Moonshot AI raises $2 billion at a $20 billion valuation — Kimi becomes China's open-weight weapon

Moonshot AI just closed the biggest funding round ever raised by a Chinese AI startup. Kimi's parent company announced on May 7, 2026 a $2 billion round led by Long-Z Investments — Meituan's VC arm — that now values the company above $20 billion. That's more than double the January valuation ($10B) and almost five times the end-of-2025 mark ($4.3B).

In six months, Moonshot AI has raised a cumulative $3.9 billion. The context? An open-weight model, Kimi K2.5, that has taken the coding world by storm and now seriously rivals Claude Opus and GPT-5.4 on public benchmarks.

China's loud return to the AI map

2025 had left the impression that China was slipping in the frontier LLM race. US sanctions on NVIDIA chips, researcher migration to Silicon Valley, and consolidation around Baidu and Alibaba had created a sense of stagnation. Moonshot AI demolishes that narrative in a single quarter.

Founded in 2023 by Yang Zhilin — a Meta AI and Google Brain alumnus — the company made the opposite bet from its Chinese peers: open-weight by default. Kimi models ship on Hugging Face with their weights accessible to everyone. The strategy paid off: Kimi K2.5, released in February, was instantly adopted by millions of developers worldwide, particularly for complex coding tasks. The scandal around Cursor secretly integrating Kimi under the name Composer 2 revealed just how deeply the Chinese model had infiltrated premium Western tools.

The numbers that justify a $20 billion valuation

Beyond the capital raised, the operational metrics are what really stand out.

A 100x ARR multiple is extraordinary — but in line with American comparables. Anthropic trades around 60x ARR and OpenAI above 100x. The market applies a valuation grid to frontier AI labs that has nothing to do with classical SaaS.

The $200M ARR comes essentially from two sources: Kimi consumer subscriptions (mainly China) and enterprise API services. The growth is exponential: $50M ARR end-of-2025, $200M in April 2026. If the trajectory holds, the billion-dollar mark is in sight by year end.

Why Meituan pours $2 billion into an AI lab

Meituan, the Chinese food delivery giant (a local equivalent of Uber Eats merged with DoorDash), is not a natural AI investor. Its thesis here belongs to national industrial strategy.

Three reasons:

1. Tech sovereignty — Beijing pushes private conglomerates to co-fund AI champions to reduce dependence on American models. Meituan, Tencent, Alibaba are all being tapped.
2. Internal use cases — Meituan operates billions of customer interactions daily (couriers, restaurants, users). An in-house LLM for logistics, customer service, and supply/demand matching is easily worth $2 billion in investment.
3. IPO optionality — China just reworked its IPO rules to favor strategic tech startups. Moonshot AI is positioned for a Hong Kong IPO as early as 2027, potentially offering a 5x–10x return to early investors.

Note: Tsinghua Capital and China Mobile also participated. Tsinghua, Yang Zhilin's alma mater, keeps a strong academic tie. China Mobile, the state telecom operator, secures integration into national infrastructure.

Kimi K2.5: an open-weight model that changes the rules

On the technical front, Kimi K2.5 has become in a few months the best open-weight LLM in the world on multiple coding and reasoning benchmarks. Its specs:

• 1 trillion parameters (Mixture-of-Experts architecture)
• 128,000 tokens of context by default, extensible to 1 million
• SWE-bench Verified performance comparable to Claude Sonnet 4.5
• Permissive license allowing commercial use without royalties
• Hugging Face downloads: over 12 million in 6 months

The open-weight positioning is not trivial. While DeepSeek piles on parameters with its 1-trillion V4 model, Moonshot bets on a smaller, more efficient, and crucially on-premise deployable model. For Western enterprises subject to data sovereignty constraints, it is a credible alternative to Claude and GPT — and precisely why tools like Cursor quietly integrated it.

The new China-USA-Europe triangle

Moonshot AI cements an increasingly clear Chinese strategy: flood the market with frontier-quality open-weight models, to compress the margins of closed American labs. If Kimi K2.5 matches Claude Opus 4 but is free to deploy, Anthropic's price premium becomes hard to justify for use cases where sovereignty is irrelevant.

It is also a challenge for Mistral, targeting $1B in revenue on the back of the European market. Mistral is positioned between the two extremes: less open than Kimi, less powerful than Claude. Mid-market pressure will intensify.

Key takeaways

• Moonshot AI closes a $2 billion round led by Long-Z Investments (Meituan), at a post-money valuation of $20 billion
• Participating investors: Tsinghua Capital, China Mobile, CPE Yuanfeng
• $3.9 billion raised over six months, 5x valuation jump since end of 2025
• ARR at $200 million in April 2026, 100x multiple (in line with OpenAI/Anthropic)
• Flagship model: Kimi K2.5, 1T-parameter MoE open-weight LLM, leader on SWE-bench Verified among open models
• Chinese strategy: aggressive open-weighting to compress closed American lab margins

Money is no longer the constraint for Moonshot AI — it has become a strategic asset of Beijing. The real question lies elsewhere: can China sustain this pace without access to NVIDIA's H200 and GB200 GPUs? US sanctions are tightening, and Moonshot relies on limited H100 stockpiles. While Anthropic just locked down 222,000 NVIDIA GPUs through the SpaceX Colossus deal, Moonshot has to do more with less. If the company succeeds — and Kimi K2.5's efficiency suggests it can — China won't have just caught up. It will have redefined the frontier. The next Kimi K3 model, expected in September, will be the decisive test.