SAN FRANCISCO, March 5, 2026, 13:15 (PST)
- Nvidia has halted production of its H200 AI chips for China, moving capacity at TSMC over to Vera Rubin, the Financial Times reported.
- H200 shipments to China haven’t moved past U.S. export control licensing, and according to a U.S. official, not a single unit has been sold into the country.
- Nvidia is projecting quarterly revenue around $78 billion, while making clear it’s not counting on any data-center compute sales from China.
NVIDIA Corporation has stopped making H200 AI chips meant for China and shifted that production at Taiwan Semiconductor Manufacturing Co to focus on its upcoming Vera Rubin line, the Financial Times said Thursday. Just last week, Nvidia disclosed it had obtained U.S. licenses to supply “small amounts” of H200 chips to Chinese customers. Reuters was unable to independently confirm the FT’s account; TSMC wouldn’t comment, and Nvidia did not reply to Reuters’ inquiry. Reuters
The reported change comes as U.S. export controls continue to block Nvidia’s China business, throttling sales of its advanced chips. Asked about H200 deliveries to China, David Peters, assistant secretary for export enforcement, told a House hearing last week, “My understanding is that none so far.” Last month, the Trump administration officially approved H200 chip sales to China, but under the current framework, shipments are still on hold. Reuters
Nvidia, the top name in AI chip supply, last week said it expects first-quarter revenue to hit around $78.0 billion—not counting any data center compute sales from China in that number. For the quarter ended Jan. 25, revenue landed at $68.1 billion. “Vera Rubin will extend that leadership even further,” CEO Jensen Huang said in the earnings release. NVIDIA Newsroom
The H200 isn’t your standard gaming card. It’s a data-center GPU, designed for the demanding calculations that drive today’s AI engines. Here, the high-stakes competition centers on these processors—ones destined for massive server farms, not home PCs.
Should Nvidia be diverting TSMC capacity from H200 chips meant for China, that move could tighten future availability if export rules ease up. It’s a sharp example of how a hangup in licensing can ripple fast through the entire foundry schedule.
Nvidia CFO Colette Kress told analysts on last week’s earnings call the company has shipped out “our first Vera Rubin samples,” and is targeting broader rollout of those chips in the back half of 2026. “We expect every cloud model builder to deploy Vera Rubin,” Kress said. In its annual report, Nvidia also warned that future revenue could take a hit if there’s a squeeze on data centers, energy, or capital. Business Insider
Nvidia plans to reveal more details from its roadmap during the upcoming GTC conference in San Jose, set for March 16-19. CEO Jensen Huang will take the stage for a keynote on March 16. “AI is no longer a single breakthrough or application — it is essential infrastructure,” Huang said in a company statement dated March 3 that announced the event. NVIDIA Newsroom
Enforcement is now at the center of the China debate. U.S. officials have flagged concerns over chip smuggling, while China hawks in Washington point to the risk that advanced processors might end up repurposed from commercial projects for military applications.
Nvidia ticked up roughly 0.2% to $183.30 late in the U.S. session. TSMC’s U.S.-listed shares fell close to 1%.
Still, politics drives the real outcome here, not engineering. Nvidia hasn’t said it will alter production, and with U.S. export licenses or China’s import signoff able to flip fast, the market could swing open or shut—never mind that Nvidia left it out of this year’s guidance.
Right now, Nvidia is channeling limited manufacturing resources into Vera Rubin. The upcoming product line gets priority, while the China pipeline remains knotted in red tape and requirements.
