On March 12, China’s lawmakers approved the country’s 15th Five-Year Plan, a sweeping blueprint for 2026 to 2030. Washington, meanwhile, is still arguing about whether Nvidia should be allowed to sell last-generation chips to Beijing. The mismatch tells you everything about who is playing offense and who is reacting.
The plan mentions artificial intelligence more than 50 times. Semiconductors? Barely a handful. That ratio is deliberate. Beijing has decided that chips are an input, not the destination, and it has organized its entire national strategy around what comes after the silicon: AI systems, quantum computing, fusion energy, brain-computer interfaces, 6G, and embodied robotics.
None of this means China has stopped caring about chips. It absolutely has not. But the framing has shifted. Semiconductors now sit inside a larger story about “technological self-reliance,” alongside biotech, advanced materials, and foundational software. The plan calls for “rapid breakthroughs in critical areas” across all of these fields simultaneously. The goal is not to win the chip war. The goal is to make the chip war irrelevant.
What the Plan Actually Says
The Five-Year Plan sets a target for core digital industries to reach 12.5% of GDP by 2030. It establishes an “AI+” action plan as a cross-cutting national program spanning science, industry, governance, and global cooperation. Computing power, which was nearly absent from the previous plan, now gets its own dedicated chapter.
The plan also commits to establishing a global organization for AI cooperation and building AI collaboration platforms under the Belt and Road Initiative. Beijing wants to set the rules of the AI era, not just participate in it.
On the semiconductor side, the plan takes a pragmatic approach. It calls for refining mature-node production, advancing high-performance processors and high-density memory, and accelerating wide-bandgap semiconductorsA semiconductor material that operates at higher voltages and temperatures than silicon. Silicon carbide (SiC) and gallium nitride (GaN) are common examples, used in power electronics and high-frequency devices. like silicon carbide and gallium nitride. It is a plan to build a complete, self-sustaining chip ecosystem, not to chase TSMC’s cutting edge at any cost.
The Chip War That Washington Is Still Fighting
Since October 2022, the US has progressively tightened export controls on semiconductor technology to China. The policy aimed to impair Chinese capabilities in AI and supercomputing by cutting off access to high-end chips, blocking advanced design tools, and restricting chipmaking equipment.
The controls worked, to a point. Export controls have made China a marginal producer of AI chips, and without EUV lithographyA chip manufacturing technique using extreme ultraviolet light to etch transistors below 7nm. Only one company (ASML) makes the machines, making them a key export control target. machines from the Netherlands’ ASML, Chinese foundries cannot manufacture at the most advanced nodes.
But the controls also did something the architects did not plan for: they lit a fire under China’s domestic chip industry. Beijing launched an all-out, government-backed push for self-sufficiency. SMIC, China’s largest chipmaker, posted record revenue of $9.3 billion in 2025, with $8.1 billion in capital expenditure and aggressive 12-inch wafer capacity expansion. The state-backed Big Fund III injected $47.5 billion into the semiconductor supply chain, targeting not just fabs but the weak links: lithography tools, inspection systems, etching platforms, EDA softwareElectronic Design Automation: software tools used to design, simulate, and verify integrated circuits before fabrication. A critical chokepoint in semiconductor supply chains and a target of export controls., photoresists, and specialty materials.
And then Washington reversed course. In December 2025, the Trump administration announced it would allow Nvidia to sell H200 chips to China, with a 25% surcharge and volume caps. By early 2026, the formal rules were in place, with Chinese companies gaining access to H200 chips.
The zigzag in US policy, as the Council on Foreign Relations noted, reduces the leverage of future controls. China has learned that Washington will back down under industry pressure. The lesson is clear: self-reliance is the only reliable path.
Why the Controls Are Not Enough
The fundamental problem with the US approach is that export controls are a holding action, not a strategy. As CSIS researchers put it, they are “at best a short-term palliative for the long-term strategic challenge posed by China.”
The evidence supports this. Chinese labs have produced AI models that rival American ones despite chip restrictions. DeepSeek’s R1 model, trained on downgraded Nvidia H800 GPUs at a reported cost of $5.6 million, roughly matches capabilities of models from OpenAI, Google, and Anthropic. DeepSeek’s founder Liang Wenfeng has said that chip access is his biggest constraint, but the constraint has pushed his team to innovate in efficiency rather than brute-force compute.
Meanwhile, circumvention is rampant. Huawei reportedly used shell companies to obtain 2 million chipletsA small, modular semiconductor die designed to be combined with other chiplets in a single package. This approach allows mixing components from different manufacturers or manufacturing processes. from TSMC. A Singapore-based ring bought $390 million worth of servers containing banned Nvidia GPUs and smuggled them into Malaysia. Semiconductor chips are tiny and produced by the millions. Unlike Cold War-era technologies like aircraft engines, they are essentially impossible to fully control.
The Research Pipeline
Perhaps the most underappreciated dimension is research. China is not just buying or smuggling chips. It is investing heavily in inventing the next generation. According to Georgetown University’s Emerging Technology Observatory, Chinese scholars published 160,852 chip-related papers from 2018 to 2023, more than the next three countries combined. The US came in second with 71,688.
Chinese institutions claimed nine of the top ten spots in chip research output and eight of the top ten in highly cited publications. This is not just volume. It is influence.
The CSIS report warns that export controls “are relevant only so long as the United States and its allies possess chip technologies that China wants and needs.” If Chinese researchers develop alternative approaches, whether through new materials, novel architectures like RISC-V, or carbon nanotube transistors, the entire premise of the control regime collapses.
What Washington Is Missing
The 15th Five-Year Plan is not a semiconductor plan. It is an AI-era industrial strategy that treats semiconductors as one component among many. While Washington debates Nvidia export licenses, Beijing is building a unified national computing power network, setting global AI governance standards, and positioning itself as the default technology partner for the Global South through BRI-linked AI platforms.
The plan also reveals a shift in how China thinks about competition. It is no longer just about catching up. The emphasis on “new quality productive forces” signals that Beijing intends to leapfrog, not follow. Quantum technology, biomanufacturing, hydrogen and fusion energy, embodied intelligence: these are the six designated growth engines for the next decade.
The US, for its part, has the CHIPS Act and the newly formed National Center for the Advancement of Semiconductor Technology. These are meaningful investments. But they are investments in the current battlefield, not the next one. As chip export controls cool down amid diplomatic priorities, the US risks ceding the strategic initiative to a competitor that is already playing a different game.
The chip war is not over. But it may already be beside the point.
On March 12, China’s National People’s Congress ratified the 15th Five-Year Plan, the country’s economic and industrial blueprint for 2026 to 2030. For anyone still framing the US-China technology rivalry primarily through semiconductors, the document is a corrective. Artificial intelligence appears more than 50 times. Integrated circuits are mentioned roughly four times. The ratio is not accidental. It reflects a strategic reorientation that Washington has been slow to recognize.
The Architecture of the Plan
The CCP’s proposal, translated in full by Georgetown’s Center for Security and Emerging Technology, lays out a layered technology strategy. At the foundation: “new quality productive forces,” a framework for replacing legacy industries with technologically advanced successors. Above that: a commitment to “significantly improve scientific and technological self-reliance” through basic research, original innovation, and key core technologies.
The plan identifies six “future industries” as designated growth engines: quantum technology, biomanufacturing, hydrogen and fusion energy, brain-computer interfaces, embodied intelligence, and 6G communications. Semiconductors are not on this list. They are treated as enabling infrastructure, not as the strategic objective itself.
Chapter 13 establishes the “AI+” action plan as a cross-cutting national program modeled loosely on Germany’s Industry 4.0 but far broader in scope, covering science and technology, industrial development, consumer markets, social welfare, governance, and global cooperation. Computing power, nearly absent from the 14th Five-Year Plan, now receives a dedicated chapter focused on unified national computing power networks and industrial digital ecosystems.
The headline target: core digital industries at 12.5% of GDP by 2030.
What the Semiconductor Section Actually Specifies
The semiconductor provisions in the plan are pragmatic rather than aspirational. The draft calls for efforts to “refine and strengthen mature-node, enhance capabilities in advanced process technologies, accelerate development of key equipment, materials and components, and advance high-performance processors and high-density memory.” It singles out wide-bandgap semiconductorsA semiconductor material that operates at higher voltages and temperatures than silicon. Silicon carbide (SiC) and gallium nitride (GaN) are common examples, used in power electronics and high-frequency devices., specifically silicon carbide (SiC) and gallium nitride (GaN), for industrial development alongside high-performance AI chips.
This is a significant strategic signal. Rather than an all-out push for sub-5nm logic at any cost, Beijing is prioritizing a vertically integrated domestic supply chain. The Big Fund III, established in May 2024 with $47.5 billion in registered capital, reflects this approach. Its capital is flowing into the weak links of the supply chain: lithography tools, inspection systems, etching platforms, EDA softwareElectronic Design Automation: software tools used to design, simulate, and verify integrated circuits before fabrication. A critical chokepoint in semiconductor supply chains and a target of export controls., photoresists, specialty gases, wafer materials, and power modules.
The strategy at the fab level is similarly pragmatic. SMIC posted record revenue of $9.327 billion in 2025, driven by aggressive capacity expansion. The company added roughly 50,000 12-inch wafers per month of capacity in 2025, with another 40,000 planned for 2026. Capital expenditure hit $8.1 billion, with 2026 spending expected to remain flat. SMIC’s 7nm-equivalent node is in mass production, though yields reportedly weigh on margins (gross margin slid to 19.2% in Q4 2025). China-related revenue accounted for nearly 90% of total sales.
ChangXin Memory Technologies (CXMT) is expanding into DRAM. In five years, China’s share of the global memory chip market went from virtually zero to 5%, with projections for doubling. Alibaba’s RISC-V-based C930 CPU offers an alternative architecture to proprietary Arm and x86 designs, potentially circumventing export-control bottlenecks entirely.
The Export Control Regime: Technical Assessment
The US export control regime, initiated in October 2022 and tightened in October 2023 and December 2024, targets three vectors: AI chip exports, chipmaking equipment, and semiconductor design tools. Chris Miller’s comprehensive assessment for AI Frontiers offers a useful framework for evaluating their impact across three dimensions.
Chipmaking capability: Controls have been most effective here. By blocking EUV lithographyA chip manufacturing technique using extreme ultraviolet light to etch transistors below 7nm. Only one company (ASML) makes the machines, making them a key export control target. tools from ASML and restricting DUV immersion tools from Japan’s Tokyo Electron and others, the US and allies have kept SMIC locked to multi-patterning DUV at 7nm-equivalent. Without EUV, achieving high yields below 7nm is economically impractical. Export controls have made China a marginal producer of AI chips, with Huawei projected to produce only 200,000 AI chips in 2025 versus the millions Nvidia ships globally.
AI model quality: Controls have been largely ineffective. Chinese labs have produced models that are, at worst, fast followers in capability benchmarks. DeepSeek trained its R1 model on 2,000 H800 GPUs (a China-specific downgrade of the H100 with halved interconnect bandwidth) for approximately $5.6 million. The constraint on compute pushed algorithmic innovations in training efficiency, not capability.
AI infrastructure deployment: Controls have been moderately effective internationally. Huawei lacks the scale to compete with Nvidia-based cloud infrastructure outside China. But the H200 policy reversal may change this. The Trump administration formalized the H200 export policy in early 2026, approving H200 exports with a 25% surcharge and volume caps. CFR analysts warn the policy could at least triple China’s aggregate AI computing power additions in 2026 and enable a potential “AI Belt and Road” infrastructure push.
Circumvention and Its Implications
The enforcement regime has structural weaknesses that are well-documented. Semiconductor chips are produced by the millions, are physically small, and are easily incorporated into end-use equipment. Huawei used shell companies to procure over 2 million chipletsA small, modular semiconductor die designed to be combined with other chiplets in a single package. This approach allows mixing components from different manufacturers or manufacturing processes. from TSMC for its Ascend 910 AI processors before the fraud was detected. A separate ring purchased $390 million in servers containing banned Nvidia GPUs and smuggled them to Malaysia. CSIS notes that the Cold War analogy does not hold: the technologies at issue are small, evolve rapidly, and flow through deeply intertwined global supply chains that did not exist during the Soviet era.
The Research Dimension
The most consequential long-term development may be in research output. Georgetown’s Emerging Technology Observatory found that Chinese scholars published 160,852 chip-related papers from 2018 to 2023, more than the next three countries combined. The US was a distant second with 71,688. Chinese institutions held nine of the top ten spots in total output and eight of the top ten in highly cited publications.
This research pipeline is not theoretical. Peking University researchers announced a 2D transistor operating 40% faster than TSMC’s 3nm devices while consuming 10% less energy. A separate team disclosed the world’s first carbon nanotube-based chip running AI tasks using ternary logic, a potential leap beyond binary silicon. These are early-stage results, but they illustrate CSIS’s warning: export controls “are relevant only so long as the United States and its allies possess chip technologies that China wants and needs.”
The Strategic Gap
The core mismatch is temporal. US policy remains focused on restricting inputs to China’s current technology stack: sub-7nm logic chips, HBM, EUV tools. China’s 15th Five-Year Plan is organized around outputs for the next technology stack: AI systems, quantum networks, fusion energy, embodied robotics, and global digital governance.
The US has countervailing strengths. The CHIPS Act, the National Center for the Advancement of Semiconductor Technology, and the existing dominance of Nvidia, TSMC (with its Arizona fabs), and ASML are real advantages. But these are investments in sustaining the current paradigm. The 15th Five-Year Plan is a bet on disrupting it.
China’s plan also includes an explicit international dimension: establishing a global organization for AI cooperation, setting AI governance standards, and deploying technology platforms across the Global South. As US chip export controls cool down amid diplomatic priorities and industry lobbying, the risk is not that China catches up in the chip war. It is that Beijing defines the next battlefield while Washington is still fortifying the last one.
