Nvidia did not lose China because Huawei suddenly became better than Nvidia. Nvidia lost China because the rules of the game changed. When the best chips became hard or impossible to buy, being the top performer stopped mattering as much as simply being available.
That is what a lot of people still do not fully appreciate. Huawei never needed to beat Nvidia on every single test or feature. It just needed to be good enough, right when Nvidia's latest chips got blocked or delayed by export rules. Once that happened, the whole contest shifted. It was no longer only about who had the fastest, most powerful technology. It became a choice between the best chip you could not reliably get and a solid local option that was ready to ship today, backed by the government.
In that new setup, timing and access won out over raw performance.
How Huawei Made It Work
Take Huawei's Ascend 910C chip. They built it by combining two of their older 910B processors into one package. The result delivers around 60% of the performance of Nvidia's H100, particularly for inference tasks that run AI models in production environments.
This was not a fundamental breakthrough that left Nvidia behind. It was a smart, practical engineering fix. Chinese companies needed chips to train and run AI models, and they could not always wait for Nvidia. Huawei gave them something that worked well enough, immediately available.
The narrative that Huawei magically caught up and passed Nvidia is not quite right. The truth is more precise: Huawei became a real choice in a market that was closed off to outsiders. Inside China, being available and locally supported mattered more than being the absolute best.
What Happened to the Market
A few years back, Nvidia effectively owned China's AI chip market. After the export restrictions hit, that changed quickly. By late 2025, Nvidia's share had fallen to approximately 55%, with Chinese manufacturers capturing nearly half the market. Huawei led the domestic pack.
This is not only a numerical shift. It is creating two separate AI worlds. Outside China, Nvidia remains dominant — its software ecosystem (CUDA), deep developer community, and top-tier performance remain unmatched. Inside China, everything is moving local. The chips may not always match the very best global technology yet, but they are sufficient for training models, running data centers, and meeting the country's strategic infrastructure goals.
Why the Export Rules Did Not Work as Planned
Nvidia's own CEO, Jensen Huang, has been direct about this. He called the U.S. export controls a failure. Instead of slowing China down, they accelerated the shift to domestic suppliers.
Chinese firms still needed enormous computing capacity for AI. When they could not reliably access Nvidia chips, they turned to Huawei and others. Demand did not disappear — it redirected to whatever could deliver. That outcome surprised many who assumed the restrictions would freeze China's AI development. Instead, they helped split the market in two and accelerated China's own technology push.
What This Means Going Forward
The significant development here is not that Huawei suddenly became better than Nvidia globally. It is that these policies created a distinct Chinese AI market — one where "good enough, local, and available now" can outperform "the absolute best, but uncertain delivery."
This split is producing two different supply chains, two software stacks, and two frameworks for building AI infrastructure. Over time, it may mean two different definitions of what winning at AI looks like.
- Two separate hardware ecosystems, each with its own manufacturing base and supplier relationships.
- Two software environments — CUDA-based outside China, Ascend-native inside — creating long-term switching costs for developers and enterprises.
- Two capital allocation stories: one centred on U.S. and allied AI infrastructure, one centred on Chinese sovereign compute.
For investors, company leaders, and executives watching the AI race, this is the important signal. It is not a simple story of one winner taking everything. It is about the world dividing into different arenas, each with its own rules and leaders. Geopolitics pushed China to build a resilient local compute system faster than competition alone would have done.
If you are an investor, operating executive, or board member trying to make sense of these AI infrastructure shifts and what they mean for capital allocation in European mid-market companies, reach out directly.