• Low-orbit satellites exceeding 100 Gbps of individual capacity, direct-to-handset satellite services, and deeper 5G–BeiDou integration.

Implementation, per the document, rests on cross-department coordination, ministry–province linkage, and government–enterprise cooperation, with dynamic monitoring, a mid-term evaluation, and a final review of plan execution.

What it means

An explicit 9,800-EFLOPS target paired with a ¥3.8 trillion five-year budget converts China’s compute ambition from directional rhetoric into a procurement curve that any company with China AI exposure now has to plan against rather than guess at. Three implications stand out.

First, the domestic silicon question is now unavoidable. The plan’s call to adapt infrastructure to domestically produced chips — coming in the same week that Alibaba Cloud and Cambricon joined the PyTorch Foundation as platinum members, and following DeepSeek’s deployment of a 160,000-card Huawei Ascend cluster — signals that the buildout is expected to run increasingly on Chinese accelerators. How much of the 9,800 EFLOPS must ride on domestic hardware, and whether the plan attaches any explicit local-content ratio to the ¥3.8T pool, are the key open questions.

Second, utilization discipline matters as much as capacity. The 71.4% rack utilization figure and the sub-1.2 PUE requirement show MIIT is pairing quantity targets with efficiency targets — a response to criticism that parts of China’s earlier data-centre boom produced underused facilities. The “orderly deployment” phrasing is itself a brake against the speculative buildout that left some western-region centres dark.

Third, the number sets a benchmark for the U.S.–China compute race. If China reaches 9,800 EFLOPS of national AI compute by 2030 on largely domestic chips, it demonstrates a sanctioned supply chain can still deliver frontier-scale aggregate compute. If it falls short, the gap will reveal exactly where export controls bit. Either outcome makes this one of the most consequential infrastructure documents of the current planning cycle.

The plan lands in a busy week for China’s AI infrastructure story — and it is the document most likely to be cited when historians trace where the compute arms race of the late 2020s was formally budgeted.