Key Moments
- Moonshot AI has been in talks with Microsoft, Amazon and Alphabet’s Google on revenue-sharing deals to host its Kimi K3 model on their cloud platforms, according to three sources.
- The Chinese startup is seeking up to a 30% share of revenue from K3-related services on Azure, AWS and Google Cloud, mirroring terms it uses with major open-weight model customers.
- Negotiations remain at an early stage, with key issues such as revenue allocation, data access, and auditing of token usage still unresolved, the sources said.
Talks Aim to Bring Kimi K3 to Major U.S. Cloud Platforms
China’s Moonshot AI is in negotiations with Microsoft, Amazon and Alphabet’s Google on revenue-sharing arrangements that would enable the U.S. cloud providers to offer hosting for its Kimi K3 model, according to three people familiar with the matter.
If completed, such agreements could represent the first significant revenue-sharing structure between a Chinese artificial intelligence developer and a large U.S.-based cloud firm.
The discussions underscore growing interest in leading Chinese AI models in the U.S. market, even as Washington’s national security concerns have prompted restrictions on exports of AI chips to China and as senior U.S. officials have publicly criticized Moonshot.
Commercial Terms and Early-Stage Negotiations
Sources said IPO-bound Moonshot is targeting as much as a 30% share of revenue produced from services tied to Kimi K3 on Microsoft’s Azure, Amazon Web Services and Google Cloud. That proposed level aligns with the revenue-sharing conditions the startup has set out for major users of its open-weight model, according to the same people.
The talks remain preliminary, and there is no assurance they will lead to finalized agreements, the sources added.
Moonshot did not respond to a request for comment on the potential partnerships. Microsoft, Google and AWS declined to comment.
Regulatory Scrutiny and Allegations
Moonshot has drawn criticism from U.S. Treasury Secretary Scott Bessent, who said last month that he might place the company on a trade blacklist. U.S. officials have accused the Beijing-based firm of improperly drawing on Anthropic’s most advanced model, Fable, in developing Kimi K3 and of illegally obtaining Nvidia chips.
Moonshot has pushed back against claims that Kimi K3’s performance relies on distillation. The company told China’s National Business Daily last month that its model’s improvements stem from original changes to its underlying architecture.
Key Issues: Revenue Split, Data and Token Auditing
One of the sources said several points remain unresolved in the discussions between Moonshot and the U.S. cloud providers. These include how revenue would be divided among the parties, the degree of data access to be granted, and the mechanisms for auditing token usage.
Tokens represent units of text processed by AI models, and tracking token consumption is central to determining fees and revenue under usage-based billing structures.
Performance Benchmarks and Model Scale
Kimi K3 has shown strong performance in third-party assessments. Arena.ai has placed it first in a benchmark that evaluates capabilities for building web interfaces. Artificial Analysis reports that Kimi K3 offers performance on par with OpenAI’s GPT-5.5 and Anthropic’s Claude Opus 4.8, particularly in evaluations of complex, multi-step tasks.
For large-scale AI model providers such as Moonshot, major cloud platforms are a primary channel for enterprise deployment.
Although Kimi K3 is an open-weight model that users can download and customize, analysts note that few customers are expected to operate a model with 2.8 trillion parameters solely on their own infrastructure because of the substantial computing requirements.
| Metric / Aspect | Detail |
|---|---|
| Model name | Kimi K3 |
| Parameter count | 2.8 trillion parameters |
| Model type | Open-weight, downloadable and modifiable |
| Revenue share sought | Up to 30% on K3-related services |
| Key third-party rankings | First in Arena.ai web interface benchmark; performance comparable to GPT-5.5 and Claude Opus 4.8 (Artificial Analysis) |
Existing Partnerships and Expansion Strategy
The sources said Moonshot has already entered into similar revenue-sharing arrangements with some smaller cloud platforms, though they did not disclose further information.
Separately, Chinese IT services provider Chinasoft International stated in July that it had reached a revenue-sharing agreement with Moonshot, without revealing the specific split.
Funding, Backers and Market Positioning
Moonshot was established in 2023 by AI researcher Yang Zhilin, who trained at Carnegie Mellon. The company is supported by several major Chinese technology firms, including cloud computing provider Alibaba.
The startup secured more than $2 billion in funding in May and is preparing for a potential listing in Hong Kong, according to sources.
Sources also said that Alibaba is pursuing revenue-sharing deals with significant users of its own new open-source AI model.
TradingPedia.com is a financial media specialized in providing daily news and education covering Forex, equities and commodities. Our academies for traders cover 
