Chinese AI startup Moonshot is in early-stage negotiations with Microsoft Azure, Amazon Web Services, and Google Cloud to host its flagship Kimi K3 model, seeking up to a 30 percent share of revenue from services built on the technology. If successful, the agreement would represent the first major revenue-sharing partnership between a Chinese AI company and a major U.S. cloud provider, signaling a potential shift in how international AI companies collaborate despite rising geopolitical tensions.

Why Is Moonshot Pursuing These Cloud Partnerships?

Moonshot, founded in 2023 by Carnegie Mellon-trained AI researcher Yang Zhilin, has emerged as one of China's most ambitious AI startups. The company raised more than $2 billion in May and is preparing for a potential Hong Kong listing. By partnering with major U.S. cloud providers, Moonshot would gain access to the infrastructure needed to serve global customers while maintaining control over its model's intellectual property and revenue streams.

The negotiations come at a critical moment for Chinese AI companies. While models from China are often cheaper than Western alternatives, they face significant barriers to international expansion, including U.S. export restrictions on advanced AI chips and ongoing scrutiny from American officials. Hosting Kimi K3 on Azure, AWS, or Google Cloud would allow Moonshot to reach U.S. and international customers without requiring them to download and run the massive model themselves.

What Are the Key Sticking Points in These Negotiations?

The discussions remain in early stages, and sources familiar with the talks identified several unresolved issues that could determine whether agreements are reached. These include how revenue would be divided between Moonshot and the cloud providers, what access each party would have to customer data, and how to audit the use of tokens, which are units of text processed by AI models and used to calculate revenue under usage-based billing.

Kimi K3 is an open-weight model, meaning it can be downloaded and modified by users. However, analysts note that few customers are likely to operate a system with 2.8 trillion parameters on their own infrastructure because of the prohibitively high computing costs involved. This reality makes cloud hosting partnerships essential for Moonshot's business model.

How Do Revenue-Sharing Agreements Work in AI?

• Model Hosting: Cloud providers host the AI model on their servers, handling infrastructure costs and maintenance while the AI company retains ownership of the underlying technology.
• Usage-Based Billing: Customers pay based on how many tokens they process, with revenue split between the cloud provider and the AI company according to negotiated terms.
• Data Access and Auditing: Agreements must specify what customer data each party can access and how usage will be monitored to ensure accurate revenue calculation and prevent misuse.
• Precedent Setting: Moonshot has already signed similar agreements with smaller cloud platforms, and the terms it is seeking align with those offered to major customers using its open-weight model.

Moonshot has already demonstrated success with this model. Chinese IT services provider Chinasoft International announced in July that it had a revenue-sharing agreement with Moonshot, though the company did not disclose the specific revenue split.

What Challenges Does Moonshot Face From U.S. Officials?

The startup's path to partnership with major U.S. cloud providers is complicated by significant political and regulatory headwinds. U.S. Treasury Secretary Scott Bessent said last month that he might add Moonshot to a trade blacklist, citing concerns about the company's practices. U.S. officials have also accused Moonshot of stealing from Anthropic's Fable model to help create Kimi K3 and of illegally acquiring Nvidia chips, allegations the company has rejected.

Moonshot has pushed back against these claims, telling China's National Business Daily that Kimi K3's performance gains resulted from original changes to the model's underlying architecture, not from distillation or model theft. The company's ability to secure partnerships with Microsoft, Amazon, and Google will likely depend on how these regulatory concerns are resolved and whether U.S. officials view such partnerships as a national security risk.

The negotiations represent a critical test of whether Chinese and American AI companies can find common ground despite escalating tensions over artificial intelligence, data security, and chip exports. If Moonshot succeeds in securing even one major partnership, it could open the door for other Chinese AI startups to pursue similar arrangements, fundamentally reshaping the global AI market landscape.