Skip to content
Advertisement
Moonshot AI reportedly ends $150 million deal for Nvidia Blackwell chips in Indonesia

Moonshot AI reportedly ends $150 million deal for Nvidia Blackwell chips in Indonesia

Photo: Pixabay / Pexels

The Kimi developer's reported exit from a B300 access arrangement adds to a growing list of questions about how it gets its compute

Moonshot AI, the Beijing-based developer behind the Kimi family of large language models, has reportedly ended a $150 million deal to access Nvidia Blackwell Ultra B300 chips in Indonesia.

The transaction has not been publicly confirmed. It still lands at a delicate moment for a company whose access to top-tier Nvidia hardware has drawn sustained attention from Washington.

What reportedly happened

The arrangement centered on Nvidia’s Blackwell Ultra B300, one of the most advanced GPUs the chipmaker sells. Moonshot was not buying the chips outright. The deal covered access to B300 capacity hosted in Indonesia.

Blackwell-generation GPUs fall under strict US export controls. Selling them to Chinese companies without a proper license is prohibited. Remote access through overseas data centers has become one of the gray areas regulators keep circling.

No reason for the reported termination has been made public. Moonshot has stayed quiet on many of the allegations tied to its chip access, and it has not commented on this one either.

A pattern of scrutiny

The Indonesia report does not exist in a vacuum. It follows a string of episodes that have placed Moonshot near the center of the US-China fight over AI hardware.

Advertisement

In July 2026, a White House official accused Moonshot of accessing Blackwell GPUs in Thailand in violation of US export restrictions. That same month, the company released Kimi K3, its latest model.

Then came August. Financing documents valued at approximately $65 million were reported for servers housing B300 GPUs located in Qingyang, China.

Nvidia pushed back on the link. The chipmaker stated that customers named in the prior financing documents disproved an association with Moonshot.

By November 2026, Moonshot was reportedly in talks with European cloud operators about clusters of more than 8,000 Blackwell GPUs. Those clusters were described as being focused primarily on inference, the work of running a trained model rather than building one.

Nvidia’s balancing act

Nvidia sits in an awkward spot. It wants to sell as many GPUs as possible, and it also has to stay on the right side of US export rules.

The company’s guidance reportedly requires that more than 51% of new GPU capacity be reserved for clients outside of China. The idea is to keep the majority of fresh Blackwell supply flowing to customers that do not trigger export concerns.

When capacity is leased through intermediaries in Southeast Asia or Europe, tracing who is actually running workloads becomes far harder.

Why Moonshot keeps attracting attention

Moonshot is not a small player hiding in the margins. Its valuation reportedly reached up to approximately $35 billion by mid-2026, a sign of how much investor money is riding on its Kimi models.

The company has kept up a rapid release cadence. Kimi systems are positioning themselves as serious competitors in global AI markets, not just inside China.

That ambition creates a problem with no easy fix. Competing at the frontier requires frontier hardware, and the most capable hardware is precisely what US policy aims to restrict.

What this means

For Moonshot, walking away from a $150 million compute arrangement, if confirmed, could cut either way. It may reflect a decision to reduce regulatory exposure after months of scrutiny. It could also simply mean the company found capacity elsewhere, such as the European clusters it was reportedly discussing.

For Nvidia, the episode underscores how much of its compliance burden now sits downstream. Selling to a cloud provider is one thing. Knowing who rents that provider’s capacity six months later is another.

Disclosure: This article was edited by Diego Almada Lopez. For more information on how we create and review content, see our Editorial Policy.