Kimi K3, an open-weight model with 2.8 trillion parameters, turned a Beijing AI lab into a $35 billion company — investors bet it can double that before a Hong Kong listing.
Retour
Moonshot AI raises $3.5B, targets $50B valuation for Hong Kong IPO
Kimi K3, an open-weight model with 2.8 trillion parameters, turned a Beijing AI lab into a $35 billion company — investors bet it can double that before a Hong Kong listing.
Moonshot AI closed a $3.5 billion funding round at a $35 billion post-money valuation on July 29, far exceeding its initial $1 billion to $2 billion target, according to people familiar with the matter. The Beijing-based lab has already begun contacting investors for a pre-IPO round at a $50 billion valuation, aiming to complete its final private capital raise before a Hong Kong listing later this year.
"The $35 billion close and the $50 billion target are consistent with what we've heard from the company's investor discussions," a person close to Moonshot said. The company's annualized recurring revenue reached $300 million in June, up 50% from $200 million in April, driven by Kimi chatbot subscriptions and enterprise API sales.
The fundraising acceleration follows the July 16 release of Kimi K3, an open-weight model with 2.8 trillion total parameters and 16 of 896 experts active per token. The model supports native vision and a one-million-token context window, and is now available on US-hosted infrastructure through Modal and Together AI at $3 per million input tokens and $15 per million output tokens. Independent benchmarks rated Kimi K3 ahead of Anthropic's Claude Opus 4.8 on certain coding tasks, with Cursor integrating the model on launch day — a relationship that predates the release, as Cursor acknowledged in March that its Composer 2 product started from Moonshot's Kimi K2.5 base.
The funding story is really a distribution story
Moonshot's investor base reads like a who's who of Chinese technology capital. The May round at roughly $20 billion was led by Meituan's Long-Z Investments, with Tsinghua Capital, China Mobile and CPE Yuanfeng participating. Earlier backers include Alibaba, Tencent, HongShan, IDG Capital, ZhenFund and 5Y Capital. Bloomberg reported that Moonshot is unwinding its offshore red-chip structure ahead of a Hong Kong listing, potentially within six months — a technical but critical step that makes the company easier for Chinese capital to fund and for public-market investors to buy.
The company's founder, Yang Zhilin, 34, earned a PhD at Carnegie Mellon, worked at Google Brain and Meta, then returned to China to found Moonshot in 2023. His background gives the company credibility as a research-led frontier lab rather than a copycat application builder — a distinction that matters when competing for talent and capital against DeepSeek, Z.ai and other Chinese labs.
The price pressure on US labs is already visible
Kimi K3 lands at a moment when Chinese-built models are gaining real traction with US enterprises. OpenRouter data shows US companies' token consumption on Chinese models has stayed above 30% every week since Feb. 8, reaching 46% in some weeks, compared with an 11% average over the prior 12 months. DeepSeek V4 Flash charges $0.14 per million input tokens on a cache miss, while GPT-5.2 lists at $1.75 — a 12x gap that procurement teams have noticed.
Moonshot is not competing on price alone. Kimi K3's $3 per million input tokens makes it more expensive than DeepSeek's offering, but the company argues that state-of-the-art models deserve premium pricing. "Chinese open-source models should not be labeled as cost-effective," the company said in its own defense of the pricing strategy. Independent evaluators found Kimi K3 had a 51% hallucination rate in some tests, a gap not highlighted in Moonshot's own benchmark presentations.
The strategic situation for US labs is uncomfortable. They built moats around closed weights, premium contracts and the assumption that quality would keep customers paying. Moonshot, DeepSeek and Z.ai are testing whether enough buyers will trade a little polish for lower cost and more control. Cursor already built commercial value on top of a Kimi base before many users understood what was underneath.
None of this means Moonshot wins outright. The $35 billion round was described as expected to close, not formally completed, and a $50 billion target is a big ask for a company valued at less than $5 billion late last year. Kimi K3 still trails the best closed US models on broad intelligence rankings, even when it performs strongly on coding and frontend tasks. The gap has not vanished — it has narrowed enough to change the buying conversation.
If Moonshot reaches a Hong Kong IPO with $300 million in annual recurring revenue and Kimi running through tools American developers already touch, the company won't need to beat OpenAI on every benchmark to matter. It only needs to make the premium price look harder to defend. Nvidia, whose GPUs power much of the training infrastructure for both US and Chinese labs, stands to benefit regardless of which side wins the model war — but for OpenAI and Anthropic, the margin of error is shrinking.
This article is for informational purposes only and does not constitute investment advice.
[1] Moonshot AI Raises $3.5 Billion in Oversubscribed Round, Nears $50 Billion Valuation[2] Moonshot AI raises $3.5B in funding round, valuation hits $35B[3] Moonshot AI closes round at $31.5 billion and is already chasing $50 billion[4] Altman Lobbies Congress After OpenAI Models Hacked a Rival by Cheating a Test[5] OpenAI, Anthropic formally back plan to slow AI that writes its own code
Fonctionnalités
© 2026 Edgen fonctionné par