According to insights shared with financial outlet Cailianshe, Moonshot AI is currently in revenue-sharing discussions with Microsoft, Amazon, and Google around its Kimi K3 AI model. During these early-stage talks, the company is aiming to secure a cut of up to 30% of the revenue generated from K3-related services.
If a deal gets sealed, this would be the very first large-scale revenue-sharing agreement for a major AI model between a Chinese AI company and American cloud giants.
That said, everything is still very much in flux. Key details like how revenue gets calculated, the rollout timeline, and the boundaries of the service are all still up in the air. So far, all three cloud providers and Moonshot AI have declined to comment publicly on the talks.
Back in July, ChinaSoft International had already announced a similar revenue-sharing pact with Moonshot AI, though no specific percentage was ever disclosed.
If Moonshot lands that 30% ceiling, it would blow past market expectations—most insiders had figured 20% was the absolute max.
The core confidence behind pushing for a revenue cut comes down to how differentiated Kimi K3 truly is on the tech front. On July 17th of this year, Moonshot AI officially launched Kimi K3,witha whopping 2.8 trillion total parameters and roughly 104 billion activated parameters. That makes it the world’s first open-source 3T-level model, and it natively handles text, image, and video multimodal understanding.
Word on the street is that Kimi K3 had nearly 3,700 developers queued up to download it before launch. Within just 30 minutes of going live, it racked up over 4,000 likes on Hugging Face and shot straight to the top of the trend charts. Even Elon Musk left a comment on a related post saying “Impressive.” Demand was so overwhelming that compute resources ran short, and Kimi K3 even had to pause new consumer subscriptions for a bit.
Out in the global market, most general-purpose LLMs still struggle with long-context processing. Tons of overseas businesses are on the hunt for precise, efficient AI services tailored to their needs. Kimi K3 brings serious strengths to the table in ultra-long text handling, complex logical reasoning, multimodal adaptation, and Chinese-language optimization—which is exactly why those three cloud giants are genuinely enthusiastic about bringing Kimi K3 on board.
Beyond the big three cloud providers, Moonshot AI’s own developer hub already features an “Access Kimi models anywhere” entry point, and it’s made clear that K3 can be called via third-party inference partners. Public info shows Kimi K3 is already live on platforms like Together AI, Fireworks, DigitalOcean, Modal, Baseten, and DeepInfra.
Moonshot AI chose to open up the model weights, but the Kimi K3 license has some pretty specific terms aimed at large MaaS (Model-as-a-Service) platforms: if a company offers MaaS publicly and that company plus its affiliates pull in over $20 million in combined revenue over a rolling 12-month period, they must sign a separate agreement with Moonshot AI before using K3 for commercial purposes.
At the AWS China Summit, Huang Zhenxin, who heads up Moonshot’s Kimi enterprise business, revealed that Kimi’s API revenue has grown 400% and now accounts for over 70% of the company’s total revenue—and that number keeps climbing. On top of that, Kimi’s overseas paying users have also jumped 400%, with the product now available in more than 180 countries and regions.
Operating data shows that as of mid-June, Kimi’s annual recurring revenue (ARR) had blown past the $300 million mark. Back in March, ARR had only just crossed $100 million—so that’s a threefold surge in just three months.
Industry experts argue that regardless of how these talks ultimately shake out, the Kimi K3 overseas revenue-sharing negotiations are a landmark moment. It signals that Chinese-made LLMs have evolved from just shipping out tech and open-source contributions to a whole new phase of standardized commercial monetization and global value-sharing.
As domestic AI tech keeps evolving, we’re going to see more top-tier Chinese models heading overseas, carving out pricing power within the global AI supply chain, and pushing the whole industry toward a more competitive, mutually beneficial landscape.
OpenRouter data shows that as of July 26th, Chinese models held roughly 63.5% of the global share over the previous 28 days—about two-thirds of the worldwide market. Chinese AI models are also leading the pack in weekly call volumes,with the top performers being Xiaomi MiMo-V2.5, DeepSeek V4 Flash, Tencent HY3, Zhipu GLM-5.2, and DeepSeek V4 Pro.
