Nvidia Chips Reach China's AI Labs Through a Legal Rental Route

Chinese AI labs can legally train on Nvidia's most advanced chips today — not by buying them, but by renting them from data centers in Southeast Asia. That is the core finding of a Bloomberg Odd Lots segment that walks through an arrangement U.S. export rules simply do not cover, and it lands at a moment when Nvidia's stock and Washington's China policy are both under strain.

The stakes are straightforward. Nvidia traded at $160.56 on the morning of November 12, down 7.6% on the day and 29.1% over the past month, with the company's leadership in flux after its chief executive resigned. Behind the stock sits a policy question with far larger consequences: are the controls designed to keep advanced chips away from Chinese AI labs actually working?

What did John Chambers say?

Former Cisco Systems (NASDAQ:CSCO) chief executive John Chambers joined CNBC's Fast Money on November 11 and delivered a blunt verdict on U.S.-China technology ties: "The capability we give them and the trust factor is broken." He added a personal note about his own positioning: "But when the current leadership came in, I pulled out of China."

Chambers did not call for a blanket decoupling. For very advanced technologies that matter to defense, he said, "we have to be very selective," and he described himself as conservative on where to draw lines. The remark matters because export enforcement depends on exactly the trust Chambers says is gone: Washington has to trust that chips sold to allies stay put, and that cloud operators renting them out know who their customers are.

In plain words: a veteran tech CEO says the foundation of the entire control regime — trust — has cracked.

How does the rental loophole work?

Odd Lots laid out the mechanism with a concrete example. A company builds data centers in Malaysia and leases that capacity to DeepSeek and other leading Chinese AI companies. As far as anyone knows, the show said, "there is nothing per se illegal about that arrangement."

The reason is jurisdictional. Chips sold into Southeast Asia for use by Chinese customers sit outside the current rules. "There are no restrictions on selling chips into Southeast Asia to be used to train the most advanced open source models from China," the episode stated. "There's actually nothing illegal about it."

The show distinguished this from the Megaspeed case, where U.S. scrutiny has bitten. Remote rentals, the hosts said, are "extraordinarily common across the Southeast Asian data center ecosystem." What drew official attention there was "the suspicion that the company has not just rented compute to Chinese offtakers, but actually diverted the physical hardware" into mainland China. Renting the compute from abroad appears legal; moving the silicon across the border does not.

There was, briefly, a rule aimed at this gap. The Biden administration sought to limit such access through the AI Diffusion rule, which was set to take effect in May 2025. The Trump administration is not applying it.

Takeaway: the loophole is not an oversight being exploited in the shadows — it is an open channel the current administration has declined to close.

Why does Chinese demand flow to Malaysia?

"Offtake" refers to committed buyers who agree ahead of time to purchase a data center's capacity. According to Odd Lots, that committed Chinese demand is what physically built the region's infrastructure: Nvidia chips are "filling all these data centers in Southeast Asia that in many cases wouldn't exist if not for the offtake demand from Chinese companies."

The show went further on geography, calling Johor, Malaysia — the state adjacent to Singapore — in some ways "the backbone of Chinese compute." That is a striking label for a Malaysian state, and it reframes what export-control maps show: the choke point is no longer only the Chinese border.

Domestic supply cannot replace this demand yet. Bloomberg reported, via Odd Lots, that "Huawei's overall production capacity of its best AI chips is in the low hundreds of thousands," and that Huawei senior executives say "it'll take until 2030 to satiate domestic demand." Until Chinese fabrication catches up, Southeast Asian rental capacity remains the path of least resistance — and the legality is not in question.

Takeaway: Chinese AI labs need compute now, Huawei cannot supply it at scale until the end of the decade, and Malaysian data centers fill the gap.

What do Nvidia's own filings show?

Nvidia's numbers expose the tension between its China rhetoric and its regional growth. In its Q2 fiscal 2027 earnings release, the company guided third-quarter revenue to $108 billion, plus or minus 2%, explicitly excluding any Data Center compute revenue from China. On the August 26 earnings call, CFO Colette Kress said Hopper (Nvidia's prior-generation AI chip architecture) shipments to China-based customers made up "less than 1% of our total data center revenue."

Yet the broader business keeps compounding. Data Center revenue reached $89.02 billion in the quarter, up 117% year over year.

Nvidia is also formalizing a rental model of its own. Kress described a revenue-sharing arrangement with regional NeoClouds — third-party cloud operators built around Nvidia hardware — in which "we get paid twice: once on the hardware sale and again through the share of rental revenue." YTL AI Cloud in Malaysia is one of the named partners. Jensen Huang, CEO at the time of the call, said the platform "is fungible and durable and can be redeployed to support other customers."

Nothing on the call showed Nvidia equipment being leased into China. But regional ownership and third-party operation make it harder to see who ultimately ends up using the chips — the same visibility problem that defines the Megaspeed scrutiny.

Kress also said sovereign AI — government-backed AI infrastructure deals, mostly sold through regional NeoClouds — more than tripled year over year in Q2. That growth is the exposure investors should watch if the rules tighten.

Takeaway: Nvidia collects rent twice on hardware it cannot sell into China, through partners whose customers it does not fully control.

What happens next in Washington?

Two developments will decide whether the loophole survives. The first is regulatory: any move to close the remote-access rental channel, whether by reviving something like the AI Diffusion rule or adding licensing requirements for overseas cloud operators, would change who can lease Nvidia compute in Southeast Asia.

The second is disclosure. More regional detail in Nvidia's upcoming filings or on its next earnings call would show how exposed the tripled sovereign AI growth really is to a policy shift. Chambers's warning about broken trust now has a financial counterpart: the fastest-growing slice of Nvidia's business runs through the very arrangement Washington has left unregulated.

For now, the chips flow — legally — and the policy question remains open.