Nvidia tightens restrictions on cloud GPU access for Chinese AI firms
Chinese artificial intelligence companies face mounting pressure to secure Nvidia compute capacity as the chipmaker enforces stricter policies on overseas cloud rentals and export-compliant hardware.
Chinese AI laboratories are encountering significant obstacles in obtaining Nvidia processing power through their two primary channels: purchasing export-approved processors and leasing capacity from international cloud providers. According to reporting by the Financial Times, Nvidia's enforcement actions are contributing to these constraints. The chipmaker has instructed its cloud service partners across the globe to allocate a minimum of 51% of their resources to customers outside China. In response, Chinese firms have acquired over 1 million Nvidia gaming and workstation cards during the current year to compensate for the shortage. Pricing pressures are intensifying as well: a B300 rack that commands approximately $600,000 on the international market reportedly sold for $2.8 million within China in recent weeks.
The demand for export-compliant alternatives underscores the severity of the supply crunch. Chinese purchasers have placed orders exceeding 6 million units of Nvidia's RTX Pro 5500 cards, which meet export regulations. Manufacturing capacity presents a critical bottleneck, with Nvidia reportedly capable of producing only around 3 million of these cards by 2027. According to one source who spoke with CNBC, at least one Chinese company, Moonshot, received a proposal requiring it to pay 30% to 50% more in upfront costs relative to pricing offered to non-Chinese entities.