Beijing’s Cold Shoulder: Why Nvidia’s H200 Chip Still Can’t Crack China

When Jensen Huang made a last-minute dash to Beijing this week, joining other American executives at the Trump-Xi summit, industry watchers saw a flicker of hope. Maybe, after months of regulatory gridlock, Nvidia would finally get the green light to sell its advanced H200 AI processors into China. That hope evaporated by Friday. The summit ended with no breakthrough, and the H200 remains a product in name only—zero units shipped, zero approvals granted. Beijing is signaling it would rather nurture its own chip champions than open the door for U.S. silicon.

The Trump administration gave the green light for H200 sales late last year, but Chinese authorities never issued the final purchase permits. U.S. Trade Representative Jamieson Greer acknowledged the standoff in a Bloomberg interview, calling it a sovereign decision for China. “Obviously we think it could be helpful to them,” he said, “but they’ll just have to make their decision.” The fact that chip export controls never even came up in high-level talks underscores how little leverage Washington holds. Beijing is focused on building domestic alternatives.

Just days before the summit, Chinese startup DeepSeek announced its latest AI model had been optimized specifically for Huawei chips—a symbolic win for state-backed efforts to reduce reliance on foreign technology. Huawei’s Ascend series is gaining traction among local developers who once defaulted to Nvidia. Meanwhile, the H200 sits in regulatory limbo. Chinese customs blocked initial shipments earlier this year, and parts suppliers paused production. Regulators told tech giants they could buy the chips only under “special circumstances,” like university research. The directive was deliberately vague.

Additional conditions attached to U.S. approval add friction: third-party labs must test each batch, shipments to China can’t exceed half the volume sold domestically, and buyers must prove strong security measures with no military use. At $30,000 a unit, the financial stakes are high. Congress has also waded in, advancing bills to review future licenses. Bipartisan frustration targets both the White House and Nvidia, with some lawmakers arguing that even limited sales accelerate China’s military AI ambitions.

Trump struck an optimistic note after the summit, saying China “chose not to” buy the H200 because officials want to develop their own technology. “I think something could happen on that,” he added. But for now, the stalemate holds. Nvidia reported no data-center revenue from China tied to the H200 in recent quarters. Domestic champions like Huawei grow stronger by the month, backed by government procurement policies, subsidies, and talent programs that recruit overseas engineers.

The broader picture: national security priorities now shape billion-dollar technology decisions on both sides of the Pacific. Neither capital appears ready to blink first. For Nvidia, the H200 is approved on paper but collecting dust in practice. For Chinese cloud providers, the dilemma is acute—they need massive compute clusters today to match American rivals, but waiting for domestic chips risks falling behind, and buying American invites regulatory whiplash. The result is a fragmented market where strategic competition trumps commercial opportunity.