Under the revised terms, customers must pay the full amount at the time of ordering, with no option to cancel, seek refunds or alter configurations after orders are placed. In limited cases, buyers may be allowed to provide commercial insurance or asset collateral instead of cash, one of the sources said. Both people spoke on condition of anonymity because the policy is not public.
While Nvidia has historically required advance payments from Chinese clients, those arrangements previously allowed partial deposits in some cases. The H200, however, is subject to unusually strict enforcement, given the lack of clarity surrounding regulatory approvals in China. Nvidia and China’s industry ministry did not respond to requests for comment at the time of publication.
Chinese technology firms have already placed orders for more than two million H200 chips, each priced at about $27,000, according to a Reuters report last month. That demand far exceeds Nvidia’s current inventory of roughly 700,000 units, underscoring the importance of the Chinese market even as regulatory hurdles persist.
Domestic alternatives remain limited. Although Chinese chipmakers such as Huawei have introduced AI processors like the Ascend 910C, those products still trail Nvidia’s H200 in performance for training large-scale, advanced AI models, making the U.S. firm’s chips highly sought after by China’s leading internet and technology companies.
Regulatory signals from Beijing remain mixed. Bloomberg reported on Thursday that China may approve some H200 imports as early as this quarter, allowing purchases for select commercial uses while barring the military, sensitive government agencies, critical infrastructure and state-owned enterprises on security grounds. At the same time, Chinese regulators have reportedly asked some tech firms to temporarily pause H200 orders while officials determine how many domestically produced chips must be purchased alongside each Nvidia chip. The Information first reported the pause earlier this week.
Nvidia chief executive Jensen Huang said on Tuesday that demand for the H200 was “quite high” and that the company had accelerated its supply chain to increase output. He added that he did not expect a formal approval announcement from China, noting instead that the ability to place purchase orders would signal regulatory clearance.
The tougher payment structure highlights Nvidia’s attempt to balance surging Chinese demand against regulatory uncertainty on both sides of the Pacific. Although the Biden administration had previously banned exports of advanced AI chips to China, President Donald Trump reversed that policy last month, allowing H200 sales subject to a 25 per cent fee paid to the U.S. government.
Nvidia’s caution is informed by past losses. Last year, the company wrote down $5.5 billion in inventory after a sudden ban on sales of the H20 chip to China. While the U.S. later reversed that restriction, China subsequently blocked H20 shipments, leaving Nvidia exposed to shifting policy decisions.
By requiring full upfront payment for the H200, Nvidia effectively transfers much of the financial risk to its customers, who must commit capital without certainty that imports will be approved or that deployment plans will proceed as intended. For Chinese buyers, however, the chip represents a major upgrade: the H200 delivers roughly six times the performance of the now-blocked H20 and is currently Nvidia’s second-most powerful AI processor.
Initial H200 orders are expected to be fulfilled from existing inventory, with the first shipments anticipated before the Lunar New Year holiday in mid-February. Nvidia has also approached Taiwan Semiconductor Manufacturing Co to expand production capacity, with additional manufacturing slated to begin in the second quarter of 2026.
Expanding capacity comes at a challenging time for the company, which is simultaneously transitioning from its Blackwell chips to the next-generation Rubin architecture and competing with major players such as Alphabet’s Google for advanced manufacturing slots at TSMC.