Moonshot AI Cancels $150 Million Nvidia Chip Leasing Contract

nashnova research
今天发布阅读约 8 分钟

Moonshot AI has canceled a $150 million contract to lease Nvidia B300 chips through an Indonesian data center, after new U.S. rules this week blocked Chinese AI firms from remotely accessing restricted chips overseas — a move that threatens to unravel the entire offshore computing strategy of China's AI industry.

01

Why did this deal fall apart?

Moonshot AI planned to lease a cluster of Nvidia Blackwell Ultra B300 chips via an Indonesian data center, starting as early as next year.
New U.S. Commerce Department rules now bar Chinese firms from remotely accessing restricted chips at overseas facilities. This means → even parking the chips in Southeast Asia no longer works.
To lock in the deal, Moonshot had already paid roughly 50% upfront — far above the industry norm of 15%. In plain terms = the money was already half-spent, and the contract is still dead.
02

How far does the new rule reach?

Overseas data centers serving Chinese clients above a certain computing threshold must obtain a U.S. license within 90 days.
That threshold is remarkably low — roughly 300 top-tier AI chips, according to law firm Sidley Austin. Chinese firms typically contract clusters of thousands.
This means → virtually every existing offshore computing deal involving Chinese AI companies will trigger the licensing requirement, not just Moonshot's.
03

Has the chain reaction already started?

Moonshot has paused all new contract negotiations while it assesses the full impact of the rules.
Separately, a $480 million financing deal for Singapore-based cloud firm Aolani — intended to purchase Nvidia chips for ByteDance's Malaysian data center — is stalling as five lenders flag compliance risk.
In plain terms = Moonshot is the first domino. Projects still under construction and not yet delivered face the highest cancellation risk.
04

How deep is the hit to Moonshot itself?

Moonshot's K3 model, released in July, approached the capability of leading U.S. frontier models. Its training relied on Blackwell chips sourced from outside China.
The company was actively seeking more compute for its next-generation model, Kimi K4 — the new rules cut off that supply line directly.
Shortly after K3's release, the U.S. Bureau of Industry and Security launched a formal investigation into Moonshot's advanced-chip procurement channels. This reflects a regulatory posture that goes beyond closing loopholes — it is tracing deals already completed.
05

What does this mean for China's AI industry?

Bernstein analyst Qingyuan Lin wrote this week that the new rules will materially raise training costs for Chinese AI firms.
This means → China's access to cutting-edge compute has shifted from "expensive but available" to "money alone may not be enough."
Whether any viable alternative path remains for offshore computing will be a defining variable in the next phase of the global model-development race.

市场有风险,内容仅供研究参考,不构成投资建议。