> ## Content Index > Fetch the complete content index at: https://bytevyte.com/llms.txt > Use this file to discover other available public pages before exploring further. # Kimi K3 revenue-sharing talks with US clouds hinge on audit - URL: https://bytevyte.com/kimi-k3-revenue-sharing-talks-with-us-clouds-hinge-on-audit/ - Published: 2026-08-27T14:11:53.000Z - Updated: 2026-08-27T14:11:53.000Z - Description: Moonshot is negotiating a Kimi K3 revenue-sharing deal with Microsoft, Amazon and Google to host the model on Azure, AWS and Google Cloud for US enterprises. - Author: Bytevyte Editorial - Tags: ai-beats Microsoft, Amazon and Google are in preliminary talks with Moonshot AI about selling access to Kimi K3 through Azure, AWS and Google Cloud. Kimi K3 is the Beijing startup's 2.8-trillion-parameter open-weight model. The proposed **Kimi K3 revenue-sharing** arrangement would give US enterprise customers a managed route to the model. Reuters reported on August 26, 2026, that Moonshot's proposal caps its share at 30% of the revenue the three clouds generate from Kimi K3 access. None of the four companies has confirmed a deal. Chinese AI firms have not previously signed a hosting-revenue agreement of this size with a US cloud provider. A deal would give Moonshot, valued at $35 billion as of July, a distribution channel into US accounts that its own infrastructure cannot serve. For the hyperscalers, the upside is a frontier-class model with proven demand, packaged as billable cloud usage rather than a downloadable artifact. Under the proposed split, Moonshot would collect the minority slice while the hosting platform keeps the larger share of gross service revenue before its own serving costs. That inversion is the point. In a conventional API model, the developer sets the price and keeps it. In a distribution model, the platform owns the customer relationship and the model owner gets a royalty-like fee. For Moonshot, 30% of someone else's hosting revenue is smaller than the retail price of direct API access, but it buys distribution that a Beijing lab cannot build alone. The structure resembles marketplace economics more than software licensing, with the platform paid for hosting, billing and customer acquisition and the model owner paid for the product itself. The deal would also invert the usual direction of AI trade flows. The dominant pattern so far has been US models moving outward through US clouds. This arrangement would put a Chinese model at the center of US infrastructure, sold by American companies to American customers. ## Inside Moonshot's Kimi K3 revenue-sharing proposal The structure under discussion is simple: Azure, AWS and Google Cloud would host Kimi K3 and sell access to it, and Moonshot would collect up to 30% of the resulting service revenue. Because the weights are public and any well-funded team can already run them directly, the deal is a distribution arrangement rather than a conventional API licensing play. The revenue share is, in effect, a fee for routing usage through an official, commercially licensed channel, and the 30% ask extends the pricing policy Moonshot already applies to its largest open-weight customers. Moonshot's contribution is licensing and support. The model's license is proprietary: organizations earning more than $20 million a year must negotiate a commercial license, while smaller teams can run the open weights at no licensing cost. That segmentation lets Moonshot capture grassroots adoption and enterprise money at the same time. For Moonshot, the threshold turns the open-weight release into a lead-generation mechanism: free adoption below the line, paid licensing above it. A cloud-hosted offering bundles managed inference, support and compliance into one purchase, which is exactly the packaging enterprises buy from hyperscalers. Kimi K3 is one of the largest open-weight models yet released. It arrived in July 2026 with 2.8 trillion parameters, and it has beaten rivals such as GPT-5.6 Sol and Claude Fable 5 on some benchmark tests. The model rattled markets when it landed, and Moonshot's July valuation of $35 billion reflects that momentum. | Kimi K3 at a glance | Detail | | ---------------------------- | ------------------------------------------------------- | | Parameters | 2.8 trillion | | Open-weight release | July 2026 | | License | Proprietary; commercial terms above $20M annual revenue | | Benchmark results | Beat GPT-5.6 Sol and Claude Fable 5 on some tests | | Moonshot valuation | $35 billion (July 2026) | | Revenue ask on cloud hosting | Up to 30% of cloud hosting revenue | ## Why the cloud giants are listening Enterprise demand for Kimi K3 already exists. Customers can download the weights and run them on their own infrastructure, so the question for Microsoft, Amazon and Google is where that demand converts into revenue. An official hosted version would capture that demand inside each platform, with Moonshot's license terms supplying the compliance wrapper that self-hosted deployments lack. Whichever cloud signs first also gets a differentiation point against rivals selling the same frontier models from US labs, and a supply line for open-weight workloads that does not depend on any single US model developer. For US customers, the practical change is in procurement. Instead of assembling the hardware and expertise to serve a 2.8-trillion-parameter model in-house, enterprises would buy Kimi K3 through the same cloud marketplace and billing relationship they already use, with support and licensing handled by the vendor. That lowers the barrier to running a frontier-class open model for organizations that could never justify the infrastructure cost of self-hosting. Moonshot's side of the bargain is partly about hardware. The company has already stopped accepting new subscribers because it could not secure enough GPU capacity, and a 2.8-trillion-parameter model is expensive to serve at scale. The pause showed how far demand had outrun supply and why Moonshot is willing to trade margin for capacity. The US clouds solve both problems in one deal: they bring inference capacity Moonshot cannot acquire quickly enough, and they bring the enterprise sales motion that turns model quality into recurring revenue. ## What is still on the table Negotiations are early, and the hard parts are unresolved. The revenue allocation behind the headline 30% figure, data access, and auditing of token usage are all still under discussion, and the talks could end without an agreement. None of the four companies has commented publicly on the negotiations. The token-auditing question is the one to watch. Moonshot's payment is a percentage of usage-based revenue, so both sides need a trusted method for counting tokens and verifying bills, and that mechanism does not yet exist. Usage-based payment aligns incentives, since both parties earn more when the model is used more, but it puts measurement at the center of the deal. Data access terms decide what Moonshot can observe about how its model is used inside US enterprises, a point with compliance implications for every party involved. The talks also proceed against a backdrop of US export restrictions that have limited Chinese AI firms' access to advanced American chips, which makes a Chinese frontier model sold through American clouds a politically sensitive arrangement. A completed deal would give Moonshot a direct monetization channel in the largest enterprise cloud market and set a template other Chinese open-weight labs could follow. If the talks fail, the open weights remain available for self-hosting, so the model's reach does not depend on this outcome. Only Moonshot's revenue is at stake. ## Why this matters The **Kimi K3 revenue-sharing** talks are the clearest sign yet that Chinese open-weight AI monetization is shifting from API licensing to distribution deals with Western infrastructure providers. For enterprises, the outcome decides whether frontier open models become a standard cloud product, bought through an existing marketplace, or remain a self-hosting project. Either way, the weights stay available; what is unresolved is whether Moonshot turns them into recurring revenue. *AI-generated image.* ## Related Articles - [Moonshot AI IPO Push Turns Kimi K3 Capacity Crunch Into a Market Structure Test](https://bytevyte.com/moonshot-ai-ipo-push-turns-kimi-k3-capacity-crunch-into-a-market-structure-test/) - [Moonshot AI's Kimi K3 Open Weights Reshape the Frontier Model Economics](https://bytevyte.com/moonshot-ais-kimi-k3-open-weights-reshape-the-frontier-model-economics/) - [Kimi K3 valuation impact: $314B cut from AI pre-IPO estimates](https://bytevyte.com/kimi-k3-valuation-impact-314b-cut-from-ai-pre-ipo-estimates/) ✔Human Verified --- *Researched and cross-referenced against primary sources by the Bytevyte editorial team. This article was generated with the assistance of artificial intelligence and reviewed by the Bytevyte editorial team.*