Here’s the uncomfortable irony buried in a July 2026 AEI report: American export restrictions on AI chips, designed to limit China’s compute capabilities, may have pushed Huawei to move faster than it otherwise would have. According to Huawei Central, a new analysis suggests Huawei could meet between a third and half of China’s total AI chip demand by 2028, and potentially exceed 50 percent under favorable conditions.
The figures come from an American Enterprise Institute report titled “China’s Path to Chip Independence.” The AEI’s modeling is notably measured in its language, which makes the conclusions land harder. Even in the most pessimistic scenario, the report puts Huawei’s share of China’s compute demand at roughly one-third by 2028. If memory and packaging production scales quickly, and if Huawei redirects more of its manufacturing capacity away from mobile chips, that number climbs past 50 percent. Full self-sufficiency, the report predicts, could arrive by 2030.
The chip in question is Huawei’s Ascend line of AI accelerators. Huawei is expected to release the Ascend 950DT, positioning it as a direct alternative to Nvidia’s top-tier AI processors. Nvidia currently dominates the global AI accelerator market with its H100 and B200 series, and has been the de facto choice for Chinese cloud providers and AI labs up until export restrictions effectively cut off supply. Huawei’s Ascend 910B was the first credible domestic substitute, though it still trails Nvidia on raw performance. The 950DT is meant to close that gap further.
To understand why this matters, consider the scale of China’s AI infrastructure buildout. Major cloud providers like Alibaba Cloud, Baidu, and Tencent are running large language model training and inference at a pace that demands enormous chip volumes. If Huawei can reliably supply half that demand, it changes the economics and politics of AI development in China significantly. It also reduces the leverage that US export policy has over China’s AI trajectory.
There are real constraints worth watching, though. Huawei’s Ascend chips rely on SMIC’s most advanced nodes, and SMIC still lags behind TSMC and Samsung in yield rates and process maturity. Packaging and high-bandwidth memory remain bottlenecks. The 50 percent figure is the optimistic ceiling, not a baseline forecast. Still, even the floor estimate of 33 percent would have seemed far-fetched three years ago.
For developers and AI infrastructure buyers evaluating supply chain risk, the picture is shifting. China is no longer purely dependent on Nvidia, and that has implications beyond the country’s borders. If Huawei’s Ascend ecosystem matures enough to support serious model training workloads, it becomes a genuine alternative platform, one with its own software stack, compiler toolchain, and growing developer adoption inside China.
The broader trend here is real: export controls as a policy tool have a mixed track record when the target country has both the capital and the industrial base to respond. China clearly has both. Whether Huawei hits 33 percent or 50 percent by 2028, the direction of travel is no longer in doubt.
