Nvidia to demand full upfront payment for H200 GPUs from China customers, report claims — more than two million chips may have been ordered despite uncertain Beijing stance
Nvidia now demands full advance payment for its H200 GPUs for AI applications from its customers in China amid uncertainties with approvals of H200 imports, reports Reuters, citing sources with knowledge of the matter. Clients in the People’s Republic also cannot cancel orders even if the government bans them from importing them to the country.
Under the new terms, Chinese buyers must pay 100% of the H200 order value at placement with no option to change configurations afterward. In limited cases, customers may substitute cash with commercial insurance or asset collateral, but the overall approach is far stricter than Nvidia’s terms for Chinese clients earlier, which sometimes allowed partial deposits rather than full prepayment, according to Reuters.
Although it is expected that the Chinese government will approve H200 imports early in 2026, the situation remains particularly uncertain as authorities are projected to allow H200 imports only for selected customers in the commercial sector and under certain conditions, so it is natural for Nvidia to hedge its risks.
Purchases by the military, sensitive government organizations, critical infrastructure operators, and state-owned enterprises are expected to remain barred due to security concerns. Meanwhile, regulators have asked certain Chinese firms to temporarily pause orders while they are figuring out how many domestically produced AI accelerators must be bought alongside each imported H200.
Despite strict terms on Nvidia’s side, demand for H200 remains robust in China. Chinese technology companies have ordered more than two million H200 processors priced at roughly $27,000 per unit, which exceeds Nvidia’s available inventory of around 700,000 units.
Nvidia plans to fulfill initial Chinese H200 orders from existing stock, and the first deliveries are expected before the Lunar New Year in mid-February. To get the remaining 1.3 million units or more, Nvidia has to place new orders with TSMC (or rather allocate its pre-paid production capacity from other products to H200), make the silicon, and then package it using CoWoS-S technology, which takes over three months. As a result, the company will be able to ship additional H200 units to Chinese clients only sometimes in Q2 at the earliest.
