This page gives you a quick introduction to Kimi. Kimi is the flagship AI assistant from Chinese startup Moonshot AI. The Beijing company was founded in 2023 by former Google Brain researcher Yang Zhilin and made its name in the Chinese-language market through the ability to read exceptionally long documents. One distinction worth keeping straight from the start: Kimi is the product name; Moonshot AI is the company name — the two are often used interchangeably, but they’re not the same thing.
The company’s direction took a notable turn: from an assistant that played up ultra-long context windows to open-source large models — most prominently Kimi K2, with successive updates — and agentic coding, meaning the model plans, writes code, and calls tools autonomously to complete tasks.
One line to remember it by: a scholar-founded Beijing AI startup that pivoted from “reading really long documents” to “open-source LLM plus agents.”
The aspect that gets the most attention is how fast the valuation has climbed: media and investment-advisor reports pegged a late-2025 round at around US$4.3 billion and a mid-2026 round at around US$20 billion. But these are media and advisor figures — Moonshot AI hasn’t confirmed each one, the company has no audited financials, and fundraising advisors have an incentive to pitch high numbers. Keep “reported valuation” and “confirmed figure” as separate categories.
Key Data Snapshot
The key figures in one place. Moonshot AI is a private company with no public financials, so quite a few numbers below come from media or investment-advisor sources — we try to label them clearly.
| Item | Data |
|---|---|
| Product name | Kimi (AI assistant app + API platform) |
| Parent company | Moonshot AI |
| Founded | 2023 |
| Headquarters | Beijing, China |
| Founder / CEO | Yang Zhilin (former Google Brain researcher, CMU PhD); co-founders Zhou Xinyu, Wu Yuxin |
| Company type | Private company (not public, not a PBC) |
| Latest valuation | ~US$20B (mid-2026, media / investment-advisor framing, not audited); previous round ~US$4.3B |
| Flagship products | Kimi App (consumer-facing), Kimi API (OpenAI-compatible), open-source Kimi K2 series |
| Open-source strategy | Kimi K2 released under Modified MIT license (subsequent versions vary in open-source status); API is paid |
| Primary use cases | Chat, long-document and deep research, agentic coding, enterprise (finance, AI for Science) |
| Key investors | Alibaba (~30% preferred shares per its 2024 disclosure, financial investment not control), Tencent, IDG, ZhenFund, and others |
| Compute variables | Relies on large-scale GPU clusters (proprietary Mooncake serving architecture); actual chip types and sources not disclosed |
Two reminders when reading these numbers. ① Moonshot AI’s valuation and ARR mostly trace back to media and fundraising advisors (who have an incentive to pitch numbers high), and none of it has been audited. Label it “reported / advisor framing” when you cite it. ② Kimi is the product; Moonshot AI is the company. When the internet talks about “Kimi’s valuation,” it means the company Moonshot AI — don’t conflate the product with the company.
A Quick Tour of Seven Dimensions
Seven dimensions are a useful frame for getting to know an AI company. Penchan will break the key ones into standalone pieces over time.
① Technology and product direction: Moonshot AI first built its reputation on ultra-long-context assistance, then shifted to open-source large MoE models — most prominently Kimi K2 (Modified MIT license, with K2.5, K2.6, and beyond to follow). The company developed a serving architecture it calls Mooncake to reduce inference compute costs, and the product roadmap is heading toward agentic coding. One note: the “read millions of Chinese characters at once” claim from early marketing is not the same concept as the model’s token limit.
② Customer structure and market positioning: Kimi serves three tiers: a free app for general users, an API for developers, and enterprise customization. Its differentiating path started with “ultra-long context” and has turned toward “open-source LLM plus agentic coding.” The China competitive field includes DeepSeek, Qwen, Doubao, GLM, and MiniMax; internationally it faces OpenAI, Claude, and Gemini. This page presents them in parallel without picking a winner.
③ Ecosystem and partnership strategy: Moonshot AI’s investor roster draws attention in its own right — ZhenFund, IDG, and Wuyuan Capital are in there along with Alibaba and Tencent. Per Alibaba’s SEC disclosures (2024), Alibaba holds roughly 30% preferred shares as a financial investment, not a controlling stake. The open-source play has landed Kimi K2 across third-party developer tools and platforms, widening developer reach.
④ Valuation and financial model: Moonshot AI is private and has moved fast through funding rounds, lifting its reported valuation sharply in a little over a year. The circulating valuation and ARR figures mostly trace back to media and fundraising advisors (who have an incentive to pitch numbers high), haven’t been officially confirmed by Moonshot, and aren’t audited. Opening its flagship model to open source may compress direct API revenue in the short term, traded against broader developer ecosystem reach.
⑤ Commercialization risk and regulation: Several pressures are running simultaneously. China’s LLM market is in a price war, and the high compute costs of training and inference make gross margin a long-term variable. The China regulatory side requires generative AI registration and content safety compliance (the standard compliance backdrop for the industry). Media reports have also pointed to allegations that several Chinese AI apps including Kimi were collecting data unrelated to their stated functions; Moonshot did not respond promptly at the time. Claims like this should be cited with sources rather than taken as settled.
⑥ Geopolitics and supply chain: Moonshot AI’s published Mooncake architecture confirms it depends on large-scale GPU clusters, and the ongoing tightening of US export controls on high-end AI chips for China is a structural risk shared across companies in this space. Its actual chip types and sourcing remain undisclosed — that part stays black-box; no inferences here.
⑦ Leadership and governance: Founder Yang Zhilin has an academic background (former Google Brain, CMU PhD, with notable involvement in Transformer-related research). Co-founders also carry academic and big-tech credentials — the “academic-founder” archetype is common across Chinese AI startups. As a private company, key investors include Alibaba (preferred shares), Tencent, and IDG, but board seats and control terms haven’t been made fully public.
Key Milestones
The pivotal moments that brought Moonshot AI to where it is today (valuation figures are all media / investment-advisor framing):
| Date | Milestone |
|---|---|
| 2023 | Moonshot AI founded; launches Kimi (emphasizing ultra-long context) |
| 2024 | Multiple large funding rounds with Alibaba, Tencent, and others participating (media reporting) |
| Jan 2025 | Publishes k1.5 technical report (enhanced reasoning, arXiv) |
| Jul 2025 | Releases open-source large model Kimi K2 (Modified MIT) — a pivot-defining moment |
| Dec 2025 | Series C (IDG lead), valuation ~US$4.3B (media framing) |
| Mid-2026 | New round reported; valuation ~US$20B (media / investment-advisor framing, not audited) |
Milestones will continue to be updated; figures based on the latest announcements (table last compiled: May 2026).
Further Reading and Upcoming Standalone Pieces
Penchan will break the key dimensions into standalone pieces, rolling them out over time:
- How is Moonshot AI’s valuation calculated? Reported framing vs. the audit reality
- What is Kimi K2 open source? The Modified MIT license explained plainly
- Yang Zhilin and Moonshot AI: how a scholar-founder’s Chinese AI startup grows up
- Comparing two open-source flagships: the different paths from DeepSeek
- Want to use Kimi for free, or get a step-by-step walkthrough? Penchan will cover that separately (link to follow once live)
- For the full hardware chain: The AI Hardware Supply Chain, End to End