Nvidia denies plans for China-specific LPU chip as Supermicro fires staff over alleged GPU smuggling

Two separate developments this week underscore the intensifying scrutiny around US export controls on advanced AI hardware to China, with Nvidia pushing back against claims it is preparing a new China-bound chip and Supermicro revealing the results of its own internal probe into alleged violations.

Nvidia's statement on Thursday directly contradicted a report from The Information, which cited two Nvidia employees and claimed the company had already received orders from Chinese customers for a custom LPU variant. The chip in question was described as a version of the Groq 3 LPU—a decode co-processor for the Vera Rubin platform—that is already compliant with US export rules because its silicon is unchanged. According to The Information, Nvidia's engineers rewrote software to allow the LPU to work alongside processors available in China, bypassing the fact that the full Vera Rubin system cannot be sold there.

This is not the first time Nvidia has denied such reports. In March 2026, after similar rumors surfaced, CEO Jensen Huang called the story 'totally false.' Thursday's statement is narrower in scope, addressing current and planned sales rather than future possibilities. Nvidia's spokesperson said the company has 'no LPU sales in the China market today' and no China-specific LPU on its roadmap, though it did not rule out future adaptations that comply with US law.

Meanwhile, Supermicro released the findings of its internal investigation into allegations that its executives participated in a scheme to ship $2.5 billion worth of GPU-laden servers to China in violation of export controls. The company stated that no current senior management had knowledge of the alleged diversion, and that it found no evidence of direct sales to restricted parties or that its financial statements were unreliable. However, Supermicro acknowledged that some employees failed to follow export policies and company code of conduct, leading to 'personnel actions' including terminations within sales, technical support, and business development. The three individuals named in the March DOJ indictment—co-founder Wally Liaw, Taiwan-based executive Steven Chang, and broker Willy Sun—'no longer have any relationship with Supermicro,' effectively confirming their firing. The board has also adopted recommendations to strengthen the company's export compliance program.

Both cases illustrate the high stakes of US-China technology competition. The US government has steadily tightened export controls on advanced AI chips since October 2022, restricting sales of Nvidia's A100, H100, and later H20 GPUs to China. Beijing has also blocked some imports, recently permitting H200 shipments to select companies like ByteDance and Tencent. Whether Beijing would approve an LPU-based product, even if Nvidia were to build one, remains unclear.

Why This Matters

Both stories directly affect the supply chain for AI hardware, which underpins the generative AI boom that companies worldwide are racing to exploit.
US export controls are a central tool in the technology competition with China; any perceived loophole—whether through adapted chips or diversion schemes—could undermine policy and trigger further restrictions.
Supermicro's firing spree and Nvidia's denial signal that companies are under immense legal and reputational pressure to demonstrate compliance, but also that the line between compliant and prohibited products remains contested.

Background

The US government first imposed export controls on advanced AI chips in October 2022, followed by a series of updates that have incrementally restricted the performance thresholds allowed for sales to China. Nvidia has created several China-compliant versions of its GPUs, including the A800, H800, and H20, but Chinese authorities have at times blocked even those. In March 2026, the DOJ indicted three individuals connected to Supermicro for allegedly routing GPU servers through intermediaries to Chinese end users. Supermicro promised an independent investigation, the results of which were released on August 21. Separately, rumors of a China-specific LPU have circulated since at least March 2026, when Jensen Huang first denied them.

Key Perspectives

Nvidia: The company maintains it has no current or planned China-specific LPU product, pushing back against the Information's reporting as incorrect. Nvidia argues its existing compliant chips serve the Chinese market within legal boundaries, and that any new product would have to meet US export rules. Supermicro: The company acknowledges compliance failures and has taken action, but insists no current senior executives were involved in the alleged diversion. It portrays the investigation as thorough and cooperative with authorities. US Government (DOJ/BIS): The indictment of Supermicro executives signals that the Biden (or subsequent) administration is actively prosecuting export control violations. The government likely views the internal firings as insufficient and may pursue further charges against other employees. Critics/Skeptics: Some observers argue that Nvidia's denial may be carefully worded to leave room for future adjustments, and that the LPU story reflects persistent demand from Chinese customers willing to pay premiums for any advanced AI hardware. They also note that Supermicro's history of governance issues—including previous accounting scandals and export probes—raises questions about the depth of its compliance reforms.

What to Watch

Whether Nvidia releases any future statement clarifying its roadmap for China-compliant accelerators beyond the H200.
The progress of the DOJ's case against the three indicted individuals and whether any additional Supermicro employees face charges.
Beijing's response: Chinese regulators previously blocked H20 imports; they could similarly block any new Nvidia products, regardless of US compliance.
Upcoming earnings calls: Nvidia and Supermicro will likely face analyst questions on their China exposure and compliance risk.

ZOTPAPER