Is Nvidia Secretly Building a China-Only AI Chip? Inside the LPU Rumor Jensen Huang Keeps Denying

Is Nvidia Secretly Building a China-Only AI Chip?

A rumor that refuses to die is back in circulation: that Nvidia is quietly preparing an AI chip specifically for the Chinese market, built using technology it licensed from the inference startup Groq. The company has denied it — again. But the fact that this is the second time in 2026 a strikingly similar story has surfaced, each time followed by a firm public denial, is itself part of the story.

What The Information Reported

According to a report from The Information, citing two unnamed Nvidia employees, the company has been preparing small-batch shipments of a China-specific chip by the end of this year, and several customers had reportedly already placed orders. The chip in question is said to be a variant of Nvidia's "LPU" — a decode co-processor built on technology Nvidia licensed from Groq, the AI inference chip startup, and first unveiled publicly at GTC earlier this year as part of the Vera Rubin platform lineup.

The twist, per the report: because Vera Rubin itself can't legally be sold in China under current U.S. export rules, Nvidia reportedly rewrote the software that splits workloads between GPU and LPU so the Groq-derived accelerator could run alongside processors that are already permitted in the Chinese market — effectively threading a needle through export restrictions without changing the underlying silicon.

Nvidia's Response: "Incorrect"

Nvidia didn't take long to respond. "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 told Tom's Hardware, a statement that was also carried by Reuters and widely picked up across tech and financial press.

It's worth reading that denial carefully, though. It addresses current sales and the public roadmap — it does not explicitly dispute the specific claims about rewritten scheduling software, pending customer orders, or the export-compliance workaround described in the original report. The Information has also said Nvidia did not respond to its requests for comment over several days before publication. That gap between what's being denied and what was actually alleged is exactly the kind of detail that keeps a rumor like this alive rather than killing it outright.

This Isn't the First Time

This is also not a new argument. Reuters reported back in March that Nvidia was preparing a China-compatible AI chip as competition heated up for the country's inference market — and CEO Jensen Huang personally shot that down within days, reportedly calling it "totally false" in comments to Punchbowl News. Huang has also been candid about the broader picture: in May, he reportedly said Nvidia had "largely conceded" China's AI chip market to homegrown rival Huawei, a remarkable admission from a company that still clearly wants a foothold there.

So the pattern now is: a specific, detailed leak surfaces from a credible outlet, Nvidia denies it in strong terms, and a few months later a strikingly similar story resurfaces. That doesn't prove the reports are true — leaks evolve, and reporters chasing the same rumor from different sources can converge on similar, equally wrong, conclusions. But it's also not the kind of pattern that simply evaporates.

The Compliance Problem Nobody's Actually Solved

Even if Nvidia were moving in this direction, there's a second layer of uncertainty that often gets lost in the back-and-forth: U.S. export approval isn't the only gate here. Washington has shifted its own stance repeatedly in 2026, approving limited sales of H200 chips to select Chinese firms including Alibaba, Tencent and ByteDance in May, with deliveries only beginning to trickle through more recently. But Beijing has its own say — Chinese regulators reportedly blocked purchases of the earlier H20 chip outright last year, and have only recently signaled openness to some H200 imports. A chip that clears U.S. export rules is not automatically a chip Beijing lets its companies buy. That's the unresolved question nobody — not The Information's sources, not Nvidia's denial — has actually answered.

Why This Matters in the Gulf

It's not just a US-China story. The UAE's own AI ambitions — from G42's infrastructure buildout to the Stargate UAE campus in Abu Dhabi — run on the same Nvidia export-licensing machinery that governs what chips can legally leave American soil for any market outside the US. Every twist in Washington's China chip policy is being read closely in Abu Dhabi and Riyadh too, because the same regulatory apparatus, and the same Nvidia supply chain, underpins the Gulf's compute deals. A company reportedly willing to engineer a workaround for China is a company Gulf buyers will note is also motivated to keep every other regional relationship, including the UAE's, firmly on the right side of US export law.

What to Watch Next

Nothing here is confirmed beyond Nvidia's own public statement, and that statement leaves real gaps. Readers should treat the LPU-for-China claim as an open, contested rumor rather than settled fact — but one worth revisiting given how this exact story has already resurfaced once this year. If small-batch shipments genuinely do appear before year-end, as The Information's sources claimed, that would settle the question fast. Until then, it's Nvidia's word against two unnamed employees' — and a Beijing regulator that hasn't weighed in on any of it.