The word for lithography machine does not appear once in the draft outline of China’s 15th Five-Year Plan. Neither does wafer fab, extreme ultraviolet, or chip manufacturing. In the 141-page planning document submitted to the National People’s Congress on March 5, the entire vocabulary of the “chip war” as debated in Washington is absent.

What appears instead is a different strategic vocabulary. Artificial intelligence outnumbers references to integrated circuits by roughly 13 to 1. Computing power, nearly absent from the previous plan, now receives its own dedicated chapter. A new planning term has entered the document for the first time in Five-Year Plan history: 模芯云用, meaning “model-chip-cloud-application,” a four-character compound that encodes a formal architectural decision. The chip is one layer of four, alongside AI models, cloud infrastructure, and application deployment. China’s strategic objective is not the chip; it is the system that contains the chip.

The National Data Administration has already published technical standards for a national computing grid, including resource pooling across facilities, unified scheduling and dispatch, and crisis-override provisions guaranteeing minimum computing services “during extreme market fluctuations or supply crises.” The National Development and Reform Commission (NDRC) has published implementation commentary using the same “reservoir” metaphor for strategic computing reserves, and the State Council has issued policy guidance confirming the architecture. The domestic system, in other words, has moved from planning language into implementation specifications. 

Meanwhile, the 70 percent semiconductor self-sufficiency target established under Made in China 2025, a target China missed by roughly 50 points, has been quietly deleted from the 15th Five-Year Plan and replaced with a deployment metric: digital economy value-added at 12.5 percent of GDP by 2030. Beijing is no longer measuring success by how many chips it produces. It is measuring success by how deeply computing infrastructure penetrates the economy.

None of this means China has abandoned chip fabrication. The plan calls for “extraordinary measures” on a list of key core technologies that includes integrated circuits, in language comparable to wartime mobilization in American budgetary discourse. Advanced process manufacturing remains a stated goal. However, the tonal contrast between the semiconductor sections and the computing infrastructure sections is revealing. The lead phrase for integrated circuits translates as “refine and polish mature process nodes”: defensive, incremental. The lead phrase for computing infrastructure translates as “solidify the foundation of digital-intelligent development”: confident, architectural. The hardware emergency is real. It is also structurally subordinate to an architectural objective.

The evidence suggests that this reorientation is already being operationalized beyond China’s borders. 

In May 2025, Malaysia’s deputy communications minister announced the country as the “first nation to activate Huawei Ascend GPU-powered AI servers at national scale.” The announcement described a $160 million deployment including 3,000 GPUs, a full-stack sovereign AI infrastructure built by a Malaysian company called Skyvast, and a Malaysia-China Trusted Data Zone linking Cyberjaya to Shanghai. 

Within 24 hours of a public statement by the White House AI adviser, the Malaysian government retracted the announcement. The Ministry of Trade, Industry, and Digital Economy stated that the initiative “was not developed, endorsed, or coordinated by the Government of Malaysia.” But the government retraction did not halt the deployment; the private sector operator stated that the project was “moving forward as scheduled.” 

The pattern that emerged, from government endorsement through U.S. pressure to official retraction and private sector continuation, is worth studying as a template. China’s Ministry of Commerce publishes country-by-country investment guides that explicitly map foreign investment screening regimes. Malaysia’s guide states plainly that the country has no security review procedures for foreign direct investment. The 15th Five-Year Plan formalizes what the Malaysia episode previewed: a strategy centered on offshore computing facilities and multilateral AI institutions, including a proposed World AI Cooperation Organization and a Belt and Road AI cooperation platform, explicitly supporting “Global South countries in strengthening AI capacity building.” China is building a name for itself not as a rival to cutting-edge American AI, per se, but as a provider of computing infrastructure to the developing world.

DeepSeek demonstrated in early 2025 that frontier AI capability is achievable with constrained hardware. However, a single model does not constitute the developer ecosystem required for sustained competition. Chinese financial filings corroborate this assessment. A Shanxi Securities equity research report from May 2025 described NVIDIA’s CUDA ecosystem as “the main obstacle” to domestic AI chip substitution. Moore Threads, in its December 2025 IPO prospectus, described NVIDIA’s ecosystem as “not easily surpassable.” These are legally mandated financial disclosures. The software ecosystem gap that the Five-Year Plan leaves conspicuously unmetriced – the only major strategic domain with aspiration and no measurement – is one that Chinese industry acknowledges as a weak spot in its own regulatory filings.

Two additional findings from the Chinese text merit attention from security analysts. The first is the military-civil fusion provision. The chapter on military-civil integration gives the order to “promote efficient fusion of new quality productive forces and new quality combat capability, with bidirectional pull.” The concept that organizes the entire civilian technology strategy, “new quality productive forces,” is explicitly paired with military combat capability.

The computing grid, the AI deployment framework, and the unified scheduling system described throughout the plan are designed to serve military modernization as well as civilian development. Offshore computing facilities built for partner nations under the banner of AI capacity building operate on the same architecture that the plan explicitly links to military capability. The dual-use implications extend well beyond individual items on existing control lists; the entire computing architecture is dual-use by documented mandate. Any assessment of Chinese computing infrastructure projects in partner nations that treats them as purely civilian risks underestimating the architecture’s intended scope.

The second finding concerns the diplomatic framing of the document itself. The United States is not named once. Neither is decoupling, nor any foreign company. The adversary is addressed only through its instruments. The national security chapter contains a three-part counter-strategy formula: counter-sanctions (反制裁), counter-interference (反干预), counter-long-arm jurisdiction (反长臂管辖). These map precisely to the three primary tools of U.S. coercive economic statecraft. Beijing does not name the adversary but prepares specifically for each of the adversary’s methods.

The 15th Five-Year Plan describes a technology strategy that current U.S. export controls address only partially. Controls target the fabrication layer, one component of a four-layer architecture whose other three layers proceed without equivalent countermeasures. Whether the strategic reframe represents genuine evolution or a response to the semiconductor target failure does not change the operational reality. 

The computing infrastructure is being built domestically, deployment has already been attempted in Southeast Asia, the military-civil fusion mandate is explicit, and the software ecosystem vulnerability that China’s own industry identifies as the binding constraint remains the one layer U.S. policy has not addressed. 

The official English translation of the plan will follow weeks after the NPC session closes on March 12. By then, the analytical window for first-mover assessment will have narrowed considerably.