Prices have surged from roughly 4 million yuan late last year to 7 million yuan ($1 million) today, driven by the Supermicro co-founder’s arrest, tighter enforcement, and voracious demand from Chinese AI companies unwilling to hold restricted hardware on their own books.
NVIDIA’s B300 AI servers are now selling in China for approximately 7 million yuan, roughly $1 million per unit, nearly double their US list price of around $550,000, according to Reuters, which cited four industry sources.
The near-doubling from around 4 million yuan late last year reflects a scarcity premium that has intensified sharply since early 2026, as a crackdown on chip smuggling closed the grey-market supply channel that had been a critical workaround for Chinese companies unable to purchase Nvidia hardware through official channels.
Rental prices for B300 servers in China have also surged alongside sale prices, with short-term contracts reportedly reaching as high as 190,000 yuan per month, a figure that underlines the extremity of compute scarcity.
The 💜 of EU tech
The latest rumblings from the EU tech scene, a story from our wise ol’ founder Boris, and some questionable AI art. It’s free, every week, in your inbox. Sign up now!
Chinese technology companies are scrambling for the hardware needed to monetise their AI models and compete on inference efficiency, the cost of generating tokens, the basic unit of AI text output, as commercial AI deployment accelerates.
Many are deliberately structuring arrangements to avoid holding Nvidia hardware directly on their balance sheets, fearing exposure to US sanctions.
The Supermicro bust that choked the grey market
The inflection point in the price surge traces directly to the arrest of Yih-Shyan ‘Wally’ Liaw, co-founder of Supermicro, on 19 March 2026.
US federal prosecutors unsealed an indictment alleging that Liaw, Supermicro’s Taiwan general manager Ruei-Tsang Chang, and contractor Ting-Wei Sun conspired to divert roughly $2.5 billion in Supermicro servers, containing Nvidia’s export-controlled Blackwell-class AI chips, to Chinese buyers via a Southeast Asian pass-through company.
The scheme involved swapping serial numbers using heat guns, fabricating paperwork, and using fake replica servers to deceive government auditors. Liaw and Sun subsequently pleaded not guilty; Chang remains a fugitive.
Supermicro’s shares fell 33% on the day the indictment was unsealed, erasing more than $6 billion in market capitalisation. Liaw resigned from the board immediately.
The prosecution is the highest-profile enforcement action yet under the AI chip export control regime the US has been tightening since 2022, and its effect on grey market supply appears to have been immediate: the arrest signalled to the broader network of brokers, logistics companies, and intermediaries that the enforcement risk had materially increased, sharply reducing the volume of hardware being diverted through unofficial channels.
NVIDIA has publicly distanced itself from the grey market supply chain, stating that the B300 is restricted from sale in China and that partners must comply strictly with export rules.
The company has incorporated export compliance features into its chips, though the extent to which these can be defeated or circumvented has not been publicly established.
The B300 scarcity compounds a separate restriction that hit Chinese AI companies in early 2025. On 9 April 2025, the US government informed Nvidia that it required a licence to export its H20 chips, the China-specific GPU Nvidia had designed to comply with earlier export control rules, to the Chinese market.
NVIDIA took a $4.5 billion charge in Q1 FY2026 on excess H20 inventory and purchase obligations, and was unable to ship an additional $2.5 billion of H20 revenue that quarter.
The H20 had been Chinese companies’ primary access point for compliant Nvidia GPU hardware; its restriction eliminated that option.
The combined effect, H20 restricted, B300 grey market disrupted, domestic Chinese AI accelerators not yet competitive with Nvidia’s highest-end systems for frontier model training, has produced the scarcity premium now reflected in the B300 price.
Chinese tech giants are reported to be particularly focused on B300 hardware for inference workloads, generating AI outputs at scale, where per-token cost is the defining competitive metric.
The current price dynamic is unfolding against an escalating policy backdrop. Last week, the US House Foreign Affairs Committee advanced the MATCH Act on 22 April in what lawmakers described as the largest semiconductor export control markup in congressional history.
The bill would require the Netherlands and Japan to align their own chip equipment restrictions with US rules, cutting off ASML’s remaining China sales and banning the servicing of already-installed machines. If enacted, it would further constrain China’s ability to manufacture advanced chips domestically and deepen its dependence on restricted foreign hardware.
China’s Ministry of Commerce has warned the legislation would ‘severely disrupt’ global supply chains.
For Chinese AI companies, the message embedded in the B300 price is simple: advanced AI compute will only get more expensive and harder to obtain through any channel.
The $1 million price point is not a market equilibrium. It is the cost of urgency in a market where the alternative, falling behind on AI inference economics, is considered more expensive still.
