Moonshot AI targets $2 billion revenue run rate as open-weight K3 gains traction

Moonshot AI, the Chinese company behind the Kimi assistant, is targeting a $2 billion annualized revenue run rate by the end of 2026, according to reporting by Bloomberg cited by TechCrunch. The goal would represent roughly twice the company’s reported August run rate and reflects the commercial momentum the company says it has built around its K3 model.

The target puts a major financial benchmark on an open-weight AI strategy. Moonshot makes K3 model weights freely available, a distribution approach that can accelerate adoption but generally gives providers less control over pricing, hosting and margins than closed models. At the same time, the company’s growth ambitions are emerging alongside allegations from Anthropic about how Moonshot obtained training data—an issue that could create legal and reputational risks for the business.

K3 turns open weights into a revenue test

K3 was released this summer and has become the center of Moonshot’s commercial plans. TechCrunch, citing OpenRouter data, reported that K3 models were generating as many as 300 billion tokens per day on the platform. The same report said usage had declined slightly in recent months, suggesting that high activity has not necessarily translated into continuously accelerating demand.

OpenRouter is an important distribution and measurement layer for developers comparing and routing requests across AI models. Its figures provide an indication of model usage on that system, but they do not establish Moonshot’s total global usage, revenue or profitability. The available reporting also does not specify how much of Moonshot’s projected revenue would come from hosted API access, enterprise agreements, model licensing or other products connected to Kimi.

The $2 billion figure is therefore best understood as a company target or reported internal projection, not a confirmed financial result. TechCrunch said the target was reported by Bloomberg and that Moonshot’s August revenue run rate was approximately half that level. Neither the supplied evidence nor the linked wire items provides audited financial statements or a detailed revenue breakdown.

Open-weight economics are central to the bet

Moonshot’s strategy illustrates both the opportunity and the pressure facing open-weight model developers. Releasing weights can attract researchers, application developers and infrastructure providers without requiring every user to rely on a single company-hosted endpoint. That broader availability can increase experimentation and create downstream demand for hosted services, fine-tuning, support and enterprise deployment.

The trade-off is that freely available weights can make price competition more intense. Developers may run K3 on their own infrastructure, use competing inference providers or modify the model without paying Moonshot directly. TechCrunch noted that Moonshot’s margins are likely to be lower than those of closed-weight competitors because the company does not retain the same level of control over access.

That makes the reported target significant beyond Moonshot itself. If achieved, it would suggest that an open-weight lab can convert large-scale model usage into substantial commercial revenue even without the economics of a fully closed platform. It would not, however, demonstrate that the model is as profitable as leading closed-model businesses.

TechCrunch compared the projected revenue with figures recently reported for OpenAI and Anthropic, which were described as roughly $40 billion and $65 billion, respectively. Those comparisons are not like-for-like financial disclosures, and the evidence does not establish the same accounting basis for each company. They do show the scale gap Moonshot would still face in the global AI market.

Adoption signals come with important limits

The strongest usage signal in the available reporting comes from OpenRouter, which was cited as recording up to 300 billion daily tokens for K3 models. That is a platform-specific measurement rather than a verified statement of paying customers, active enterprise deployments or recurring revenue. The report also acknowledged a recent modest decline in usage, making the trajectory more difficult to interpret from the available data alone.

The $2 billion objective is similarly unverified. Bloomberg’s report, as summarized by TechCrunch, supplies the target and the comparison with August’s run rate, but the source evidence does not include a Moonshot financial filing, an executive quote, customer contracts or an independent audit. Moonshot’s actual progress will depend on whether usage converts into paid inference, enterprise commitments and sustainable margins.

There is also a separate dispute over the provenance of Moonshot’s training data. Anthropic alleged earlier this week that Moonshot had conducted a long-running model-distillation effort involving nearly 300,000 requests routed from Kimi to Claude Opus. Anthropic further alleged that more than 23 million responses from its models were collected for use in Moonshot’s training.

Those claims remain allegations in the supplied evidence. The reporting does not establish a final legal finding, and it does not include Moonshot’s response. Nevertheless, the dispute matters commercially: model developers, enterprise buyers and infrastructure partners increasingly scrutinize how training data was obtained and whether a model’s development practices could expose users to legal or compliance risks.

What K3 means for builders and buyers

For AI builders, K3’s appeal is likely tied to the combination of open access and high observed activity. Teams can evaluate the model through platforms such as OpenRouter, compare it with other systems and potentially deploy weights under their own infrastructure choices. That can improve negotiating leverage and reduce dependence on a single API provider.

The same flexibility shifts more responsibility to the buyer. A team running an open-weight model must evaluate inference costs, hardware requirements, latency, version stability, security controls and the obligations attached to its use. It must also determine whether the model’s training provenance is acceptable for its application, particularly in regulated environments or products that handle sensitive data.

For Moonshot, the business challenge is converting attention into durable revenue while preserving the distribution advantages of open weights. Kimi can serve as a direct product channel, while K3 can create indirect demand through hosted endpoints and developer infrastructure. But the company will need to show that this ecosystem produces repeatable payments rather than only large volumes of low- or zero-margin traffic.

What to watch next

The clearest follow-up signal will be whether Moonshot reports revenue or bookings that substantiate progress toward the $2 billion annualized goal. A breakdown between Kimi subscriptions, API usage, enterprise contracts and other services would help distinguish model popularity from monetization.

Developers should also watch OpenRouter usage over the coming months, including whether K3’s recent decline continues and whether demand becomes concentrated among a small number of applications. New K3 releases, pricing changes, hosted deployment partnerships and evidence of enterprise adoption would provide additional clues about the model’s commercial durability.

Finally, the Anthropic dispute could affect how customers assess Moonshot. Any response from Moonshot, legal action, settlement or independent investigation would be relevant to procurement teams considering K3 for production workloads.

Creati.ai perspective

Moonshot’s target is a meaningful test of whether open-weight AI can support a large business without relying on closed-model pricing power. The reported usage figures show that distribution can be substantial, but they do not by themselves prove profitable demand.

For builders and enterprise buyers, the practical question is not simply whether K3 is popular. It is whether Moonshot can turn that popularity into reliable service, transparent economics and a defensible training record. Until the company provides more financial and operational evidence, the $2 billion figure should be treated as an ambitious target rather than an established outcome.