Skip to content
Sıfırıncı Dakika
Latest

Nvidia cracks down on overseas cloud AI GPU rentals for China

Hardware1 min read

In brief

Nvidia has told cloud partners worldwide that at least 51% of their services must go to non-Chinese firms. Chinese AI labs are turning to overseas GPU rentals as an alternative, while B300 rack prices in China reportedly surged to $2.8 million.

  • 51% non-Chinese customer quota for cloud partners
  • Rule communicated verbally and applied globally
  • Existing deals unchanged, new ones must comply
  • Fleet Intelligence tracking required for Southeast Asian providers
  • B300 rack prices in China reportedly hit $2.8 million

A quiet but global restriction from Nvidia

Nvidia has told cloud partners worldwide that at least 51% of their services must be allocated to non-Chinese customers. According to the Financial Times, the instruction was communicated verbally through sales and regional teams; existing deals are unaffected, but new agreements must comply.

Both routes are narrowing for Chinese labs

Chinese AI companies are caught between buying export-compliant cards and renting compute abroad. Nvidia's whitelist more than halved its authorized Asian buyer count, pushing Chinese groups toward overseas leases. Now that route is being restricted too.

Prices and demand

A B300 rack that sells for about $600,000 internationally reportedly fetched $2.8 million in China last month. Chinese customers have ordered more than 6 million export-compliant RTX Pro 5500 cards, at least twice the 3 million Nvidia could reportedly manufacture in 2027.

Moonshot and others look to Europe

Moonshot has reportedly held talks with cloud providers hosting Nvidia Blackwell systems in Europe for clusters of 8,000 or more GPUs. Tencent has a deal with Oracle to lease 100,000 GPUs in Southeast Asia; ByteDance and Alibaba are pursuing similar long-term leases.

Domestic supply falls short

China's domestic AI chip supply is expected to reach about 3.5 million units this year against demand exceeding 5 million. Huawei's Ascend 910C outperforms lower-end cards in most areas, but reliance on the CUDA ecosystem keeps Chinese firms tied to Nvidia solutions.

Why it matters

This shows how AI compute is being allocated and how export controls work in practice, which could affect model training costs and access going forward.

Sources

Related stories