Key points
- Moonshot AI drops a $150 million Nvidia chip deal
- A US rule covering Nvidia chips in Indonesia took effect Nov. 19
- Moonshot AI has looked abroad for Nvidia chips before
- ByteDance's chip loan stalls as its Azure talks advance
Moonshot AI, the Beijing company behind the Kimi models, has scrapped a $150 million deal to access Nvidia (NVDA) chips through a data center in Indonesia, The Information reported on Nov. 19. Moonshot canceled the deal after the US announced a rule this week that restricts Chinese companies' remote access to advanced chips, according to the report.
The contract was with a Southeast Asian neocloud, a cloud company that specializes in renting out AI chips. It covered a set of Nvidia's Blackwell Ultra B300 chips, with service planned to begin as early as next year, according to ChainCatcher's summary of the report. The summaries we read don't name the provider or say how many chips the deal covered. We found no public comment from Moonshot or Nvidia on the report.
Nvidia shares were at $152.06 at 10:54 a.m. ET on Nov. 19, up 0.7% from the Nov. 18 close of $150.96.
Indonesia is one of five countries named in the rule
The Commerce Department's Bureau of Industry and Security (BIS) released the rule on Nov. 16. It was published in the Federal Register on Nov. 19 and took effect the same day.
The rule requires a license to export or reexport advanced computing items to Indonesia, Malaysia, Singapore, Thailand, and Brazil, or to transfer them within those countries. The covered items include chips classified under ECCN 3A090 (except 3A090.c), the category that includes Nvidia's data-center AI chips, plus 4A090 computers that contain them and related software and technology. BIS says those countries "present a higher-than-normal risk of diversion."
The five-country requirement doesn't apply when the ultimate consignee or end user is headquartered in the US, or its ultimate parent company is. To use that exclusion, the exporter, reexporter, or transferor must file the reports required under Section 743.10 of the export rules.
Separately, the rule can require a license for advanced chips destined for data centers anywhere in the world that give companies headquartered in Macau or in Country Group D:5, which includes China, access to more than 12 million "aggregate TPP" of computing power. That equals the combined processing power of about 758 of Nvidia's H100 chips, as our guide to the remote-access rule shows.
Data centers that were providing a "significant volume of compute power" to those companies before the rule was published and want to keep doing so must get a license within 90 days of the effective date, which works out to Feb. 17, 2027, or "face future export restrictions," the rule says. That clause doesn't give a number. The rule's summary uses the same "significant volume" wording to describe the 12 million TPP license requirement. Shipments already en route on Nov. 19 can go ahead under the old terms if they're completed by Dec. 21, 2026. Comments on the rule are due by Jan. 19, 2027.
Moonshot has sought Nvidia chips outside China before
Moonshot discussed renting clusters of 8,000 or more Blackwell chips from neoclouds in Europe this year, CNBC reported on Nov. 5. One proposal had Moonshot paying 30% to 50% more upfront than a non-Chinese customer would. CNBC didn't name a neocloud or report a signed deal.
In July, White House science director Michael Kratsios said Moonshot "acquired GB300-equipped servers and has accessed GB300s in Thailand, likely to train its AI models." China's embassy in Washington rejected the claim. Thailand is also one of the five countries in the new rule.
ByteDance's chip loan stalled and its Azure talks advanced
Lenders paused work on a $480 million loan for Nvidia chips in a Malaysian data center that ByteDance would use, Bloomberg reported on Nov. 17. Bankers at five lenders worried the loan could violate the rule, according to that report. The borrower, Singapore-based Aolani Cloud, said it was working with regulators and its financing partners to comply.
A day later, The Information reported that ByteDance is in advanced talks to spend about $500 million more on Microsoft (MSFT) cloud services, mainly for access to OpenAI models through Azure. That report doesn't say the rule applies to the deal. The rule's main test is access to computing power, and it doesn't say how a hosted model service would count toward the 12 million TPP limit.












