Nvidia is requiring Chinese customers to pay in full upfront for its H200 artificial intelligence chips, according to two sources familiar with the matter. The move aims to hedge against ongoing uncertainty over whether Beijing will approve the shipments. The U.S. chipmaker has imposed unusually strict terms for the H200: orders cannot be canceled, refunded, or reconfigured once placed. In rare cases, Nvidia may accept commercial insurance or asset collateral in lieu of cash.
While Nvidia has previously required advance payments from Chinese clients, it often allowed deposits rather than total upfront costs. Sources say the company is tightening enforcement because it remains unclear if Chinese regulators will greenlight the imports. Nvidia and China’s industry ministry did not immediately respond to requests for comment.
Regulatory Tug-of-War The payment structure effectively transfers financial risk from Nvidia to its customers. Chinese tech giants, including ByteDance, view the H200 as a vital upgrade. It offers roughly six times the performance of the H20 chip—a lower-spec model Nvidia designed for China that was recently banned by Beijing. The H200 demand is massive. Reuters reported last month that Chinese firms have ordered over 2 million units, priced at roughly $27,000 each. However, Beijing recently asked some companies to pause orders while regulators decide how many domestic chips (such as Huawei’s Ascend 910C) customers must buy alongside Nvidia products.
Navigating Policy Shifts Nvidia is walking a tightrope between surging demand and volatile geopolitics. • U.S. Policy: The Biden administration originally banned advanced AI exports to China. President Donald Trump reversed this last month, allowing H200 sales subject to a 25% U.S. government fee. • Past Losses: Nvidia remains cautious after writing down $5.5 billion in inventory last year when the U.S. abruptly banned the H20 chip.
Supply Chain Pressure Nvidia CEO Jensen Huang confirmed Tuesday that demand is "quite high" and the supply chain is ramping up. Initial orders are expected to ship from existing stock before the Lunar New Year in mid-February. To meet the 2-million-chip backlog, Nvidia has asked Taiwan Semiconductor Manufacturing Co (TSMC) to increase production, with additional manufacturing slated for the second quarter of 2026. However, capacity remains tight as Nvidia also transitions to its next-generation "Rubin" architecture while competing for factory space with firms like Google.
