Analysis
Nvidia plans to begin shipping small volumes of a new AI chip designed for Chinese customers by the end of 2026, The Information reported Thursday, citing two employees. The report says the chip uses technology licensed from Groq, modified so it complies with current US export rules and works with processors used in China, and that certain Chinese customers have already placed orders. Nvidia disputed the report the same day, with a spokesperson saying: "We have no LPU sales in the China market today, and no China-specific LPU product in our roadmap." As of publication, that disagreement between The Information's sourced reporting and Nvidia's on-record denial remains unresolved -- it isn't yet possible to say definitively which account is accurate.
A separate, better-documented problem: the loophole
Independent of that dispute, CNBC reported that Chinese AI firms including Moonshot AI, ByteDance, Alibaba and Tencent have been accessing restricted Nvidia chip compute by routing through data centers in Thailand, Malaysia and Japan -- a loophole that exploits how current US export-control rules are written. The regime controls physical ownership of restricted chips; it does not currently cover remote access to that chip's compute power over the cloud, meaning a Chinese company can legally rent time on Nvidia hardware located in a third country without ever taking possession of the chips themselves. White House Office of Science and Technology Policy Director Michael Kratsios accused Moonshot AI in July of using this exact mechanism, accessing Nvidia GB300 chips through a facility in Thailand to train its Kimi K3 model.
“## Separating fact from inference What's verified: the Southeast Asia remote-access loophole is real and documented by multiple named officials and companies.”
What's being done about it
US legislators are discussing the proposed Remote Access Security Act, or RASA, which would bring cloud-based remote access to restricted chips and software under the same regulatory authority that currently governs physical chip exports. The bill faces real hurdles before it could take effect, and in the meantime the loophole remains open -- meaning restricted Nvidia compute is reaching Chinese AI labs training frontier models regardless of whether Nvidia itself ships a China-specific chip.
Separating fact from inference
What's verified: the Southeast Asia remote-access loophole is real and documented by multiple named officials and companies. What's disputed: whether Nvidia is actually building a China-specific LPU chip, which the company has explicitly denied on the record. Reading these two stories together, as some coverage has, risks implying Nvidia is deliberately working around US policy on two fronts at once, when only one of those two claims currently has Nvidia's own denial attached to it.
