Nvidia has denied a report that it plans to ship a China-specific AI inference chip by the end of 2026. The company says the product, described in the report as a language processing unit built on technology licensed from Groq, is not on its roadmap.
Nvidia has denied a report that it plans to begin shipping a China-specific language processing unit by the end of 2026, saying the product is not on its roadmap. An Nvidia spokesperson told Reuters the company has no LPU sales in China and no China-specific LPU product planned.
The report Nvidia disputes
The denial followed a report by The Information that said Nvidia was preparing small-batch shipments for Chinese customers. According to that report, several customers had already placed orders for the chip, which it described as an inference processor built with technology licensed from Groq. The Information said the product was designed to work alongside graphics processing units and comply with U.S. export controls.
Nvidia's shifting position in China
Nvidia's China business has faced shifting export restrictions. Washington approved limited H200 sales to several Chinese companies in May. Yet CEO Jensen Huang has said Nvidia had largely conceded the country's AI chip market to Huawei.
A licensing deal built for a different purpose
Nvidia's LPU business rests on a licensing agreement with chip startup Groq worth roughly $20 billion, announced in late 2025. Nvidia's GPUs handle high-throughput training and batch inference, processing large amounts of data at once. The Groq-based LPU line targets a different job: low-latency, high-concurrency inference, the kind of workload where a user waits in real time for a response.
The Groq 3 LPU was introduced publicly at Nvidia's GTC 2026 event on March 16, 2026, as part of the company's Vera Rubin AI platform. Each chip carries 500 MB of on-chip SRAM and delivers 150 TB/s of bandwidth for inference tasks, and rack-scale deployments integrate 256 Groq 3 LPUs per rack. Nvidia has said the LPU is meant to work alongside its GPU lineup, not replace it, so GPUs still handle bulk training while LPUs manage interactive AI applications.
Source: Crypto Briefing
Trading involves risk.
Risk Disclaimer
