Nvidia's CEO Jensen Huang recently confirmed at a shareholder meeting that his company has yet to generate any revenue from China's AI chip market and is uncertain whether Beijing will allow imports of its products at all. That is a striking reversal. Just a few years ago, Nvidia held roughly 95% of China's market for the specialised computer chips used to build and run AI systems.
The story behind that collapse is straightforward. Starting in 2022, the U.S. government restricted exports of high-end AI chips to China, citing national security concerns. The restrictions tightened steadily, eventually covering even the stripped-down versions Nvidia had designed specifically to comply with the rules. Nvidia took a $4.5 billion write-off in early 2025 on chips it could no longer sell. By the time the Trump administration signalled it might allow some chip sales back into China, Beijing had already made its choice: buy domestic.
Huawei is the main beneficiary. Its Ascend chip line, which barely registered as competition a few years ago, now leads China's domestic market. Bernstein, the research firm, estimated Huawei held about 40% of China's AI chip market in 2025, roughly matching Nvidia. This year, Bernstein projects Huawei's share will grow to around 50%, while Nvidia's falls to about 8%. In revenue terms, Huawei expects its AI chips to bring in around $12 billion in 2026, up from $7.5 billion last year.
Huawei's individual chips are still not as powerful as Nvidia's most advanced products on a chip-by-chip comparison. The current flagship Ascend 910C delivers roughly 60% of the performance of Nvidia's H100 per chip, and is built on older manufacturing processes because China cannot access the most advanced chip-making equipment, which is also under export restrictions. But Huawei is compensating through scale: it links thousands of chips into enormous computing clusters called SuperPoDs, and claims those systems can match or exceed Nvidia at the total system level for certain tasks.
The more consequential development is what happened with DeepSeek. When China's fastest-growing AI company released its V4 model in April 2026, it was explicitly built for Huawei's hardware. DeepSeek spent months working directly with Huawei engineers to rewrite portions of the model's code so it would run efficiently on Ascend chips. Huawei's entire Ascend product line was ready on launch day. China's three largest tech platforms, Alibaba, ByteDance, and Tencent, promptly placed large orders for the new chips. That combination, a top-tier AI model designed to run on domestic hardware, is the clearest sign yet that China's AI supply chain is becoming genuinely self-contained.
That matters beyond China's borders. Huawei already operates in 170 countries, with deep relationships across Southeast Asia, the Middle East, and Africa built through decades of selling telecommunications equipment. It has already approached buyers in the UAE, Saudi Arabia, and Thailand for its AI chips. As production capacity grows and prices become more competitive, Huawei chips could appear in AI infrastructure across markets that are not covered by U.S. export restrictions.
For businesses outside China, the practical takeaway is about the direction of travel. The world's AI hardware is splitting into two distinct supply chains. One runs through Nvidia, AMD, and American cloud providers. The other runs through Huawei and Chinese technology companies. Which AI tools a business can use, where they can run, and what rules apply to them will increasingly depend on which supply chain those tools are built on.
Nvidia is not in trouble globally. Its revenue this quarter alone is expected to reach around $91 billion, driven by AI investment everywhere outside China. But China is the world's second-largest economy, and losing it entirely as a market for its most advanced products is a real and permanent cost.
The bigger question, for businesses and governments alike, is whether the two supply chains will remain separate or eventually converge. Right now, the pressure is firmly toward separation.