SAN FRANCISCO, March 5, 2026, 03:36 PST
- Nvidia has halted H200 chip production for China. Some of the freed-up capacity is being redirected to its upcoming Vera Rubin hardware.
- Nvidia’s trimmed-down offerings are still getting caught up in export rules, which continue to throttle the company’s sales pipeline into China.
- This change comes just before Nvidia’s GTC conference later this month, with the company expected to detail what’s next for its AI stack.
Nvidia has halted output of its H200 AI chips designed for China and shifted some of its Taiwan Semiconductor Manufacturing Co production lines to focus on upcoming Vera Rubin hardware, the Financial Times said Thursday. 1
This is significant: Nvidia counted China among its biggest customers for data-center accelerators, the key chips powering large AI models. With U.S. export rules changing, Nvidia has been working to hang onto its slice of the Chinese market.
There’s also the near-term snag: TSMC’s production slots aren’t unlimited. Giving priority to one chip family knocks delivery timelines, reshuffles customer orders, and can disrupt plans for the next round of products.
The FT report points to weak near-term sales hopes for China as the driver behind the H200 move. Reuters was unable to confirm the story on its own, and requests for comment from Nvidia and TSMC were not answered right away.
Nvidia disclosed last week that it got U.S. government approval to supply “small amounts” of H200 chips to Chinese clients. Still, according to Reuters, a Commerce Department official said last month that no sales to China had actually gone through. The report also noted that although President Donald Trump’s administration formally authorized H200 exports to China back in January, shipments hadn’t moved forward due to procedural hurdles.
Nvidia’s production call lands just before its GTC developer conference in San Jose, set for March 16-19. The company will be unveiling new hardware and software aimed at what it’s branded as AI factories—data centers specifically designed to generate AI output at scale. 2
CEO Jensen Huang called AI “essential infrastructure” in a company statement this week, as he announced the event. “AI is no longer a single breakthrough or application,” Huang said.
Just a day before, Huang took a different stance on Nvidia’s investment plans, hinting that the wave of large equity injections into leading AI labs could be winding down as these firms look toward going public. “The opportunity to invest $100 billion in OpenAI is probably not in the cards,” he remarked Wednesday during Morgan Stanley’s Technology, Media and Telecom conference, as quoted by Reuters. 3
Competitive heat is rising fast. Broadcom on Thursday pegged its 2027 AI chip sales at over $100 billion, sharpening its push into Nvidia’s territory. AMD, too, is working to carve out more of the fast-growing data-center AI pie. 4
The China move isn’t without danger. Should Washington relax licensing rules or Beijing begin allowing wider imports, Nvidia might not have enough China-tailored stock on hand, having wound down H200 production. Bringing foundry output back up takes time—it’s not a quick flip. And the longer the freeze lasts, the more room competitors and local upstarts have to entrench themselves in what used to be a core growth market for Nvidia.
