# Moonshot AI Wants Microsoft, Amazon and Google to Pay It a Cut of Kimi K3 > Source: > Published: 2026-08-26 22:58:18+00:00 *Moonshot AI wants a share of the money Microsoft, Amazon and Google make from hosting Kimi K3. The awkward part is simple: the same Chinese model Washington is watching could soon be sold through America's biggest cloud platforms.* Moonshot AI is trying to turn open weights into cloud revenue. According to Reuters, the Beijing startup is in talks with Microsoft, Amazon and Alphabet's Google over deals that would let Azure, Amazon Web Services and Google Cloud host Kimi K3, its 2.8 trillion-parameter model. Moonshot wants up to 30% of the revenue tied to K3 services on those platforms, Reuters reported on August 26, 2026, citing three people familiar with the talks. No deal has been signed. That matters here because the talks are still sitting on hard questions: the revenue split, access to data, and how token usage gets audited across three cloud systems that don't bill or meter everything in exactly the same way. Moonshot, Microsoft, Google and AWS all declined to comment to Reuters. Frankly, it's an awkward spot: these are companies that have spent two years telling investors and regulators they're racing China. Now they're weighing whether to help distribute one of its strongest models. ## The open model still needs a cloud bill Kimi K3 is why anyone is at the table. Moonshot's own GitHub materials describe it as a 2.8 trillion-parameter mixture-of-experts model with 104 billion active parameters, 896 experts, native vision and a 1,048,576-token context window. Reuters described it as open-weight, meaning customers can download and modify it. That sounds like freedom until you price the hardware needed to run a model of that size properly. [Moonshot's Bet on a Bigger Kimi K3 Is Paying Off in China's AI Race](https://startupfortune.com/moonshots-bet-on-a-bigger-kimi-k3-is-paying-off-in-chinas-ai-race/) Moonshot AI's 2.8 trillion parameter Kimi K3 topped a coding leaderboard ahead of Claude Fable 5 and GPT-5.6 Sol, then forced the company to pause new subscriptions as demand outran its GPUs. The launch also wiped hundreds of billions off Nvidia's market cap, reviving DeepSeek-style fears about AI infrastructure spending. - [Kimi K3 model China performance](https://startupfortune.com/moonshots-bet-on-a-bigger-kimi-k3-is-paying-off-in-chinas-ai-race/) - [Moonshot AI compute capacity constraints](https://startupfortune.com/moonshots-bet-on-a-bigger-kimi-k3-is-paying-off-in-chinas-ai-race/) Most companies won't run K3 on their own machines. They may like control, but they don't want the GPU bill, the infrastructure work, and the operational risk that come with serving a 3T-class model at scale. Reuters made the same point plainly: for large AI model providers, major cloud platforms are the main route into enterprise adoption. If you're Moonshot, giving away weights doesn't kill the business. It changes where the meter runs. The model has also earned enough attention to make the cloud companies listen. Arena.ai ranked Kimi K3 first in a benchmark for building web interfaces, while Reuters reported that Artificial Analysis found its performance comparable to OpenAI's GPT-5.5 and Anthropic's Claude Opus 4.8, especially on complex, multi-step tasks. Moonshot doesn't need every benchmark to say it has beaten the US frontier. It only needs enough customers to believe K3 is close enough and cheap enough to test. The market already got the message. On July 17, the Nasdaq fell about 1.4% and the S&P 500 dropped about 1%, while the Philadelphia Semiconductor Index lost 1.63%, according to Korean financial press reports that tied the selling to renewed concern over Chinese AI models and AI infrastructure spending. Nvidia fell 2.21% that day and briefly lost the top market-cap spot to Apple. One trading day doesn't prove a new AI order. It does show you how nervous the old one has become. ## The politics are built into the deal Moonshot has used this playbook before. Reuters reported that Chinese IT services provider Chinasoft International said in July that it had signed a revenue-sharing agreement with Moonshot, though it didn't disclose the split. A similar structure with Azure, AWS or Google Cloud would be a different event entirely. Chinasoft is a domestic channel. The US cloud giants are the enterprise internet's pipes. The company is also moving fast on capital. Bloomberg reported in July that Moonshot had told investors it was preparing for a Hong Kong listing in as early as six months, and separate Bloomberg reporting said the startup was seeking pre-IPO funding at a valuation of as much as $50 billion after a round that valued it around $31.5 billion. Its revenue story is no longer a side note: 36Kr reported that Moonshot's annual recurring revenue passed $100 million in March, $200 million in May and $300 million by mid-June. Those numbers need context. K3 launched in July, so the March-to-June jump wasn't caused by K3 alone. It does show that Moonshot already had a real paid business before the model that put it back in the center of the AI race. Alibaba's own annual report said it invested about $800 million for roughly 36% of Moonshot in fiscal 2024, a stake that now looks far more important than a passive bet on another chatbot company. Washington is the hard part. Reuters reported that US Treasury Secretary Scott Bessent said last month he might add Moonshot to a trade blacklist, and that US officials have accused the company of illegally acquiring Nvidia chips and distilling from Anthropic's Fable model. Moonshot has rejected the distillation claim, telling China's National Business Daily that K3's gains came from original architectural changes. [Moonshot AI Moves Toward a Hong Kong IPO After Kimi K3 Stuns Rivals](https://startupfortune.com/moonshot-ai-moves-toward-a-hong-kong-ipo-after-kimi-k3-stuns-rivals/) Moonshot AI is dismantling its offshore holding structure to prepare for a Hong Kong listing, days after its Kimi K3 model rattled global markets and closed the performance gap with OpenAI and Anthropic. The restructuring, reported by Bloomberg, sets up a possible filing as soon as the third quarter of 2026. - [AI startup Hong Kong IPO listing](https://startupfortune.com/moonshot-ai-moves-toward-a-hong-kong-ipo-after-kimi-k3-stuns-rivals/) - [Kimi K3 chatbot rival comparison](https://startupfortune.com/moonshot-ai-moves-toward-a-hong-kong-ipo-after-kimi-k3-stuns-rivals/) That's the real tension. If US officials are worried enough about Moonshot to talk about blacklisting it, then a cloud deal that sends K3 through Microsoft, Amazon or Google becomes more than a commercial negotiation. It becomes a test of whether the AI race is being governed by policy, platform demand, or whichever model customers want to run next. The talks may still collapse. Reuters said there is no certainty of an agreement, and that is the right note to end on. But the question is no longer whether Chinese open-weight models can get attention in the US. Kimi K3 already has it. The question is whether America's biggest cloud companies are willing to put it on the invoice. **Also read:** [IBM Releases Granite 4.2 Open Reasoning Models Free for Local Deployment](https://startupfortune.com/ibm-releases-granite-42-open-reasoning-models-free-for-local-deployment/) • [Nvidia Doubled Its Revenue to $96 Billion and Wall Street Shrugged Anyway](https://startupfortune.com/nvidia-doubled-its-revenue-to-96-billion-and-wall-street-shrugged-anyway/) • [OpenAI Says Its Own AI Agents Secretly Coordinated to Breach Hugging Face](https://startupfortune.com/openai-says-its-own-ai-agents-secretly-coordinated-to-breach-hugging-face/)