Huawei's Ascend chip series is the clear centerpiece of the national push. Last summer, Huawei held a closed-door briefing for Ding Xuexiang to unveil its latest AI chips and lay out a plan to rival Nvidia and achieve self-sufficiency in critical AI areas within three years .
The results have been dramatic:
The national campaign directly fueled a frenzy of public listings among Chinese GPU startups, all riding investor enthusiasm for homegrown alternatives to Nvidia.
Concentrated IPO window: Between December 5, 2025 and January 8, 2026, four Chinese GPU companies went public in just five weeks — Moore Threads and MetaX on Shanghai's STAR Market, Biren Technology and Iluvatar CoreX in Hong Kong .
The four companies — often dubbed the "GPU Four Dragons" — have collectively raised billions, but the scale of the hype is also a warning: they are burning cash at high rates and remain far from profitable .
Despite real market-share gains inside China, the absolute performance and scale gap with Nvidia remains stark.
Market share by volume: In 2025, Chinese domestic vendors shipped 1.65 million AI accelerator cards — about 41% of China's total ~4 million-unit market — while Nvidia still held ~55% of China's overall AI accelerator server market by shipments . However, in the advanced segment, Nvidia's share fell to roughly 8% per Bernstein
.
Performance gap: Per the Council on Foreign Relations (CFR), the best US AI chips are about 5x more powerful than Huawei's best offerings today. CFR projects that gap will widen to 17x by 2027 as Nvidia's next-generation architectures outpace what Huawei can produce under export controls .
Global market share: The four Chinese GPU IPO companies collectively hold less than 3% of the global GPU market while Nvidia dominates the global AI chip industry .
EUV gap: China built a prototype EUV lithography machine by early 2025, but it has yet to produce operational chips. The government aims for functional chips by 2028, though sources believe 2030 is a more realistic timeline — meaning China remains dependent on older fabrication nodes for years .
Overall AI gap: Brookings notes that Chinese AI models still lag US frontier models "by several months or more" across industry benchmarks from math to code generation .
The bottom line is clear: Ding Xuexiang is leading an aggressive, top-down campaign using procurement mandates, subsidies, and nationalist rhetoric to force China's AI sector off Nvidia and onto domestic chips. Huawei's Ascend line has become the primary beneficiary, and a wave of GPU IPOs has raised billions for startups like Moore Threads and MetaX. But these companies collectively hold negligible global GPU market share, and despite real gains inside China, the absolute performance gap with Nvidia's frontier chips remains large — and is projected to widen significantly under current export controls.