Chinese AI startups are posting triple-digit revenue growth yet still falling short of the commercial trajectories their valuations imply, pushing at least one lab toward revenue-sharing arrangements with the same US hyperscalers that compete with them on models.
Why it matters
MiniMax posted first-half 2026 revenue of approximately $116.6 million, a 283% year-over-year increase1. Moonshot AI is in early talks over revenue-sharing agreements with Microsoft, Amazon, and Google to host its Kimi K3 model, seeking up to a 30% share of K3-related service revenues, according to Reuters3,4. ANALYSIS The two developments trace the same underlying pressure: Chinese labs can acquire users and enterprise customers at speed, but translating that into self-sustaining economics remains a different problem, and the emerging answer is to borrow distribution from Western cloud platforms and split the proceeds.
The big picture
MiniMax posted first-half 2026 revenue of approximately $116.6 million, a 283% year-over-year increase powered by a 700% jump in its enterprise business. That growth rate accelerated from 2025's 159% year-over-year pace2. Yet the $116.6 million represented roughly 32% of the $363.77 million analysts expect for the full year, according to estimates compiled by Bloomberg. ◆ Hitting the consensus target would require the second half to deliver more than twice the first half's revenue, a back-loaded trajectory that leaves little margin for a slowdown.
On the model side, MiniMax's M3 trailed rivals such as Z.AI Co.'s GLM-5.2 in the race to release top-tier large language models. ◆ Revenue momentum and model competitiveness are diverging: enterprise demand for lower-cost AI services is driving the top line even as the flagship model falls behind the frontier.
Moonshot AI, meanwhile, is in early talks over revenue-sharing agreements with Microsoft, Amazon, and Google to host its Kimi K3 model, seeking up to a 30% share of K3-related service revenues, according to Reuters. Newsquawk characterized the 30% ask as "an opening position rather than a term sheet," noting that "comparable cloud-model arrangements have historically settled with the platform side retaining the larger economics in exchange for compute commitments, distribution, or both".
ANALYSIS Moonshot's decision to negotiate with all three US cloud giants simultaneously is structurally telling. Newsquawk observed that "parallel talks of this sort typically function as an auction dynamic that firms up pricing for the model developer and often ends with one anchor partner". Running a three-way process maximizes Moonshot's leverage on the revenue split, but it also signals that the company cannot afford to build its own global distribution channel.
The two stories illuminate different stages of the same monetization challenge. MiniMax is further along commercially, with real enterprise revenue, yet its growth still trails the forecast curve that underpins its valuation. Moonshot is earlier in the cycle, attempting to convert a competitive model into recurring revenue by plugging into platforms that already have paying customers. Both paths depend on external capital or external distribution rather than organic cash generation.
Newsquawk flagged that for the US platforms, "incremental model catalogue additions move their equities only at the margin unless tied to capex or exclusivity; the more sensitive leg has tended to be the model developer's funding and valuation narrative". ANALYSIS That asymmetry matters: the hyperscalers gain optionality by adding another model to their catalogues, while Moonshot's entire commercial trajectory may hinge on the terms it secures.
What's next
The near-term markers are concrete. MiniMax needs a materially stronger second half to close the gap to the $363.77 million full-year consensus. For Moonshot, Newsquawk identified the next catalyst: "Confirmation, denial, or a single-partner leak is the near-term follow-on, alongside any disclosure of compute or hosting terms attached to the revenue split". ◆ Whether the 30% ask survives negotiation, or settles closer to the platform-favorable norms Newsquawk described, will set a pricing reference for every Chinese lab that follows Moonshot onto US cloud infrastructure.