China’s Full-Throttle Campaign to Match American AI Chips
China is treating its shortage of advanced artificial intelligence chips as a national emergency. At the center of the response is Vice Premier Ding Xuexiang, one of President Xi Jinping’s closest confidants, who is directing a coordinated, high-pressure effort to build a domestic alternative to Nvidia’s dominant processors. The campaign draws explicit inspiration from the all-out mobilization China used in the 1960s to develop atomic bombs, hydrogen bombs, and satellites after the rupture with the Soviet Union. This time the target is semiconductor self-reliance in the age of AI.
Last summer, Huawei Technologies held a closed-door briefing for Ding on its newest AI chips. Company executives presented a roadmap that, they claimed, could deliver self-sufficiency in some critical areas of AI computing within three years. The message aligned perfectly with Beijing’s priorities. Ding, already overseeing the broader science and technology portfolio, intensified the drive. According to people familiar with the discussions, he later delivered a blunt warning to China’s largest AI firms: anyone who continued to resist using domestic chips would be regarded as a traitor.
The urgency stems from U.S. export controls that began tightening in 2022. Those restrictions blocked sales of Nvidia’s highest-performing GPUs and limited China’s access to the extreme ultraviolet lithography machines and other advanced manufacturing tools required to produce cutting-edge chips. The policy was designed to slow China’s progress in both commercial and military AI applications. Chinese companies initially responded by stockpiling chips, purchasing downgraded versions such as the Nvidia H20, and exploring gray-market channels. Over time, the government shifted strategy from passive adaptation to active substitution.
By 2025, China’s dependence on foreign AI chips had fallen from roughly 90 percent in 2021 to under 60 percent, according to industry estimates cited in recent reporting. Huawei’s Ascend series has become the primary domestic option. Manufactured mainly at Semiconductor Manufacturing International Corporation (SMIC), these chips rely on multi-patterning techniques that stretch the capabilities of deep ultraviolet lithography equipment, since China has no access to EUV machines. Huawei and its partners have also invested heavily in advanced packaging, circuit stacking, and software optimizations that extract more performance from each die. Company leaders have openly credited the U.S. restrictions with forcing innovations they might not otherwise have pursued at the same speed.
Production is scaling, though still constrained. Huawei has targeted hundreds of thousands of Ascend units for 2026, with some internal projections aiming higher as SMIC expands capacity at the 7-nanometer node and experiments with more advanced processes. Parallel efforts are underway in high-bandwidth memory, a critical companion technology. ChangXin Memory Technologies is increasing output, yet volumes remain well below those of global leaders such as Samsung and SK Hynix. The result is a persistent bottleneck: even when logic dies are available, packaging them into complete AI accelerators is limited by memory supply.
Performance gaps remain substantial. Leading Nvidia processors still deliver several times the computing power of Huawei’s best current offerings on demanding training workloads. Chinese firms compensate by deploying large clusters of domestic chips, refining algorithms for greater efficiency, and focusing more resources on inference rather than the most compute-intensive training runs. Open-source model developers have shown that clever engineering can narrow the practical difference in certain applications, but the raw hardware disparity continues to shape what is possible at the frontier.
Government pressure has been both carrot and stick. State funds have injected tens of billions of dollars into the semiconductor sector. Salary caps that once discouraged top talent have been relaxed. Companies are encouraged to list on domestic exchanges to raise capital and impose market discipline. At the same time, access to scarce domestic chips is tightly managed. Firms seeking limited approvals to buy remaining Nvidia products face requirements to increase their use of local silicon. Reports of a booming gray market, with prices for smuggled advanced chips rising sharply, illustrate the intensity of unmet demand.
The technical challenges are formidable. Chipmaking is far more complex and iterative than the nuclear and space programs of earlier decades. Yield rates on advanced processes lag those of Taiwan Semiconductor Manufacturing Company. Specialized materials, electronic design automation software, and certain process chemicals remain areas of vulnerability. Analysts note that even optimistic Chinese roadmaps project a continued performance lag through the end of the decade. Closing the gap completely could take far longer.
Yet the political commitment shows little sign of wavering. The push aligns with broader goals outlined in China’s economic planning documents, which elevate technological self-reliance as a core national priority. Huawei’s role extends beyond design; the company coordinates a wide network of suppliers, research institutes, and manufacturers, functioning as a central organizer of the effort. Other players, including Alibaba’s chip teams and various specialized startups, are also contributing under the same policy umbrella.
The stakes extend well beyond commerce. Advanced AI chips determine who can train the largest models, deploy them at scale, and apply them to areas ranging from scientific research and autonomous systems to military applications. China’s progress in AI software and open models has been notable, creating an imbalance between algorithmic capability and hardware supply. Beijing’s campaign seeks to correct that imbalance before the disparity becomes a permanent strategic liability.
Whether the current surge in domestic production and engineering workarounds will prove sufficient remains an open question. Shortages of compute capacity have already forced some Chinese AI labs to limit access to new models. At the same time, the intensity of the mobilization, the willingness to accept higher costs and lower yields, and the political framing of the issue as a matter of national loyalty all suggest that China intends to keep pressing. The United States retains a clear lead in the most advanced silicon. China is determined not to let that lead translate into lasting technological subordination.
The outcome will influence the trajectory of the global AI race for years. For Chinese policymakers, the alternative—continued dependence on American chips subject to sudden restriction—is no longer acceptable. The campaign now underway is their answer.
About The Author
