SAN FRANCISCO, March 6, 2026, 03:58 PST
- Nvidia has halted manufacturing of its H200 AI chips for China, according to a report, moving production resources over to its upcoming Vera Rubin platform.
- That move comes while U.S. officials are weighing new limits on where AI chips can be exported—even to friendly countries.
- Nvidia CEO Jensen Huang hinted the company’s bets on OpenAI and Anthropic could be winding down, with both AI startups now looking toward IPOs.
Nvidia (NVDA.O) has halted output of its H200 AI chips designed for China, rerouting manufacturing at Taiwan Semiconductor Manufacturing Co toward its upcoming Vera Rubin line, according to the Financial Times, which cited two sources familiar with the shift. Reuters was unable to independently confirm the details. TSMC declined comment. Nvidia also did not reply to a request for comment.
This shift is significant—U.S. export restrictions have transformed Nvidia’s once-simple China sales into a maze of permits and compliance. Product plans keep getting reworked, and now the uncertainty is starting to affect production choices as well, not just the deals with customers.
Buyers scrambling to set up new AI data centers are running into this at a tough time. As Nvidia shifts its top-tier chip production over to the latest model, delivery timelines can get knocked off balance region by region—even though shipments heading for China were already squeezed.
Nvidia recently said it’s secured U.S. government licenses to deliver just “small amounts” of H200 chips to customers in China, a sign that the market is unlikely to ramp up in the near term. The H200 ranks as Nvidia’s second-most advanced AI chip, just behind its latest flagship lineup, according to the Financial Times report.
U.S. policymakers are weighing a plan that could slap fresh conditions on big AI chip exports, a document reviewed by Reuters shows. Foreign buyers might be compelled to invest in domestic AI data centers or give security assurances. The draft also sharpens compliance demands for chip producers including Nvidia and Advanced Micro Devices.
The proposal would require exporters to keep an eye on shipments, while recipients must run software that prevents chips from being combined into a so-called “cluster”—industry jargon for the sprawling chip arrays behind major AI training and operations.
“The rule could help the U.S. government address chip diversion to China and ensure a more secure buildout of the most powerful AI supercomputers,” said Saif Khan, a former Biden administration national security official who’s now with the Institute for Progress. Still, he flagged the licensing requirements as “overly broad,” and questioned whether the controls might end up being used as leverage against allies.
According to a Reuters report, the Commerce Department said on X it’s considering new rules but pushed back on adopting the previous administration’s “burdensome” framework. The rules remain unfinished and may be revised.
Nvidia is working to temper investor expectations about just how much support it’s willing to throw behind its largest clients. On Wednesday at a Morgan Stanley event, CEO Jensen Huang made it clear: “the opportunity to invest $100 billion in OpenAI is probably not in the cards”—not with the company gearing up for an IPO. Huang also indicated that Nvidia’s bets on OpenAI and Anthropic could be its last with those startups. Reuters
The competition is getting tougher. If the U.S. rolls out a broader licensing regime, it won’t just hit Nvidia—AMD and other AI chip exporters would be in the mix. That could force customers to rethink how they divide up chip orders, hinging on what each supplier is allowed to deliver, and the conditions attached.
Policy remains the wild card here. The export plan isn’t final yet, and earlier chip regulations have changed fast as national security clashes with lobbying from the tech sector. Should Washington decide to tighten controls again — or if compliance gets tricky enough to put buyers off — Nvidia may face sudden swings in demand across different countries, just as it’s working to scale up its new chip lineup.
