Kimi-Maker Moonshot AI Jumps to $20B — Meituan Just Pulled It 4x in 6 Months
On May 7, Kimi developer Moonshot AI raised ~$2B led by Meituan's Long-Z arm at a $20B post-money — 4.6x its November mark of $4.3B. April ARR topped $200M and cumulative six-month raises hit $3.9B.
$20B — a Chinese open-source LLM company just hit big-five pricing for the first time
Here's the deal: on May 7, Moonshot AI (月之暗面) raised ~$2B at a $20B valuation. This is the company behind Kimi. The round was led by Meituan's VC arm Long-Z; Tsinghua Capital, China Mobile Capital, and CPE Yuanfeng joined. Six months ago Moonshot was valued at $4.3B — a 4.6x in two quarters. The bigger story isn't the markup. April ARR topped $200M, and cumulative raises over six months reached $3.9B. This is the first time a Chinese LLM lab has stepped into global big-five pricing territory (OpenAI, Anthropic, xAI, Mistral, Cohere).
The cast — Moonshot, Kimi, Meituan, Tsinghua, China Mobile
Start with Moonshot. Founded March 2023, CEO Yang Zhilin (杨植麟) is a Tsinghua undergrad with a Carnegie Mellon PhD. Flagship models: Kimi K1.5 (Jan 2025), Kimi K2 (May 2025), Kimi K2.5 (Feb 2026). The 2024 "long context" (2M tokens) wedge accelerated user share inside China. As of April 2026, the Chinese LLM chatbot share split is roughly Doubao 35%, DeepSeek 22%, Kimi 18%, Qwen 15%, Yuanbao 8% — Kimi anchored at #3.
Kimi is the chatbot brand. Differentiated since 2024 around "long context = read a book then answer." From K2.5 it adds multimodal (image / video) and Kimi Agent (browser automation, code execution). Daily actives ~120M, #3 in China.
Meituan is the Chinese super-app for delivery, reservations, and payments. Since February 2026, it has been actively investing in AI infrastructure and labs through Long-Z. The strategic logic for the Moonshot check is (1) plug Kimi tech into Meituan's in-app assistant Wanjia, (2) fine-tune Kimi on delivery / reservation / payments data, and (3) build an AI stack independent of Tencent and Alibaba — Tencent owns Doubao via ByteDance, Alibaba owns Qwen, Meituan needed its own.
Tsinghua Capital is the VC arm of Tsinghua University. From 2024 they've concentrated on AI, semiconductors, and biotech. Founder Yang is a Tsinghua alum, so this is a "home team" bet.
China Mobile is the world's largest carrier with 900M subscribers. China Mobile Capital is in to overlay an "AI assistant service" on the network. Reports indicate H2 2026 will see all new China Mobile signups receive a "Kimi Pro" 6-month free bundle.
The hard numbers — $3.9B raised, ARR $200M, open-source license
The deal, the cumulative raises, and the revenue picture in one table.
| Metric | Value | Comp |
|---|---|---|
| This round | ~$2B | Largest single LLM round in China |
| Post-money | $20B | 4.6x the $4.3B mark from 6 months ago |
| 6-month cumulative raised | $3.9B | 65% of OpenAI's full-year 2024 raise of $6B |
| April ARR | $200M+ | ~1% of OpenAI's $20B ARR |
| DAU | ~120M | ~15% of ChatGPT's 800M WAU |
| China LLM chatbot share | 18% (#3) | Behind Doubao 35%, DeepSeek 22% |
| Model license | Open source (Apache 2.0) | Same as DeepSeek |
The most interesting line is the ARR-to-valuation multiple: $20B / $200M ARR = 100x. Compare OpenAI ($20B ARR, $500B → 25x), Anthropic ($3B ARR, $183B → 60x), xAI ($1.2B ARR, $80B → 67x). Moonshot is the highest multiple in the cohort — which is either "real value the market just recognized" or "China LLM bubble." That call gets made over the next 12 months.
The open-source license is the real differentiator. Releasing Kimi K2 weights under Apache 2.0 drove ~50M Hugging Face downloads in three months. Beyond API revenue, Moonshot can monetize "enterprise self-host consulting." DeepSeek pioneered this exact playbook and it was the catalyst behind their ARR ramp. Moonshot is now running the same play with similar results — $200M ARR by April.
The $3.9B six-month raise is at hyperscaler-subsidiary scale. OpenAI raised $6B across all of 2024; Moonshot raised 65% of that in half the time. Chinese capital is moving fast on AI. Caveat: ~70% of those dollars are "state-adjacent" (Tsinghua Capital, China Mobile, Meituan), which materially raises CFIUS risk on any future US market entry.
What each side gets — Moonshot, Meituan, China Mobile, Tsinghua
Moonshot gets three things. First, capex runway. $2B fresh primary at $20B post buys ~18 months of infra. Estimated annual GPU spend at 120M DAU is ~$800M; this round funds ~2 years of operations. Second, talent leverage. The "$20B" tag is the recruiting hook for US PhDs and Silicon Valley engineers considering a return-to-China move. Third, IPO optionality. China's May 1 IPO rules opened a fast-track listing path for AI companies, and Moonshot is being floated as a 2027 candidate.
For Meituan, the wedge is "AI super-app integration." With Kimi Agent inside the Meituan app, a user can say "book me an 8 PM table at a sushi place near Gangnam tonight, pay with my card" and have it executed end-to-end. That's a differentiation Alibaba and Tencent struggle to match quickly. Internal estimates put ARPU lift at 18-25%.
For China Mobile, the "telco + AI bundle" is a new revenue stream. ARPU is roughly 1/5 of Korean carriers like SKT/KT; a "Kimi Pro 6-month bundle" could lift per-subscriber ARPU by $2-5/month. If 10% of 900M subscribers convert, that's $20-50B of incremental annual revenue.
For Tsinghua Capital, this is "home team" pride. Moonshot reaching $20B led by a Tsinghua alum boosts the value of the alumni network and makes future Tsinghua Capital fundraising materially easier.
Historical comps — DeepSeek's win and Inflection's stumble
Win #1: DeepSeek. After DeepSeek V3 dropped in December 2024, ARR ramped from $0 to $500M in six months. SeaTowner Group led an $8B-valuation round in July 2025; another at $20B in December 2025. A 5x in six months — exactly the Moonshot pattern. Markets are positioning Moonshot as the next-cycle DeepSeek.
Win #2: ByteDance Doubao. ByteDance committed $5B+ of internal capital to Doubao through 2024 and grabbed #1 in China. Parent ByteDance valuation moved from $400B to $700B in eight months — a 1.7x. The pattern: a super-app with its own AI stack gets a multiple bump.
Bust #1: Inflection AI. June 2023, $1.3B raise at $4B. Within nine months almost the entire team migrated to Microsoft, leaving the corporate shell behind. "Big raise = guaranteed value" doesn't hold. Whether Moonshot's $20B has the same risk gets verified over 18 months.
Bust #2: China's Zhipu AI. Tsinghua-founder lab that hit $5B by 2024, but lost model performance ground to DeepSeek and Moonshot, ending up in a $3B down-round in December 2025. The "Tsinghua Chinese founder" label isn't a guarantee. Moonshot's $20B "real or bubble" gets answered in the next 12-18 months.
A more balanced parallel: Mistral AI. France's Mistral went from $6B to $12B in six months on an "open source + EU data sovereignty" narrative. The "Open-source + regional sovereignty AI" wedge works. Moonshot's "China + open source" is the same wedge with a different region.
Counter-plays — DeepSeek, Doubao, Qwen, MiniMax
DeepSeek is the closest competitor. After hitting $20B in December 2025, May 2026 estimates put DeepSeek at $35-40B. ARR moved from $500M to ~$1.2B. DeepSeek V4 Pro is expected in May, and it competes head-on with GPT-5.5 Instant and Claude Opus 6. Moonshot needs K3 (expected late September / October 2026) to keep pace.
ByteDance Doubao leads on user count (35% share) but plays a different game: closed-source, tight super-app integration. Moonshot's open-source angle doesn't directly threaten Doubao but does win in "enterprise self-host." ByteDance's June 2026 launch of Volcano AI cloud is the counter.
Alibaba Qwen runs open-source plus Alibaba Cloud integration. Qwen3 and Qwen3-VL are the most-downloaded Chinese open-source models on Hugging Face. Direct overlap with Moonshot. Qwen 3.5 dropped May 8 and narrows the gap with K2.5. Qwen's distribution is single-channel via Alibaba Cloud, where Moonshot has broader reach.
MiniMax is smaller but its M2 release on May 8 took SOTA on video generation and long context. Share is sub-5% but enterprise demand for video content puts pressure on Moonshot's weaker video story.
US competitors: OpenAI, Anthropic, Mistral, Cohere. With US LLMs effectively blocked from China, direct competition is muted. The bigger question is Moonshot's US entry — and the 40% Tsinghua / China Mobile capital stack creates a CFIUS barrier.
So what changes — Korean devs, Chinese users, global capital
For Korean developers, two changes. First, Kimi K2.5 weights are usable. Apache 2.0 means Korean startups can self-host. Korean-language quality is roughly 80% of GPT-5.5 Instant at 1/10 the cost — a fit for cost-sensitive workloads (medical record summarization, legal doc classification). Second, Kimi Agent API enables automation builds. Reports indicate Korean e-commerce (Coupang, 11Street) started piloting "Kimi-powered personal shopper" bots in May.
For Chinese users, the win is "Kimi Pro price cut + China Mobile bundle." Kimi Pro list price drops from 50 RMB/month (~9,500 won) to 30 RMB (~5,700 won) starting June. China Mobile subscribers get six months free. 120M DAU could grow 50% to 180-200M within six months.
For global capital, the signal is "China LLM pricing entered the global big-five tier." Moonshot's $20B beats Mistral's $12B and is 2.5x Cohere's $8B. The "Chinese AI = cheap and local" framing is breaking. The cap on the upside is the lack of a US IPO path — only HK and Shanghai listings available.
Last note: the "open vs closed source" debate just heated up. Moonshot, DeepSeek, Mistral, and Llama on the open side; OpenAI, Anthropic, and xAI closed. As of May 7 the open-source camp has cumulative ARR of ~$2.5B vs closed-source $24B — 10:1 today. Whether that closes to 5:1 inside 12 months is the next big fork.
References
출처
관련 기사
Yann LeCun's AMI Raises $1.03B Seed — The Biggest Bet Against LLMs
Yann LeCun's AMI Labs Raises $1.03B Seed — The Biggest Contrarian Bet Against LLMs
Deeptune Raises $43M Series A Led by a16z — Building Training Gyms for AI Agents
AI 트렌드를 앞서가세요
매일 아침, 엄선된 AI 뉴스를 받아보세요. 스팸 없음. 언제든 구독 취소.