Inside the US$50 billion rise of Moonshot AI

(By Caixin journalist Yang Zirui)

Between February and August of 2026, a seismic shift rippled through the global AI industry, emanating from Beijing. Over just six months, the pre-money valuation of Chinese startup Moonshot AI quintupled to US$50 billion. Following a US$3.5 billion funding round that closed in late July, the developer of the Kimi large language model added US$15 billion to its value in a single month.

In China’s hyper-competitive venture capital market, large model AI has become the ultimate prize. Moonshot AI and its rival, DeepSeek, are currently the only two premium targets left in the primary market, one investor familiar with the company said, prompting a frenzy among investors.

DeepSeek’s valuation is hovering near 500 billion RMB (US$74.5 billion), making it exceedingly difficult for new capital to squeeze in. By comparison, Moonshot AI is viewed as a bargain, the investor said. With a Hong Kong listing anticipated within six months, private market funding has essentially transformed into a mad dash for early IPO subscriptions, the investor said.

“Moonshot AI fits the current high perception and high sentiment for AI companies in the Hong Kong stock market,” a Hong Kong broker told Caixin, adding that Moonshot AI’s revenue is comparable to MiniMax Group Inc. and operates an overseas business, while its model capabilities are on par with or even higher than Z.AI Co. Ltd., formerly Zhipu. “If Kimi can go public, everyone will probably invest in it.”

The meteoric rise of Moonshot AI represents a fundamental turning point in the global artificial intelligence arms race, a phenomenon that industry insiders are calling the “Kimi Moment”. It is a story of technological breakthroughs, brilliant young architects, domestic price wars and a strategic pivot that is reshaping how AI is monetised worldwide.

Global shockwaves

The catalyst for Moonshot AI’s valuation explosion was the July release of its Kimi K3 model. Boasting a staggering 2.8 trillion parameters, Kimi K3 became the world’s largest open-weight model to date. Within 24 hours, it decimated global leaderboards, going toe-to-toe with top-tier American models on comprehensive performance, second only to Anthropic’s Fable 5 and OpenAI’s GPT-5.6 Sol, according to evaluation platform Artificial Analysis. On the developer platform Code Arena, Kimi K3 captured the global No. 1 spot for front-end development capabilities — the first time a Chinese large model had claimed that crown.

The shockwaves were immediate. The launch caused a temporary computing power shortage, forcing Moonshot AI to pause new user subscriptions. Domestically, the stock prices of rival AI firms such as Z.AI and MiniMax plunged.

Internationally, Kimi K3 triggered panic and awe. David Sacks, tech investor and co-chair of the US President’s Council of Advisors on Science and Technology, publicly warned in July that while America “is tying itself in knots” with regulations, China was winning the AI race. Washington hardliners quickly proposed sanctioning Chinese open-source models.

This, in turn, sparked a revolt in Silicon Valley. Nearly 200 tech companies drafted a letter urging the US government not to restrict access to China’s open-weight AI. Shortly after, Nvidia Corp. CEO Jensen Huang led a coalition of over 270 companies — including Meta Platforms Inc., Microsoft Corp., OpenAI and Google LLC — in signing a letter emphasising the critical importance of open-source collaboration.

“Before the release of Kimi K3, the market generally believed that due to computing power constraints, the gap between Chinese large models and top US models had widened to eight to 12 months,” said the investor familiar with Moonshot AI, adding that the company had claimed in May that Kimi K3 would fully align with new US models.

“When it finally came out, it exceeded expectations,” the investor said. “This tells the world that Chinese model teams can maintain and even narrow the gap with advanced US models.”

Sun Wei, principal AI analyst at market research firm Counterpoint, believes that China now leads frontier open-source models. With the release of Kimi K3, global developers, labs and policymakers are viewing Chinese large models as strategic infrastructure, rather than simply low-cost alternatives.

American legal AI platform Harvey pivoted from OpenAI and Anthropic’s closed-source models to build its proprietary system using Kimi K3, marking a major milestone for China AI.

The architects

The brains behind this geopolitical tech shock belong to a bright, young team of engineers, anchored by Moonshot AI’s founder and CEO Yang Zhilin.

A star prodigy, Yang graduated first in his computer science class at Tsinghua University before earning his PhD at Carnegie Mellon University in just four years. Mentored by former Apple AI research head Ruslan Salakhutdinov, Yang authored foundational papers on machine learning and has over 52,000 citations to his name. But Yang declined lucrative offers from US tech giants to return to China and build his own empire.

In early 2023, following the global awakening sparked by ChatGPT, Yang launched Moonshot AI with the explicit goal of achieving artificial general intelligence. He brought alongside him a trio of Tsinghua alumni: Zhou Xinyu, an expert in algorithm mass production; Zhang Yutao, a chatbot specialist; and Wu Yuxin, a former Google Brain researcher.

Despite Yang’s claims that those investors initially passed on the new company because they doubted the prospects of Chinese large models, a consortium of venture capital firms — including GSR Ventures and Boyu Capital — filed for arbitration in Hong Kong in late 2024. Caixin has learned that as Moonshot AI prepares for its highly anticipated IPO, the company is simultaneously working with shareholders to negotiate a settlement with Recurrent AI investors.

From cash burn to cash cow

Moonshot AI’s early days were defined by a cash-burning consumer strategy. Following a US$1 billion Series A round backed by Alibaba Group Holding Ltd., HSG and Meituan, Kimi heavily marketed its chatbot, known for handling an unprecedented 2-million-character lossless context. By January 2025, Kimi boasted 22 million monthly active users, trailing only ByteDance Ltd.’s Doubao.

But competing with internet giants on consumer subsidies is a dangerous game. User loyalty was nonexistent and the market grew wary of Kimi’s massive advertising expenditures.

The industry experienced a paradigm shift when rival DeepSeek launched its highly efficient R1 model. Realising that AI startups could win on sheer model capability rather than scaling consumer userbases, Moonshot AI halted all advertising, abandoned updates to their older consumer models and focused their vast resources on building frontier foundation models.

This pivot completely transformed Moonshot AI’s business model. They transitioned from a consumer-facing app bleeding cash to an enterprise-driven platform. Kimi K3 was deployed with a customised open-weights licence that set clear commercial thresholds, targeting B2B clients and developers.

The strategy paid off. Following Kimi K3’s release, over 70% of Moonshot AI’s new users came from overseas. International subscription revenue surged 14-fold, and API revenue multiplied by ten, Caixin has learned.

The company quickly partnered with massive IT integrators such as Chinasoft International Ltd. and AsiaInfo Technologies Ltd. to deploy its technology across global energy and financial sectors. Moonshot AI is now also negotiating revenue-sharing deals with Amazon.com Inc., Google and Microsoft, Reuters reported.

By June 2026, Kimi’s annual recurring revenue (ARR) reached US$300 million, Caixin understands, representing a valuation-to-ARR multiple of 167 times, the market views Kimi as a generational infrastructure play.

Cut-throat price war

Despite its success, Moonshot AI is fighting a war on multiple fronts. The global large model market has entered a phase of white-hot competition, defined by both building 3-trillion-plus parameter models, and price wars with smaller, hyper-efficient “flash” models.

Shortly after Kimi K3’s release, American giants responded aggressively. Anthropic launched its Opus 5 and Fable 5.1 models, while Elon Musk’s SpaceX debuted Grok 4.6, all claiming marginal leads over Kimi in benchmark tests. OpenAI slashed the API prices of its GPT-5.6 series by up to 80%, undercutting Kimi K3’s pricing.

But the fiercest competition is domestic. DeepSeek introduced a “kill zone” — the release of its V4-Flash model, which forced competitors to either match its high intelligence or beat its rock-bottom price. Alibaba retaliated by launching Qwen 3.8-Max, a 2.4-trillion parameter behemoth, alongside an ultra-cheap small-parameter version. Z.AI followed suit, launching GLM-5.3-Flash at just one-tenth the cost of its flagship model.

The cost to execute tasks using these new Chinese flash models has plummeted to roughly US$0.10, completely undercutting top-tier American models that charge upward of US$3.69 per task. Z.AI Chairman Liu Debing recently confirmed the new reality that for equivalent intelligence, prices will continue to fall. Absolute frontier models that unlock new capabilities will command premium pricing.

Over the next year, the battle looks set to shift from benchmark vanity scores to return on investment and models that can deliver the highest ratio of successful tasks per dollar spent, Counterpoint’s Sun said. The market landscape will gradually concentrate toward a few top model platforms, she added.

This article was first published by Caixin Global as “In Depth: Inside the $50 Billion Rise of Moonshot AI”. Caixin Global is one of the most respected sources for macroeconomic, financial and business news and information about China.

Related: [Big read] China’s AI catch-up is forcing Silicon Valley to cut prices | Kimi K3’s lesson: The AI race won’t be won by containing China