Skip to content
NewsRegulation

China's MIIT sets $532 billion AI infrastructure target sized for homegrown chips

· by Pondero Newsdesk

The short version

China's Ministry of Industry and Information Technology released its 15th Five-Year Plan on September 7, targeting 9,800 eflops of AI compute by 2030, a fourfold increase built explicitly around domestically produced accelerators rather than US-exported chips.

China's MIIT sets $532 billion AI infrastructure target sized for homegrown chips

China's Ministry of Industry and Information Technology published its 15th Five-Year Plan for the information and communications sector on September 7, 2026. The headline target is 9,800 eflops of intelligent computing capacity by 2030. That figure needs to more than quadruple from the 2,185 eflops the ministry itself recorded as of June 2026, and the plan says it will get there without Nvidia.

What

The plan, which covers 2026 through 2030, sets 3.8 trillion yuan ($532 billion at current exchange rates) in cumulative information infrastructure investment over the period, with private capital expected to provide the bulk of funding, per the South China Morning Post. Storage capacity would grow from 540 exabytes today to 1,700 exabytes by 2030, per Unite.AI's summary of the plan.

Two clauses in the document carry the most strategic weight. First, the MIIT called for "orderly deployment" of intelligent computing clusters equipped with 100,000 or more accelerator cards, a cluster size that currently does not exist anywhere in China at scale. Second, the plan explicitly called for "greater efforts to adapt infrastructure to home-grown computing chips," per SCMP. That language is a direct response to US export controls that have progressively blocked Nvidia's H200, Blackwell, and Rubin lines from Chinese buyers.

China's intelligent computing base has already moved fast. By the end of July 2026, capacity had risen to roughly 2,450 eflops, per the National Data Administration, as domestic chip deployments from companies including Cambricon and Biren accelerated. The June 2026 reading of 2,185 eflops was itself a 177 percent increase from a year earlier.

The plan also sets a 2030 industry revenue target of 4.1 trillion yuan (roughly $605 billion) and calls for annual telecom business volume growth averaging 7 percent, per Unite.AI. On the telecoms side, the MIIT set a 5G adoption target of 95 percent user penetration and 50 base stations per 10,000 people by 2030, alongside a 6G commercial launch timed "at an appropriate time."

Why it matters

The plan matters less for its target number and more for what the cluster-sizing and chip-adaptation language says about how China intends to reach it. A 100,000-card cluster running Chinese accelerators would constitute a frontier training run. That the MIIT describes this as a near-term infrastructure goal, not a distant aspiration, means state-backed cloud operators such as Alibaba Cloud, Huawei Cloud, and Baidu AI Cloud are likely to receive directed support to build these facilities before 2028.

For AI tool operators and enterprise teams making cloud infrastructure decisions today, this trajectory implies faster bifurcation in the global AI supply chain. API pricing, model availability, and compliance posture will increasingly depend on which side of the supply chain a vendor sits on. Teams already evaluating whether to run inference on US-linked clouds or Chinese clouds now have a published government timeline to inform that risk modeling.

The domestic chip clause also affects hardware sourcing. If Chinese AI accelerator vendors hit the cluster-scale targets embedded in this plan, the current gap in raw performance between domestic chips and banned Nvidia products may close faster than export-control architects expected.

What to watch next

The plan sets ambitious targets but not binding ones. The near-term proof point is whether China constructs any 100,000-card clusters using domestic chips by end of 2027. State broadcaster CGTN reported in August 2026 that at least one large-scale domestic deployment was operational, but cluster sizes were not disclosed.

Two milestones will signal whether execution matches the document: the National Data Administration's quarterly capacity updates (which have outpaced the Ministry's own forecasts twice in 2026), and any procurement announcements from Alibaba Cloud or China Telecom specifying chip suppliers for new AI compute hubs.

Sources