Nvidia B300 servers are priced at $1 million in China, almost twice the cost in the United States.

      Prices have jumped from around 4 million yuan late last year to 7 million yuan ($1 million) today, mainly due to the arrest of the Supermicro co-founder, stricter enforcement measures, and a high demand from Chinese AI firms that are reluctant to hold restricted hardware on their balance sheets.

      NVIDIA’s B300 AI servers are currently being sold in China for about 7 million yuan, roughly $1 million per unit, which is nearly double their US list price of about $550,000, as reported by Reuters, citing four industry sources.

      The nearly 100% increase from approximately 4 million yuan late last year highlights a scarcity premium that has surged sharply since early 2026, as a crackdown on chip smuggling disrupted the grey market supply channel that had been essential for Chinese companies unable to legally purchase Nvidia hardware.

      Rental rates for B300 servers in China have also increased alongside sale prices, with short-term contracts reportedly reaching as high as 190,000 yuan per month, emphasizing the extreme scarcity of computing resources.

      Chinese technology firms are urgently seeking the hardware necessary to monetize their AI models and improve inference efficiency, particularly as the cost of generating tokens—the basic unit of AI text output—becomes a priority with the quickening pace of commercial AI deployment.

      Many companies are purposely structuring deals to keep Nvidia hardware off their balance sheets to avoid the potential ramifications of US sanctions.

      The catalyst for the price increase can be traced to the arrest of Yih-Shyan ‘Wally’ Liaw, co-founder of Supermicro, on March 19, 2026.

      US federal prosecutors unveiled an indictment accusing Liaw, Supermicro’s Taiwan general manager Ruei-Tsang Chang, and contractor Ting-Wei Sun of conspiring to divert about $2.5 billion worth of Supermicro servers, containing Nvidia's export-controlled Blackwell-class AI chips, to Chinese buyers through a Southeast Asian intermediary.

      The operation included altering serial numbers through heat guns, creating fraudulent paperwork, and utilizing fake replica servers to mislead government regulators. Liaw and Sun pleaded not guilty, while Chang remains at large.

      Supermicro’s stock plummeted 33% on the day the indictment was revealed, wiping out over $6 billion in market value, and Liaw promptly resigned from the board.

      This prosecution represents the most significant enforcement action under the US AI chip export control regime that has been tightening since 2022, and its impact on the grey market supply was immediate: the arrest alerted the network of brokers, logistics firms, and intermediaries that the risks of enforcement had increased noticeably, significantly lowering the volume of hardware being sold through unofficial channels.

      NVIDIA has publicly distanced itself from the grey market, asserting that the B300 is not for sale in China and that partners must adhere strictly to export regulations.

      The company has incorporated export compliance mechanisms into its chips, though the extent to which these can be bypassed has not been made public.

      The scarcity of the B300 adds to a separate restriction imposed on Chinese AI firms in early 2025. On April 9, 2025, the US government informed Nvidia that a license was needed to export its H20 chips, specifically designed for the Chinese market in compliance with earlier export restrictions.

      In Q1 FY2026, NVIDIA incurred a $4.5 billion charge for excess H20 inventory and purchase obligations and was unable to ship an additional $2.5 billion in H20 revenue that quarter.

      The H20 had served as the primary access point for compliant Nvidia GPU hardware for Chinese companies, and its restriction eliminated that avenue.

      The combination of restricted H20, disruption of the B300 grey market, and Chinese AI accelerators not yet matching the performance of Nvidia’s top-tier systems for advanced model training has resulted in the current scarcity premium reflected in the B300 price.

      Chinese tech giants are reportedly highly focused on B300 hardware for inference workloads, which involve generating AI outputs at scale, where the per-token cost is the critical competitive measure.

      The present price scenario is occurring against a backdrop of escalating policy changes. Last week, the US House Foreign Affairs Committee moved forward with the MATCH Act on April 22, regarded by lawmakers as the most significant semiconductor export control markup in congressional history.

      This bill would compel the Netherlands and Japan to align their chip equipment restrictions with US regulations, effectively stopping ASML's remaining sales to China and banning the servicing of machines that are already installed. If passed, it would further restrict China's ability to manufacture advanced chips domestically and heighten its reliance on limited foreign hardware.

      China's Ministry of Commerce has cautioned that the legislation could "severely disrupt" global supply chains.

      For Chinese AI companies, the message underlying the B300 price is clear: advanced AI computing will only become more expensive and harder to acquire through any means.

      The $1 million price tag does not reflect