Nvidia denies report of China-specific LPU shipments by year-end

Nvidia Rejects China LPU Report

Nvidia has officially denied a report from The Information suggesting it planned to ship a Language Processing Unit (LPU) specifically designed for Chinese customers by late 2026. An Nvidia spokesperson stated, "The reporting in The Information on NVIDIA's LPU is incorrect. We have no LPU sales in the China market today, and no China-specific LPU product in our roadmap."

Details of the Disputed Report

The Information's report, citing two Nvidia employees, claimed the chip was a variant of the Groq 3 LPU, announced at GTC in March. It suggested the silicon remained unchanged because it already complied with U.S. export regulations. The report indicated that software modifications would allow the LPU to function with processors available in China, despite the Vera Rubin platform, for which the LPU was originally designed, being restricted from sale in the country.

Previous Statements and Market Context

This denial follows earlier reports in March about Nvidia preparing LPUs for China, which CEO Jensen Huang then called "totally false." While Thursday's statement is more specific, addressing current sales and China-specific products, Nvidia has not clarified if its standard LPU will be available to Chinese buyers. Jensen Huang previously commented in May that Nvidia had "largely conceded" China's AI chip market to Huawei.

LPU Specifications and Export Compliance

The Groq 3 LPU is manufactured on Samsung's 4nm process, featuring 512MB of SRAM per die and no HBM. Nvidia announced at GTC that it would ship to customers like OpenAI in Q3 2026. U.S. export thresholds for China are based on compute density and bandwidth. An SRAM-only decode accelerator without an HBM stack is the type of component that can typically be exported under current regulations without requiring a cut-down SKU.