Nvidia, 중국향 H200 칩 생산을 중단하고 생산 능력을 Vera Rubin으로 전환한다고 Startup Fortune이 보도했다.

Nvidia built roughly a quarter million H200 chips meant for Chinese buyers. Now it has stopped making more and handed that factory capacity to its next AI platform instead.

Nvidia has halted production of H200 chips earmarked for China and redirected the freed-up manufacturing capacity at TSMC toward its next-generation Vera Rubin platform, according to a report from the Financial Times.

The timing is telling. Washington granted Nvidia export licenses for the H200 to China in January, and the Trump administration eased restrictions further after that. Four months on, Commerce Secretary Howard Lutnick has said Nvidia still hasn't sold a single one. Beijing, not Washington, is now blocking the deal: Chinese customs authorities have barred the chips from import, and regulators have pushed domestic buyers toward homegrown alternatives from SMIC and Huawei.

Nvidia is doing the rational thing. It is not scrapping the inventory, and it still has those 250,000 units on hand if the standoff ever breaks. But it is not building more, and it is pointing TSMC's wafer capacity at a platform with a real market waiting for it.

While Nvidia steps back from the China trade on its own terms, US officials are still debating how much of it should be allowed. The administration is weighing a cap limiting Nvidia and AMD advanced chip sales to 75,000 units per Chinese customer, according to Bloomberg. Roughly ten Chinese firms, including Alibaba, Tencent, ByteDance and JD.com, already hold licenses under that rough ceiling, with Lenovo and Foxconn authorized as distributors.

The cap debate assumes there is a live trade to regulate, and there isn't. Beijing has told Chinese companies to keep any approved H200 use confined overseas, which defeats the purpose for buyers who wanted the chips running in their own data centers. Licenses exist, customers exist, and still no chips have shipped.

This is Beijing making a choice, not a loophole closing. China would rather its AI companies run slower domestic silicon than depend on an American chip it cannot fully control, export terms and all. It is a bet on self-sufficiency over performance, and the Chinese government has made it deliberately.

**Vera Rubin is where the real money is**

Nvidia did not need much convincing to look elsewhere. Vera Rubin is expected to reach volume production in the second half of 2026, and the Western market for it comes with none of the licensing fights, customs blocks, or geopolitical complications attached to China. Demand from hyperscalers in the US, Gulf states and Europe is already outstripping what TSMC can produce, so every wafer not spent on an unsellable H200 is a wafer that ships to a customer who will actually take delivery.

This is the quiet admission beneath the FT's reporting: the China AI chip trade that Washington spent two years negotiating licenses for is functionally dead, at least for now. Nvidia did not need a press release to say so. It simply stopped making the chips and moved on.