China’s Moonshot AI has a bold number in mind: $2 billion in annualized revenue by the end of 2025. That would mean doubling what the company was generating as recently as August. According to TechCrunch, the target reflects the commercial momentum behind Kimi’s K3 model, which launched this summer and quickly drew serious usage. OpenRouter data currently shows around 300 billion tokens being generated daily by K3 models on the platform. That’s a real number. But it’s also where the clean part of this story ends.
K3 is an open-weight model, which means Moonshot’s margins are structurally lower than those of its closed-weight rivals. OpenAI is reportedly tracking toward $40 billion in annualized revenue. Anthropic sits around $65 billion. Moonshot is nowhere near that scale, and the open-weight model makes it harder to get there. But the trajectory still proves something the industry keeps debating: there’s a real business in open models, even if the ceiling is lower than what you get from locking down your weights and running everything through an API.
The more pressing issue right now isn’t the revenue gap. It’s the legal exposure. Earlier this week, Anthropic publicly accused Moonshot of running a long-term model distillation campaign that routed nearly 300,000 requests from Kimi directly to Claude Opus. The allegation is that Moonshot was essentially serving Claude responses to its own users, then collecting those outputs, over 23 million in total, to train its own models. That’s not a gray area. Distillation from a competitor’s model without authorization violates most API terms of service, and depending on jurisdiction, it could carry broader legal consequences.
This matters beyond the Moonshot-Anthropic dispute. Model distillation has been a known shortcut across the AI industry, but accusations at this scale and specificity are rare. If Anthropic pursues legal action, it could set a precedent that reshapes how open and closed model providers protect their outputs. Other frontier labs are almost certainly paying close attention.
For developers evaluating Kimi or K3 as part of their stack, the controversy creates real uncertainty. A model under legal challenge from a major competitor is a vendor risk, regardless of how competitive the benchmarks look. Moonshot has strong usage numbers and an aggressive growth plan. But building on top of a model whose training provenance is being disputed by Anthropic is a risk worth pricing in before committing to it.
