Nvidia disclosed on its earnings call that AI inference demand has driven rental prices up across GPU generations, with multi-year Hopper renewals reaching up to 2x the original price and B300 up over 60% since launch. CFO Colette Kress said token prices have fallen nearly 60% since May, spurring usage and compute demand, leaving capacity at hyperscalers and neoclouds fully utilized. Nvidia is also reported to require overseas data center partners to allocate at least 51% of services to non-Chinese customers, tightening advanced compute access for Chinese firms.
Termsslug: hopper
Hopper
Hopper 架构
Nvidia data center GPU microarchitecture featuring a Transformer Engine and high-bandwidth memory design.
Curated Intelligence
IT之家 · IT业界T2·Infra
78 pts
The Verge AIT2·Industry
78 pts
Interim CEO Colette Kress confirmed on the earnings call that Nvidia has stopped shipping licensed Hopper data center products to China, with guidance no longer including that revenue. The move follows White House approval for H200 exports that Beijing never matched with import clearance, leaving roughly 600,000 H200 chips stranded in Hong Kong and marking a sharp reversal of Nvidia's China strategy.