Small batches of Nvidia's H200 accelerators have started arriving in China after the Commerce Department's Bureau of Industry and Security moved from an outright prohibition to case-by-case review of export applications for the H200, AMD's MI325X and comparable processors [1][2]. The permission lands after the demand it was meant to serve has already been redirected: Bernstein estimates Nvidia's share of China's AI chip market could fall to roughly 8% in 2026, down from nearly 40% a year earlier, with Huawei climbing above 50% [3].
That is a share reduced to about a fifth of its prior level in twelve months [4]. The demand side of the same trade tells a consistent story. In a survey cited by TrendForce, executives said Chinese-made chips would account for 46% of their AI accelerator spending over the coming year, up from roughly 30% today [5] - a 16 point shift, or about half again as much domestic silicon by budget share [6].
The licensing regime is not a re-opening so much as a permission slip with conditions. Exporters must satisfy officials that China sales will not reduce supply for American customers, Chinese buyers must follow export rules and customer verification procedures, and the chips are subject to independent testing in the United States [7]. President Trump said on December 8, 2025 that approved Chinese customers could buy H200s under certain rules [8]. "Export controls should evolve with changes in technology, while protecting national security," said Under Secretary of Commerce for Industry and Security Jeffrey Kessler [9].
Beijing then narrowed the opening further. Cryptopolitan reported in July that Alibaba, ByteDance and DeepSeek were positioned to receive chips, but that authorities would clear orders of only about 200,000 H200s, well short of what was requested, restricted to AI training on public data and barred from inference or sensitive workloads [10].
Nvidia has already priced this in. The company reported first quarter FY2027 revenue of $81.6 billion on May 20, up 85% year over year, with $75.2 billion from data center [11]. Guidance for the following quarter of roughly $91 billion, about 11.5% sequential growth, includes no data center compute sales to China at all [11][12]. Chief executive Jensen Huang said in May that "Huawei is very, very strong" and that Nvidia had "largely conceded that market to them" after years of tightening US restrictions [13].
The replacement demand is being underwritten. TrendForce noted that China plans to invest around 2 trillion yuan, about $294 billion, in data centers over the next five years, with at least 80% of core technology including chips expected to come from local manufacturers [14] - roughly $235 billion of procurement reserved for domestic suppliers before a single order is placed [15].
What to watch: whether the 200,000 unit ceiling is actually filled, or whether Chinese buyers under-order because the training-only restriction makes the parts awkward to slot into existing fleets [10]. Watch also for the first quarter in which Nvidia puts a China data center number back into guidance [11], and whether the 46% survey figure shows up in reported capex rather than in stated intent [5].
