Nvidia's most advanced AI server, the B300, is now selling for approximately $1 million per unit in China, nearly double the $550,000 it commands in the United States, as tightening US export controls and a crackdown on chip smuggling choke off the grey-market supply chain that Chinese buyers have relied on for years.
The eight-GPU server climbed from roughly 4 million yuan late last year to about 7 million yuan now, according to four industry sources cited by Reuters, with the acceleration directly linked to the March prosecution in the United States of Yih-Shyan "Wally" Liaw, a co-founder of Nvidia partner Supermicro.
The case severely disrupted the grey-market networks that had been the primary route for restricted Nvidia hardware entering China, sources said.
"The grey market, a key supply channel, came under immense pressure following the US prosecution of a key figure in the supply chain," four people with direct knowledge of the matter told Reuters.
The scarcity premium has been compounded by uncertainty around Nvidia's H200 chip, which received export approvals from both the US and Chinese governments but has yet to ship to China as the two sides remain deadlocked over the conditions governing its sale.
That bottleneck has funnelled additional demand towards the already constrained B300, intensifying the price surge.
Demand from Chinese technology companies shows no sign of easing.
Morgan Stanley estimates that Chinese AI models now account for 32% of global token usage, meaning the country's need for advanced computing power is expanding at the same time supply is being deliberately constrained.
Some Chinese firms are shifting from purchasing servers outright to renting compute capacity, with monthly rental prices for B300 systems surging to 190,000 yuan.
The price divergence has created what analysts describe as a split market, where the same hardware carries two entirely different valuations depending on geography and compliance risk.
In the United States, B300 prices have also risen, from roughly $450,000 late last year to approximately $550,000, reflecting strong global demand from hyperscalers racing to build AI infrastructure.
However, the Chinese premium now runs at roughly 80% above the US price, a gap driven by scarcity and legal risk rather than performance differences.
The supply squeeze arrives at a sensitive moment for Nvidia's competitive position.
Google and Amazon are both preparing to commercialise their own AI chips at scale, with Google unveiling its TPU 8t and TPU 8i at Cloud Next this week and Amazon expanding deployment of its Trainium processors through major deals with Anthropic and Meta.
Related reading
- Nvidia launches open-source quantum AI models as it deepens grip on chip design workflows
- Nvidia says the open versus proprietary AI debate is the wrong argument
- From Vera Rubins, Blackwell, Nemotron, NemoClaw, to Isaac and Cosmos - here's a whistlestop guide to NVIDIA's…
Nvidia shares traded at $199.57, approximately 26% below the average analyst target of $269.17, and are up roughly 14% over the past 30 days.
The stock trades at a price-to-earnings ratio of 40.4 times, below the semiconductor industry average of 48.6 times.
The recap
- AI server prices in China have almost doubled recently.
- Shares trade at US$199.57, about 26% below target.
- Simply Wall St flagged insider selling and non-cash earnings.
