The ban on the China-only Blackwell card landed during the Trump delegation’s state visit, on which the Nvidia CEO was a late addition. Chinese AI buyers had been using the 5090D V2 as a workaround for the H200 procurement vacuum.
China stopped granting import permits for Nvidia’s RTX 5090D V2 gaming card on 15 May, the same week Jensen Huang was in Beijing alongside Donald Trump’s state-visit delegation, Financial Times reports.
The ban applies to the China-only Blackwell-architecture card NVIDIA introduced last August specifically to comply with US export controls.
The 5090D V2 had been marketed in China to gamers and 3D animators on the published-spec sheet. In practice, Chinese AI buyers cut off from the more advanced H100, H200 and Blackwell-class data-centre lines had been using the card as a workaround, because it retained the Blackwell architecture and could be racked at scale for training and inference workloads outside the explicit export-control framework.
Huang’s presence in Beijing during the ban window was itself a late addition to the Trump delegation. On 13 May Huang joined the China trip after a phone call from the president, picked up in Alaska as Air Force One refuelled.
The NVIDIA CEO sat alongside Tim Cook, Elon Musk and other US tech leaders for the formal state-visit programme.
The wider Chinese-domestic chip-procurement directive the ban sits inside has been in escalation across the spring. Beijing’s instructions to its largest AI companies is to stop acquiring NVIDIA processors, including the H20 and the RTX Pro 6000D, on the framing that Huawei’s Ascend line and Cambricon’s Siyuan accelerators now match those products on the relevant workloads.
Alibaba’s T-Head Zhenwu M890 launch is the same procurement-directive context expressed on the chip-design side, with the company’s executive claiming ‘scaled mass production’ for the domestic alternative.
The US-side framing is sharper. Trump’s own comments earlier this month stated that China is blocking H200 purchases despite the US having approved the export licences in question. The framing positions Beijing’s procurement policy as the active constraint on Nvidia’s China revenue line, rather than US export controls, an unusual diplomatic position for a US administration to take publicly.
The Trump-Xi Beijing summit on AI guardrails left the procurement and licensing questions on the same agenda. The 5090D V2 ban is the first visible enforcement step Beijing has taken since the summit ended.
On the commercial impact, NVIDIA guided to $91bn of Q2 revenue yesterday against a $86.84bn consensus, with the prepared remarks framing the China line as ‘small but material’ and the broader data-centre demand picture as still in early innings.
The 5090D V2 is, on the published unit-economics, a low-single-digit percentage of Nvidia’s quarterly revenue base. The signalling value is the more consequential read. The RTX 5090 itself, the global card, has separately climbed to $5,300 on the Korean grey market on procurement-pressure-driven demand.
NVIDIA has not issued a public statement on the 5090D V2 ban beyond confirming Huang’s participation in the delegation. China’s customs administration has not yet published a formal notice describing the permit decision in writing, and the underlying basis (national-security review, anti-competition framing, or simple procurement-policy enforcement) has not been disclosed.
The next visible proof point will be the Q2 earnings disclosure of China-specific revenue, scheduled inside Nvidia’s August reporting cycle, when the V2 ban’s contribution to the China line becomes formally visible.
