A Monthly Review essay argues that Chinese chipmakers are eroding Nvidia’s hold on China’s AI market, while AI tools are making it easier for developers to move code away from Nvidia hardware. Its case is pointed, and its evidence is worth separating from its political conclusions.
NVIDIA Watch analysis
What happened
The essay says Nvidia’s share of China’s AI-chip market fell to about a third in the first half of 2026, citing research firm IDC. It also points to Z.ai’s GLM-5.2 model, which the article says was trained entirely on Huawei chips, and argues that developers can now adapt code written for Nvidia’s CUDA platform to rival chips more quickly.
For the CUDA claim, the essay relays a startup’s account to Wired that it can get major new open models running on AMD chips within hours, at more than four-fifths of the efficiency. Those are the article’s reported examples, not a broad independent comparison of AI hardware.
Why it matters
Nvidia’s advantage is not just the chips it sells. CUDA and the software built around it have helped keep developers inside its ecosystem. If rivals can run more AI workloads without Nvidia hardware, that could weaken one part of the company’s moat, even if Nvidia remains a major supplier.
The essay also argues that US export restrictions have not stopped China’s chip industry from expanding. That is a wider political interpretation, not a settled verdict on the effectiveness or aims of the restrictions. Its useful technical question is narrower: can Chinese and other rival chips run increasingly capable models well enough to offer developers a practical alternative?
Our read
This is an argument, not a neutral market survey, and Monthly Review’s wider account of monopoly and geopolitics should be read as its analysis. But the reported examples make the software question concrete: if models can help translate workloads across chip platforms, CUDA’s switching costs may be less formidable than they once were. The proof will be in repeatable performance, not a single promising anecdote.
What to watch
- Whether independent figures support the reported change in Nvidia’s share of China’s AI-chip market.
- How well models trained on Huawei chips perform on broader, comparable tests.
- Whether developers can move CUDA workloads to rival hardware reliably, and at what performance cost.
Discussion spark: If developers can move AI workloads between chip platforms more easily, does that meaningfully weaken Nvidia’s moat, or do performance and software support still keep CUDA ahead?
Sources and evidence
- Nvidia’s monopoly breaks, and the chip blockade couldn’t save it – Monthly Review (18 November 2026, 12:01 UTC)
Independent WittyWires tracker for public updates about NVIDIA. Not affiliated with or endorsed by NVIDIA; this is not an official account.