Nvidia on Thursday pushed back on a report that it was preparing a China-specific LPU for shipment by year-end, saying the company has no such product planned or currently sold in the market.
The company rejected a report that it could begin small-batch shipments by year-end. The reported chip would use technology licensed from Groq to accelerate AI inference alongside GPUs.
Nvidia faces a long road back in China’s AI market
To understand why this story matters, it helps to remember what China once represented for Nvidia. The country accounted for roughly 20% of the company’s data center revenue before US export restrictions began tightening in earnest.
CEO Jensen Huang acknowledged in May 2026 that Nvidia had largely ceded the Chinese AI chip market to Huawei. Huawei, backed by substantial government support and increasingly capable domestic chip designs, has moved into the vacuum that export restrictions created.
The US government has approved limited shipments of its H200 processors to select Chinese companies, with ByteDance and Tencent each receiving approximately 10,000 units. The Vera Rubin architecture, Nvidia’s next-generation platform, has already been restricted from Chinese sale.
Reading between the lines
The Groq licensing angle in the original report is worth noting separately. Groq’s LPU architecture is specifically built for inference rather than training, meaning it’s designed to run AI models quickly rather than build them from scratch. Whether Nvidia ever seriously explored that avenue, and whether that exploration is now dead or simply undisclosed, is the question the denial doesn’t fully resolve.
For investors, the denial itself is the signal. Nvidia is not announcing a China re-entry strategy. It is not teasing a compliant product pipeline. It is saying, clearly, that it does not have a product for this market. Any of those readings suggests the 20% of data center revenue that China once represented should not be modeled as recoverable in the near term.