SAN FRANCISCO, March 5, 2026, 07:17 (PST)
- Nvidia has halted production of its H200 AI chips for China, the Financial Times reported, shifting TSMC’s manufacturing lines over to its next-gen Vera Rubin hardware instead.
- Still, Nvidia said it holds U.S. licenses allowing it to send “small amounts” of H200 chips into China.
- Nvidia slipped roughly 0.3% early in the U.S. session.
Nvidia has halted production of its H200 AI chips for China, according to the Financial Times, pivoting its Taiwan Semiconductor Manufacturing Co capacity to focus on the upcoming Vera Rubin line. Shares of Nvidia slipped roughly 0.3% early in U.S. trading.
If the change sticks, it’s a clear sign that export controls and approvals are now steering the company’s supply moves, despite continued strong demand for its data-center chips.
Nvidia’s H200 sits among the company’s leading data-center processors, powering server farms that handle the training and operation of large AI models. Yet, shipments to China have hit roadblocks, with regulatory clampdowns from both Washington and Beijing slowing the flow of advanced computing chips.
Nvidia and TSMC have yet to reply to requests for comment. Independent verification of the Financial Times report wasn’t possible.
Nvidia last week noted its current-quarter outlook leaves out any potential revenue from data-center chip sales to China. The company also disclosed it now holds U.S. licenses to send “small amounts” of H200 chips to certain customers in China. Reuters
David Peters, the U.S. Commerce Department’s export enforcement chief, told lawmakers last month that Chinese buyers haven’t received any H200 chips. “My understanding is that none so far,” Peters said, pointing to ongoing shipment delays due to “guardrails”—those built-in sales conditions and checks. Reuters
Back in January, Washington gave the green light to H200 sales destined for China. Seaport Research’s Jay Goldberg described the restrictions as a compromise—and a tricky one to enforce. “Put another way, this looks like a Band-Aid,” he said. Reuters
Nvidia’s manufacturing shift signals the start of its next phase. Back in January, the company announced that its Rubin platform—named for astronomer Vera Rubin—had reached “full production,” with Rubin-driven offerings set to launch through partners in the latter half of 2026. “Rubin arrives at exactly the right moment,” chief executive Jensen Huang said. NVIDIA Newsroom
Huang, speaking Wednesday, pointed out Nvidia’s recent equity moves in top clients could be tapering off, with several eyeing IPOs. A $100 billion outlay for OpenAI? “Not in the cards,” he said. According to Huang, the latest rounds in both OpenAI and Anthropic may mark Nvidia’s final investments in those two, as they get ready to hit public markets. Reuters
Still, things in China aren’t standing still. If U.S. policy shifts or Beijing alters its approach, demand could come back. On the other hand, stricter enforcement or fresh limits would likely freeze orders, sending buyers looking elsewhere.
At this point, reallocating foundry slots signals Nvidia’s focus on what it can realistically deliver. That decision leaves the H200 caught in limbo—a part designed for compliance, yet still ensnared in the political mess.
