Labs

Nvidia Denies Report
of China-Tailored LPU
as Small-Batch Shipments
Allegedly Loom

The Information said Thursday that Nvidia is preparing a Groq-licensed inference chip for Chinese customers, with orders placed and year-end shipments planned. Nvidia called the report 'incorrect' — and said no China-specific LPU is on its roadmap.

By Casey Vanderhoof Frontier labs correspondent · August 21, 2026

Nvidia on Thursday flatly denied a report from The Information that it’s preparing a China-tailored version of its Language Processing Unit, an inference chip built on technology licensed from Groq. “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,” a company spokesperson said, in a statement carried by Reuters and republished by Yahoo Finance and Investing.com.

The report the company is rejecting was unusually specific. Citing two Nvidia employees, The Information said Chinese customers had already placed orders for an LPU variant, that engineers had modified the chip’s software to pair with processors available inside China after the next-generation Vera Rubin system was blocked by U.S. export controls, and that small-batch shipments were planned by year-end. The publication also said the design complied with U.S. rules, while noting Beijing’s approval was an open question.

Investing.com, working from the same reporting, added the industrial texture: a tailored device would set up direct confrontation with domestic inference silicon, including Huawei’s upcoming Ascend 950DT. It also noted Amkor Technology packaged more than one million H200 units in the second quarter, and that Nvidia is separately developing two lower-power gaming chips for the Chinese market. Nvidia shares briefly turned positive on the news before slipping 0.5%.

This is the second time this year Nvidia has been publicly tied to a China-specific inference product. Reuters reported a similar story in March. The pattern matters because it maps onto an increasingly contested corridor. In May, Washington cleared H200 sales to a limited set of Chinese firms including Alibaba, Tencent, and ByteDance, with deliveries only recently beginning. That same month, CEO Jensen Huang told reporters Nvidia had “largely conceded” China’s AI chip market to Huawei, a striking framing from a company whose commercial center of gravity depends on not conceding anything.

On August 7, Bloomberg reported that a key U.S. agency is reviewing how Chinese AI firms acquire and access Nvidia chips overseas, following a spate of technological breakthroughs out of Chinese labs. So the denial lands inside a three-front squeeze: an export-control regime being tightened around offshore workarounds, a domestic Chinese competitor gaining silicon of its own, and a CEO who has already publicly acknowledged the strategic retreat.

Two Nvidia employees told a reporter one thing. The company’s press office says another. Both can be operationally true at once, and often are in semiconductors, where roadmap denials tend to age on a shorter clock than the roadmaps themselves.

Sources