Moonshot AI scraps $150 million cloud deal as US export rules bite
Moonshot AI canceled a $150 million Indonesian cloud deal after US rules require licenses at a threshold of ~300 AI chips, while TSMC and Samsung halt shipments to China.

Why it matters
- Moonshot AI canceled a $150 million cloud contract in Indonesia after new BIS licensing rules, The Information reports.
- New rules require licenses at a compute threshold of roughly 300 AI chips, per Sidley Austin; Chinese labs need several thousand.
- TSMC and Samsung stopped certain chip shipments to Chinese customers, including ByteDance and Baidu's Kunlun, the Financial Times reports.
- The New York Times reports Alibaba secretly financed Alice Huang's Megaspeed, which bought over $4 billion in Nvidia chips.
- Five lenders may drop out of a $480 million loan for cloud provider Aolani, which was to supply Nvidia chips to ByteDance, per Bloomberg.
Moonshot AI, a Beijing-based AI startup, has canceled a $150 million contract with a Southeast Asian cloud provider in Indonesia after the US Commerce Department closed the last major loophole in its AI chip export controls. The cancellation, first reported by The Information, signals the immediate collapse of multimillion-dollar AI data center deals across Asia — the first concrete evidence that tightened US rules are severing Chinese labs from the compute they need to train frontier models.
What changed in Washington?
Days after the US Commerce Department shut the cloud loophole in the wake of the scandal around former Nvidia CEO Jensen Huang, the Bureau of Industry and Security (BIS) — the Commerce Department agency that regulates sensitive technologies and enforces export controls — now requires licenses for data centers renting compute capacity in third countries. Chinese firms had exploited exactly that route, renting processors in Singapore, Malaysia, and Indonesia to skirt US sanctions.
Existing operators have a 90-day transition period. During that window, and afterward, they must rigorously screen their customers for ties to China.
The licensing threshold is low. According to the law firm Sidley Austin, the requirement kicks in at a compute level equivalent to roughly 300 modern AI chips. Chinese labs typically need several thousand processors to train competitive models — a gap that makes the new rule effectively prohibitive.
Why did Moonshot walk away from $150 million?
The Information reports that Moonshot canceled the Indonesian contract and paused talks on additional deals in direct response to the new rules. The stakes were substantial.
Moonshot had planned to rent a large cluster of Nvidia Blackwell Ultra B300 chips starting next year to train its next language model, Kimi K4. Because of the heavy compute demand, the company had agreed to prepay about 50 percent of the contract value — far above the industry-standard 15 percent. That prepaid capital is now exposed as the deal collapses.
Access to Blackwell chips abroad was already critical for Moonshot's current model. Its predecessor, Kimi K3, matched top US models in July and triggered a BIS investigation, The Information also reported. A lab that reached frontier performance through rented overseas compute is now cut off from the same pipeline.
Are financial markets pulling back too?
Yes. Bernstein analysts led by Qingyuan Lin are warning of rising training costs for Chinese AI firms. Banks are retreating as well.
According to Bloomberg, five lenders are considering dropping out of a $480 million loan for Singapore-based cloud provider Aolani. That provider was supposed to supply Nvidia chips in Malaysia for TikTok parent ByteDance. The loan's potential unraveling shows the controls are now hitting financing channels, not just hardware logistics.
How are physical chip shipments affected?
The new rules are also cutting direct supply chains. The Financial Times reports that contract manufacturers TSMC and Samsung stopped certain processor shipments to Chinese customers in recent days while they apply for Commerce Department licenses.
The timing is brutal for Chinese firms. They are already struggling with historically high memory chip prices and capacity bottlenecks at domestic foundries such as SMIC. Huawei recently told its customers to source their own memory for its new AI chip — a sign of how tight the domestic supply picture has become.
The rules close a specific workaround. Chinese chip designers had used independent packaging and testing firms to order high-performance processors without direct US approval. That route is now blocked.
Samsung's affected customers include ByteDance and Baidu's AI data center unit Kunlun. The reach extends beyond AI accelerators: exemptions for non-AI chips have been eliminated, so TSMC and Samsung now need licenses even for advanced smartphone processors from manufacturers including Xiaomi, Oppo, and Vivo. Industry insiders say that is likely to cause significant production delays.
What did the Alibaba investigation reveal?
A New York Times investigation into the espionage scandal involving Jensen Huang's former mistress, Alice Huang, explains why Washington tightened controls so aggressively. The findings describe a covert procurement network with military research at its end point.
According to the Times, Alibaba acted as the secret primary customer and financier behind Alice Huang's Singapore-based firm Megaspeed, which bought more than $4 billion worth of Nvidia chips. Nvidia's sales team introduced Alibaba to Alice Huang in 2022.
Through its subsidiary Speedmatrix, Megaspeed operated cutting-edge Nvidia chips on behalf of Alibaba in a data center in Johor, Malaysia, built by conglomerate YTL — according to an analysis by security firm Infrawatch. Megaspeed reportedly sourced the servers from Aivres, a US subsidiary of state-linked Chinese firm Inspur, had them maintained by technicians from Shanghai, and routed data to China through fiber-optic and undersea cables via Singapore. Some chips were reportedly also smuggled physically.
The contractual arrangement gives Alibaba leverage. A contingency contract grants Alibaba direct access to all servers, chips, and intellectual property if Megaspeed breaches their agreement. YTL's stock price fell 20 percent amid the investigations.
How deep does the military connection run?
Alibaba also has a direct institutional link to Zhejiang Lab in Hangzhou, which it co-founded in 2017. The lab connects civilian and military research. It is run by former Alibaba CTO Wang Jian — a longtime friend of Jensen Huang who also advises the People's Liberation Army's National Defense University, an institution on the US sanctions list.
Researchers at the lab use Nvidia hardware to work on AI-driven cyberattacks and automated vulnerability discovery, according to the New York Times. They also collaborate with a Chinese air force academy on war-game simulations against foreign carrier strike groups.
Alibaba's infrastructure supported that work. Through the state-owned firm Zhejiang Computility and suppliers tied to the business network of Alice Huang and her husband Kang Wei, Alibaba built data centers with export-controlled Nvidia processors for the lab. The company officially denies any military use of its services.
What comes next?
The combination of halted shipments, collapsed cloud deals, and reluctant lenders leaves Chinese AI firms with fewer options at a moment when domestic capacity cannot absorb the demand. With BIS licensing thresholds set far below what frontier training requires, and investigations like the Times' exposing how previous workarounds funneled controlled chips toward military-adjacent research, the question now is whether Chinese labs can sustain frontier-model development on domestic supply alone — or whether training costs and delays, as Bernstein warns, will widen the gap.
Original: sidley.com
More from Marcus Bennett
Show full bio
Senior reporter covering companies, deals and strategy in artificial intelligence and machine learning at AI In Context.
791 articles