On June 8, Bloomberg reported that Moonshot AI is seeking a new round of financing, aiming to raise up to $2 billion with a post-money valuation target of $30 billion.

This translates to roughly RMB 200 billion.

What kind of scale is this? It's approximately twice the market capitalization of Haidilao and higher than most A-share listed companies. And achieving this valuation is an AI company called Kimi, founded just three years ago with a total of over 300 employees.

The $30 billion valuation isn't even the highest in this track (sector)—during the same period, Zhipu went public in Hong Kong with a market cap that once exceeded $90 billion; DeepSeek is reportedly negotiating its first external funding round with a valuation of around $45 billion; even MiniMax, founded around the same time as Kimi, has already landed on the Hong Kong stock exchange with a $6.4 billion valuation.

Valuations of Chinese large model companies have collectively skyrocketed over the past six months.

How did these numbers come about? What's happening behind the scenes?

To understand this collective surge, we need to go back two years.

In 2024, the valuation anchor for large model companies was monthly active users (MAUs). At that time, Kimi was the master of traffic—with its differentiated selling point of "2 million-word ultra-long text," its MAUs once surged to 36 million, making it the undisputed top player in large models. The industry generally believed: whoever had more users would have the future.

This was the valuation inertia left over from the internet era. Over the past two decades, every financing round for BAT companies told the same story: first acquire users, then discuss monetization. Investors had been trained on this logic for an entire generation.

But large models aren't apps.

The strategy of burning money for traffic encountered a fatal problem in this sector: inference costs. Every user call for a large model involves real computational power consumption, with marginal costs that don't approach zero as user numbers grow—instead, they rise. ByteDance, Alibaba, and Tencent could make their products free with nearly unlimited subsidies, but ordinary startups simply couldn't keep up.

By March 2026, Kimi's MAUs had dropped to 8.34 million, ranking behind Doubao (345 million), Qianwen (166 million), and DeepSeek (127 million)—not even in the top five. Following internet valuation logic, Kimi should have been out of the game long ago.

Yet its valuation soared from $4.3 billion to $30 billion during this period.

What happened in between? The answer: the rules of the valuation game were quietly rewritten.

Investors stopped asking "how many users do you have" and started asking "how deep is your technical moat," along with a more direct question: "how much money are you making now, and what's your growth rate?"

Kimi's answers made the market reassess it.

In January 2026, Kimi's paid orders grew 8,280% month-over-month—over eighty times. Its Stripe payment ranking jumped from outside the top 100 to 22nd globally, then to 9th a month later. By March, the company's ARR (annualized recurring revenue) surpassed $100 million; by April, it exceeded $200 million—a speed that would be top-tier among global SaaS companies.

More notably, where the money came from. Kimi's revenue explosion relied on two main pillars: paid subscriptions and B-end API calls. The latter meant that more developers and enterprises were using Kimi's model capabilities to build their own products. The well-known code tool Cursor, for example, was revealed to be using Kimi's model at its core—incidentally, Cursor itself has a valuation of $50 billion.

This represents a new valuation logic: the value of large model technology as infrastructure. It's no longer compared to internet apps but is closer to semiconductors or EDA software—not necessarily used by the most people, but every user depends on it and is willing to pay real money.

But Kimi's journey to this point hasn't been smooth. In fact, just before this commercialization explosion, it was going through its toughest period.

In early 2025, DeepSeek emerged. With its low-cost, high-performance model, it quickly went viral domestically and internationally, becoming a phenomenon in the AI circle. For Kimi, the impact came not just from external competition but internally as well—during that time, an employee bluntly asked at the annual meeting: "Given the current situation, why would job seekers choose Kimi over DeepSeek?"

Kimi's response was a complete strategic refocus.

In terms of direction, it abandoned scattered explorations and fully committed to "model-first"—shifting its technical approach from "long-text assistant" to "agent model for complex tasks," focusing on three dimensions: token efficiency, long contexts, and agent swarms. In terms of product, Kimi K2.5 was released in early 2026, immediately topping call volumes on the OpenRouter platform and directly triggering that paid growth surge.

Interestingly, Kimi's algorithm team's first reaction to DeepSeek's viral success wasn't anxiety but "excitement." A phrase circulating internally was: "DeepSeek saved us"—because it proved the direction of the technical approach and forced the team to find its true position.

This mindset perhaps explains something fundamental about Kimi. With over 300 employees averaging under 30 years old, 80% introverted, and the company having no departments, no job titles, and no OKRs. In 2025, a 17-year-old high school intern published a paper as first author, which was subsequently forwarded and evaluated by several well-known figures in Silicon Valley. This indirectly proves Kimi's corporate culture: as long as the direction is right, age and seniority don't matter.

Kimi's story is, of course, not an isolated case.

Extending the timeline to late 2025 through the first half of 2026 reveals something interesting: nearly all leading large model companies saw their valuations surge densely within this window. Zhipu completed its Hong Kong listing and has initiated an A-share IPO; MiniMax also landed on the Hong Kong stock exchange; DeepSeek is reportedly negotiating its first external funding round with institutions like the National Integrated Circuit Fund; and Jieyue Xingchen is rumored to be about to submit its Hong Kong listing application.

Within six months, the valuation coordinate system for this sector was elevated by an order of magnitude, with the entire industry reaching the same tipping point.

This tipping point can be described with a set of numbers: in 2026, China's AI-native apps reached 440 million MAUs, with 130 million new users in a single quarter (QuestMobile data). This means large model products are transforming from "tech media buzzwords" into everyday tools for ordinary users, with genuine willingness to pay on the rise. Kimi's ARR surging from zero to $200 million in months is a clear industry signal telling the market that "large models can truly make money."

This resembles the mobile internet in 2013-2014. Back then, the trigger for market revaluation was the emergence of the first batch of truly profitable native apps—WeChat began monetization, and Didi achieved a viable unit economic model. Once someone proved "there's real money to be made on this path," the entire valuation system began to rewrite. Today's trigger in the large model sector follows the same mechanism, just faster and with bigger numbers.

Of course, there are nuances to this revaluation worth examining.

Take Kimi again: $200 million in ARR isn't small, but against a target valuation of $30 billion, the PS multiple is around 150x. This number wouldn't hold in any mature industry—even in the high-growth SaaS sector, top companies typically have PS multiples between 30x and 50x.

On the other hand, the pressure to burn cash is real. Data shows MiniMax burns about $27.9 million per month, and Kimi's scale and investments are only larger. Its $10 billion cash reserves aren't immune to pressure from sustained model training and computational power investments.

But the counter-logic supporting these valuations also exists: OpenAI reached a $300 billion valuation in 2025, and Anthropic's valuation reportedly exceeded OpenAI's by June 2026, reaching $965 billion. If Chinese model companies can maintain technical competitiveness and serve markets no smaller than any other economy, whether $30 billion is high or low is itself a question without a standard answer.

Valuation, after all, is a form of expectation. It measures not today's revenue but the market's collective judgment of where a company can reach at some future point.

When the $30 billion financing news broke, neither Kimi nor Yang Zhilin responded.

That same month, at the Tsinghua AGI Summit's closing speech, he shared a conversation he had with Kimi—he asked: "AGI's arrival might threaten humanity; should researchers continue?" Kimi replied: "Even with risks, we should continue, because abandoning AGI means abandoning human civilization's potential." Yang agreed with that answer and then shared his plan: "I hope to continue improving K4, K5, up to K100 over the next ten to twenty years."

The rules of the game have indeed changed. But the game isn't over—it's a long journey, and Kimi is on the road, with the entire industry still on the move.