China’s Ministry of Industry and Information Technology published its 15th Five-Year Plan for the information and communications industry on September 7, 2026. The plan covers 2026–2030 and sets a target of 9,800 exaflops of intelligent computing capacity — up from 1,590 exaflops in 2025, a 6.2-fold increase. It commits 3.8 trillion yuan (US$532 billion) in cumulative information infrastructure investment and explicitly calls for “greater efforts to adapt infrastructure to home-grown computing chips.” The plan calls for “orderly deployment” of intelligent computing clusters at the 10,000-GPU and 100,000-GPU scales. The document was designed around the US chip export control regime, and says so. The structural question the plan raises is not whether China will hit the 9,800 exaflop target. It is what a state-directed, half-trillion-dollar five-year commitment at this scale does to the AI technology competition when the United States is not running a comparable effort.
1. The Document
China’s Ministry of Industry and Information Technology is the primary regulatory authority for the country’s information and communications technology sector. Its 15th Five-Year Plan — published September 7, 2026, with an internal dating of August 12, 2026 — covers 2026–2030 and supersedes the 14th Five-Year Plan (2021–2025). [Established — Unite.AI, “MIIT Plan Targets 9,800 Eflops of Intelligent Compute by 2030,” 2026; South China Morning Post, “China targets fourfold boost in AI computing capacity by 2030 in major tech push,” 8 September 2026.]
The headline figure is 9,800 exaflops — a measurement in exaFLOPS (floating-point operations per second at exascale). The 2025 baseline is stated as 1,590 exaflops. The computing target therefore represents a 6.16-fold increase in five years. [Established — MIIT 15th Five-Year Plan, as reported by Unite.AI and SCMP.] The financial commitment is 3.8 trillion yuan ($532 billion at current exchange rates) in cumulative information infrastructure investment through 2030 — covering data centres, network infrastructure, computing clusters, and the domestic chip development programme. [Established — Eastern Herald, “China Maps $532 Billion AI Plan to Build Around US Chip Restrictions,” 8 September 2026.]
2. What the Plan Is Actually Doing
The most analytically significant element of the MIIT plan is not the scale of the computing target. It is the explicit acknowledgment that the plan is designed around US chip restrictions.
The US export controls introduced in October 2022 and subsequently tightened in 2024 and 2025 restrict China’s access to Nvidia’s most capable AI training processors and to ASML’s advanced lithography equipment. The controls were designed on a specific theory: that limiting China’s access to high-end semiconductors would constrain its AI development trajectory in proportion to the chip capability gap. [Assessed with established confidence — US Commerce Department export control filings, 2022–2025; standard analysis of the chip restriction regime as documented by the Semiconductor Industry Association and academic literature.]
The MIIT plan’s call for “greater efforts to adapt infrastructure to home-grown computing chips” and its emphasis on 10,000-GPU and 100,000-GPU cluster infrastructure are a direct architectural response to the restriction. [Established — Eastern Herald, 8 September 2026.] Chinese domestic chip producers — including Huawei’s Ascend series and Cambricon, which this week joined the PyTorch Foundation as a Platinum member alongside Alibaba Cloud — have been developing processors that, individually less capable than Nvidia’s restricted chips, can be deployed in large arrays to achieve aggregate computing capacity at target scale. [Established — PyTorch Foundation member announcement, September 8, 2026; industry reporting on Ascend and Cambricon trajectories.]
The plan’s 9,800 exaflop target is framed as aggregate infrastructure capacity, not individual chip performance. This framing is the strategy. If the export controls prevent China from obtaining the chips that make individual nodes highly efficient, the plan commits to deploying more nodes. [Assessed with moderate confidence — analytical interpretation of the plan’s stated measurement methodology and domestic chip deployment framing.]
3. The EUV Factor — and the Widening Gap
The same week the MIIT plan published, ASML and TSMC announced a joint industry initiative to move High NA EUV lithography from the decades-old 6-inch photomask format to a 12-inch standard, with Intel Foundry and Samsung Electronics signing on. [Established — semiconductor industry reporting, September 8, 2026.]
ASML’s High NA EUV machines — the most capable chip manufacturing equipment in existence — are already subject to US-imposed controls that prevent their sale to China. The 12-inch photomask initiative extends the frontier of what those machines can produce, in a direction that further widens the manufacturing gap between what China’s domestic fabs can build and what TSMC and Intel Foundry will be producing by 2028–2030.
The MIIT plan’s strategic response to this trajectory is not to close the chip manufacturing gap. It is to declare that chip manufacturing parity at the per-node level is not required to achieve the aggregate computing capacity target. Whether this declaration is achievable depends on domestic chip performance assumptions that have not been independently verified at the proposed cluster scales. [Assessed with moderate confidence — the plan’s own architecture separates the capacity goal from per-chip performance metrics; independent verification of Chinese domestic chip aggregate performance at 10,000+ GPU cluster scale is limited in the open literature.]
4. The Race’s New Shape
The MIIT plan raises a structural question the technology competition has not previously required answering in such explicit terms: what does the AI race look like when one superpower is running a state-directed five-year plan with $532 billion committed, and the other is not?
The United States has not published an AI computing capacity plan at national scale. The CHIPS and Science Act (2022) committed approximately $52 billion to semiconductor manufacturing and research — a one-time appropriation, not a rolling five-year infrastructure commitment. US private sector investment in AI computing is substantial: Meta, Google, Microsoft, and OpenAI have collectively announced several hundred billion in planned capital expenditure across the coming years. But these are corporate capex forecasts subject to shareholder pressure and earnings cycle constraints, not state commitments with five-year planning horizons and enforcement mechanisms. [Assessed with moderate confidence — public reporting on US AI infrastructure investment; CHIPS Act funding analysis from Congressional Research Service.]
China’s plan is not guaranteed to succeed. State-directed technology plans have a mixed empirical record: Made in China 2025 achieved some targets, significantly missed others, and produced unintended market distortions. The 9,800 exaflop target depends on domestic chip performance assumptions at cluster scale that are not independently verified. The $532 billion is a plan commitment, not a locked appropriation. [Assessed with moderate confidence — standard analysis of Chinese industrial planning history and execution gaps.]
None of this reduces the plan’s structural significance. Beijing has stated, in a formal policy document, that it intends to achieve AI computing sovereignty through state-directed investment by 2030 and is committing half a trillion dollars to the attempt. The statement itself changes what the US private sector and government must assume about the 2030 AI computing environment when pricing risk, allocating capital, and designing policy. A plan that fails at 60% of its target still produces 5,880 exaflops — a figure far in excess of any current public estimate for US national AI computing capacity. [Assessed with moderate-high confidence — mathematical implication of stated targets; no public US national AI computing capacity aggregate exists for comparison.]
Prediction: Within 18 months of the MIIT plan’s September 2026 publication (by March 2028), the US government will publish at least one formal policy assessment explicitly responding to the 9,800 exaflop target — most likely through the Commerce Department, the National AI Initiative, or the National Security Council; at least one Congressional bill will be introduced citing the MIIT plan as rationale for expanded national AI computing investment; and at least three major US technology companies’ annual SEC filings will cite the MIIT plan as evidence of intensified competitive risk in AI infrastructure.
Confidence: Moderate. The plan’s public provenance, specific targets, and dollar commitment make it precisely the kind of document that generates formal government response. The US pattern following Made in China 2025 and the 2022 chip controls suggests that formal policy responses follow strategic provocations with an 18-24 month lag.
Resolution: March 2028. Check: Commerce Department, NSC, or National AI Initiative publications; Congressional bills introduced citing MIIT 15th Five-Year Plan; SEC filings from Alphabet, Meta, Microsoft, Amazon.
Bottom line: The US chip export control regime was designed to constrain China’s AI trajectory by restricting access to the hardware that makes that trajectory possible. China has now published a five-year plan whose stated purpose is to route around that constraint through aggregate infrastructure investment and domestic chip development. The plan is the response to the policy. Whether the plan succeeds is a question for 2030 analysts. The question for September 2026 is what the US policy response to the plan will be — and whether the current architecture of chip export controls, private-sector capex, and a one-time CHIPS Act appropriation is the adequate answer to a state-directed, half-trillion-dollar five-year commitment. That answer has not been written.