A striking twist has emerged in the U.S.-China tech rivalry this year: Washington has greenlit sales of Nvidia Corp.’s (NVDA) high-end H200 AI chips to Chinese firms, yet not a single unit has been shipped—even as export licenses sit unused. The chip giant, once the unchallenged leader in China’s advanced semiconductor market, now finds itself trapped in a painful dilemma: unable to capitalize on the U.S. approval, yet unwilling to cede a market that once drove a tenth of its revenue.
The U.S. nod is hardly a goodwill gesture—it is a compromise laced with strings. In January, the Commerce Department authorized roughly 10 Chinese companies, including Alibaba Group Holding Ltd., Tencent Holdings Ltd., ByteDance Ltd. and JD.com Inc., to purchase the H200. Lenovo Group Ltd. and Foxconn Technology Group (2317.TW) secured distributor status, with each approved buyer allowed to acquire up to 75,000 chips, either directly from Nvidia or via intermediaries.
But the Trump administration’s conditions are punitive. It is demanding a 25% cut of all H200 sales revenue, a provision structured to route chips through U.S. territory first—skirting legal restrictions on direct export taxes. Chinese buyers must also prove they have “sufficient security procedures” in place and that the chips will not be used for military purposes, while Nvidia is required to certify adequate U.S. inventory. Crucially, the H200 is not Nvidia’s cutting-edge offering: it lags behind the Blackwell-architecture B200 in performance, a deliberate choice that lets Washington extract value from older technology while safeguarding its AI dominance.
What makes the situation more awkward is China’s sharp pivot from scrambling for chips to shunning them. As of May 14, deliveries remain at zero. U.S. Commerce Secretary Howard Lutnick conceded at a Senate hearing last month that Beijing is “directing investment toward its own domestic industry,” the root cause of the standoff.
Two factors underpin China’s resistance. First, new supply chain security rules from the State Council mandate audits of foreign dependencies in critical tech infrastructure, stoking fears that the U.S.-routed H200 chips could harbor backdoors or be subject to tampering. Second, domestic alternatives are gaining traction fast. Huawei Technologies Co. and DeepSeek Inc. have made rapid inroads, with Chinese manufacturers now holding 41% of China’s AI accelerator server market. Nvidia’s own share of China’s AI chip market has effectively collapsed to zero, CEO Jensen Huang has acknowledged. For Beijing, short-term computing needs can be met by homegrown chips; long-term, avoiding reliance on foreign technology is non-negotiable.
Nvidia is the biggest loser in this standoff, its market position teetering. At its peak, the company controlled 95% of China’s high-end chip market, with the country accounting for 13% of its total revenue. But since U.S. export controls tightened, its China revenue fell 21.2% year-over-year in fiscal 2025. The H200 was meant to stem that decline, yet the 25% U.S. revenue cut—applied to chips priced between $27,000 and $40,000 each—has sliced its gross margin on these sales in half.
Huang made a last-minute push to break the deadlock, joining President Donald Trump’s delegation to Beijing unannounced. But in the context of broader U.S.-China tensions, individual diplomacy is unlikely to reverse the tide.
The impasse signals a fundamental shift in the tech rivalry: Washington has moved from total blockade to “conditional approval” as a way to squeeze economic gains, while Beijing has shifted from passive acceptance to actively building a self-reliant supply chain. For Nvidia, the H200 paradox is a warning: fail to navigate the clashing demands of the two superpowers, and it could lose China for good. For China, rejecting the H200 means short-term trade-offs in computing power—but it buys critical time for domestic chipmakers to close the gap, a necessary step on its path to technological independence.
