Moonshot wants a cut when US clouds sell its Kimi K3 model
Moonshot AI is in early talks with Microsoft, Amazon and Google over revenue-sharing deals that would put its flagship Kimi K3 on Azure, AWS and Google Cloud — and the Chinese lab is asking for up to 30% of the revenue those clouds generate from K3 services, according to Reuters. The talks are early and unresolved: the sticking points include how revenue would be split and whether Moonshot gets data access to audit token usage on its model. Nothing is signed.
The significance runs deeper than one licensing deal. If it lands, Kimi K3 would become the first Chinese open-weight frontier model distributed through all three major US cloud platforms — monetized inside the same hyperscalers that Washington has spent two years keeping Chinese AI chips out of. The talks are happening despite senior US officials publicly criticizing Moonshot over intellectual-property concerns, which tells you commercial gravity is currently beating geopolitical friction. For the clouds, carrying a cheap, strong open-weights model is simply what customers want; turning that demand into an exclusive-feeling distribution moat is worth paying Moonshot's cut for.
There's also a telling detail in the timing: Moonshot is IPO-bound, and a signed distribution deal with three US hyperscalers is exactly the kind of revenue visibility a listing prospectus wants. Compare that with DeepSeek, whose leaked books this week showed a fraction of Moonshot's likely economics — we covered the leak in DeepSeek's books leak: $70.7 million revenue, $106 million loss. Open weights were supposed to be a giveaway strategy; Moonshot is trying to prove they can be a tollbooth. (We've followed K3 closely — Kimi K3 escapes testing sandbox as AI designs first novel viruses — but this is a business story, not a safety one.)
Huawei pitches Egypt 2,000 AI chips while Washington scrambles to counter
Huawei has proposed building AI data centers for Egypt's government — used by its military for surveillance and other public services — built on more than 2,000 of its Ascend 950-series chips: 1,408 for cloud services plus roughly 600 for a separate system that controls models during training, Bloomberg reports, citing sources and documents reviewed by the outlet. It would be Huawei's first major export deal for advanced AI chips since US sanctions cut the company off from Western silicon, and it lands in a country strategically positioned between three continents.
The US response is the story within the story. The State Department has been lobbying Nvidia, AMD and Microsoft to submit a competing offer for the same project — tech diplomacy as procurement bidding. If Washington can't field an alternative, the default answer becomes Chinese infrastructure in a government AI system, complete with a military surveillance use case. Either outcome sets a template for the next dozen countries shopping for AI capability: every deal from here is a referendum on whether US exporters can match Huawei's price and willingness-to-deliver.
MiniMax's revenue nearly quadruples as its model lags — and that's the interesting part
Shanghai-based MiniMax reported H1 2026 revenue of $116.6 million, up 283% from $30.4 million a year earlier, per results flagged by Bloomberg. The engine is enterprise: its Open Platform business — where companies plug into MiniMax's models — grew 703% to $73.9 million and now accounts for 63% of revenue, while consumer products including Hailuo AI doubled to $42.6 million. International markets contributed about 61% of total revenue; gross profit rose 465% to $20.8 million; and the net loss narrowed 11% to $358 million.
Here's the tension: Bloomberg notes MiniMax's M3 model lags rivals like Z.ai's GLM-5.2, yet the money keeps compounding anyway. That's the clearest evidence yet that China's model-layer market rewards distribution and price, not just leaderboard position — MiniMax listed in Hong Kong in January, serves over a million enterprise developers across more than 230 countries, and is growing faster than it did last year with a mid-tier model. Benchmarks decide prestige; billing decides survival. MiniMax is betting on the second one.
What to watch: whether any of the three US clouds confirms or denies the Moonshot talks — and whether Washington's counter-bid for Egypt materializes before Huawei's does.
Chinese labs negotiating directly with US hyperscalers while their government-funded rival bids on foreign military data centers — is open weight becoming the new oil pipeline? Tell us in the comments.
Sources: Reuters · The Economic Times · Bloomberg · Sweden Herald (TT News Agency) · Bloomberg · SCMP · Crypto Briefing · Techmeme
