Moonshot AI wants 30% of what US clouds earn from Kimi K3

Moonshot AI is reportedly in early talks with Microsoft, Amazon, and Google to host its Kimi K3 model on their cloud platforms, potentially seeking up to 30% of the revenue those services generate. If agreed, it would represent a first-of-its-kind partnership between a Chinese AI company and major U.S. cloud providers, but details remain preliminary.

Analysis

This is less a direct earnings event than a proof-of-concept that hyperscalers can become distribution rails for third-party models. If the take rate is truly on hosted revenue, the clouds are monetizing both GPU utilization and customer lock-in, which is higher-quality than selling raw compute because it compounds with usage and switching costs. Near term the move is sentiment-positive for AMZN, GOOGL, and MSFT, but the financial impact is likely de minimis unless this becomes a repeatable marketplace template.

The second-order winner is the cloud platform with the deepest enterprise sales motion and marketplace infrastructure, because model hosting turns into an app-store economics debate. That shifts bargaining power away from standalone model developers and toward infrastructure owners that can bundle hosting, inference, billing, and compliance. The loser set is broader AI software names that depend on model differentiation; if clouds can host any credible model, the moat migrates from model quality to distribution and workflow integration.

The key risk is regulatory, not commercial: a China-linked model on U.S. hyperscaler infrastructure could be interrupted by export-control, security, or data-residency pushback in days, not months. Over 1-3 months, the catalyst is whether this is announced as a scalable platform arrangement versus a one-off pilot; over 6-18 months, the question is whether clouds standardize rev-share hosting for all external models. Consensus may be overstating the revenue and understating the strategic signal: this is probably more about fill rates and platform optionality than a new profit pool.

Key Decisions for Investors

• No aggressive chase on the headline; wait 1-2 trading sessions. If AMZN/GOOGL/MSFT pull back 2-4% on regulatory noise, buy a small basket long with a 1-3 month horizon and a 6-8% upside target, since the optionality is strategic rather than immediately monetizable.
• Relative-value idea: long AMZN vs short MSFT for 1-3 months. AWS has the cleanest marketplace DNA and the most direct incentive to monetize third-party model hosting; MSFT’s AI premium is already crowded, so this incremental catalyst is less incremental. Exit if Azure announces a comparable rev-share hosting program.
• If the market extrapolates this into a broad AI monetization breakout, fade the move by shorting AI software beta against long the hyperscalers. The thesis is that value accrues to distribution and infrastructure, not to every model-layer vendor. Cover if cloud guidance starts to reference higher AI-related attach rates or utilization.
• Set a hard alert on any U.S. regulatory statement or cloud-company clarification that frames this as a pilot or excludes China-linked workloads. That would falsify the structural thesis and compress the trade back to a one-day sentiment pop.