Listing is not an option; it is a necessity.
Article author and source: Bowang Finance
From 280 billion parameters to $300 million in ARR, this 300-person company crafted a precise capital narrative in just five days—K3 as the technological leverage, API revenue as the commercial foundation, and a Hong Kong IPO as the final rivet securing the story in the capital markets.
On the early morning of July 16, Moonshot dropped a technical bomb.
Kimi K3—2.8 trillion parameters, the world’s largest open-source model—was released without any prior announcement. On the programming leaderboard Code Arena, it ranked first with a score of 1679, surpassing Anthropic’s Claude Fable 5 (1631) and OpenAI’s GPT-5.6 Sol (1618). In seven categories including frontend development, algorithms, and data structures, K3 secured first place in six.
Source: Artificial Analysis / Code Arena
The market was stunned. Within 72 hours of K3’s release, the U.S. AI sector lost approximately $470 billion, and the Philadelphia Semiconductor Index plunged 12.5% in a week, slipping directly into technical bear market territory. The Hong Kong market suffered equally severe losses—Zhipu (02513.HK) dropped 28.49% in a single day, while MiniMax fell 15.62%. Musk posted just one word on X: “Impressive.” The Kimi official account replied succinctly: “Welcome to the $2 trillion+ club.”
The show has just begun.
On the night of July 19, three days later, Moonshot announced a temporary halt to new user registrations—its computing power was overwhelmed by the surge in demand brought by K3. On the same day, the company sent investors a proposal for an initial public offering, aiming to complete a Hong Kong stock exchange IPO within as little as six months. On July 21, co-founder Huang Zhenxin held a media briefing, addressing K3 and the company’s commercialization plans comprehensively for the first time.
Three things crammed into five days. This is no coincidence.
K3 is Yang Zhilin’s leverage, and the company’s $300 million ARR is its foundation. The urgency to go public stems from the fact that Kimi has proven that Chinese large models can generate revenue through APIs—and an IPO is the only way to lock this story into the capital markets.
Total funding has exceeded RMB 37 billion (approximately USD 5 billion), backed by top-tier institutions such as Sequoia China, Xiaohongshu, Alibaba, and Tencent. This capital has given Moonshot AI the confidence to develop K3, but it also means—investors must have an exit pathway. No matter how much capital flows in from the private market, it must eventually return. An IPO is not an option; it is a necessity.
First, let’s talk about the parameters: 2.8 trillion—the largest open-source model ever, a number that speaks for itself.
But what the Dark Side of the Moon truly wants to convey is not “I’m large.” The real trump card is K3’s architecture: the KDA hybrid linear attention mechanism combined with attention residuals, and a MoE architecture with 896 experts, activating only 16 during each inference. This combination is clearly designed to drastically reduce costs through extreme sparsity while preserving performance. The 1 million token context window is no mere feature—it’s a hard requirement for enterprise clients handling long documents, codebases, and financial report analyses.
Programming ability is K3’s sharp edge. With a Code Arena score of 1679, it not only surpassed Fable 5 and GPT-5.6 Sol but also secured first place in six out of seven categories, including frontend development. Dean W. Ball, OpenAI’s Strategic Lead, offered a telling evaluation: “K3 delivers exceptional performance that cannot be achieved through distillation.” The implication is clear—this isn’t copied; it’s built from scratch.
But K3’s greatest strength isn’t its benchmark score—it’s its pricing.
Input at $3 per million tokens, output at $15 per million tokens—exactly the same as Anthropic’s Sonnet 5. Kimi is no longer competing on “value for money”; it’s now seeking “pricing power.” What does this mean? It means Kimi’s cost structure can now support this price, and the market agrees.
The Dark Side of the Moon calculated its own costs in its technical documentation: K3’s API costs are less than one-third of those of top-tier American models. The fact that it dares to price itself the same as Sonnet 5 speaks volumes about its gross margin. This isn’t a price war—it’s a profit war.
Of course, K3 is not without its weaknesses. On Artificial Analysis’s Comprehensive Intelligence Index, K3 scores 57, ranking third globally, behind Fable 5 (60) and GPT-5.6 Sol (59). In areas such as general reasoning and multimodal understanding, K3 still has room for improvement. Huang Zhenxin did not avoid this point during the July 21 forum—“We are strong in programming, but our general capabilities are still catching up.” This honesty itself is a sign of confidence.
On July 27, K3’s full weight will be open-sourced. This will further amplify its ecosystem influence and increase pricing pressure on closed-source competitors.
K3 has been online for 48 hours, and its computing power has been overwhelmed. In the late hours of July 19, Moonshot made a counterintuitive decision: to suspend new user registrations.
“Better to give up new revenue than compromise the experience of paying customers,” Huang Zhenxin originally said. It’s uncommon in the large model industry to deliberately hit the brakes during the most successful phase of commercialization. Most companies choose to scale up capacity, add servers, or even sacrifice some service quality to sustain growth. Moonshot AI does the opposite.
There are two layers of signals behind this.
The first layer is the toughness of product philosophy. Huang Zhenxin repeatedly emphasized “values” at the forum—prioritizing the experience of paying customers over the growth of free users. This may sound abstract, but in the large model industry, end-user free customers are the easiest to inflate through marketing, while API-paying customers represent real revenue. Moonshot chose the latter, demonstrating a clear understanding of its core foundation.
The second layer is the harsh reality of computational capacity. A team of 300 manages a model with 2.8 trillion parameters while simultaneously supporting API calls from over 200 countries worldwide. This ratio is among the most extreme globally. Paid users overseas have grown by 400%, and API revenue has increased by approximately 400%—demand has surged so sharply that it has outpaced the rate at which supply can scale up.
Suspending registration is a “circuit breaker” mechanism designed to protect paying customers from fluctuations in service quality. But this bittersweet situation also reveals a deeper issue: if computing power remains a bottleneck, where is the ceiling for growth?
For large model companies, compute is not a cost issue—it’s a strategic one. Training requires GPUs, and inference requires them even more. While K3’s sparse architecture does reduce some inference costs, when user growth becomes exponential, those savings are quickly consumed by scale. Moonshot needs to resolve this before going public, or its quarterly financial results won’t look good.
If K3 is a technical story, ARR is the business story—and it’s the latter that capital markets truly want to hear.
The ARR growth curve of Moonshot AI is impressive even by global AI industry standards: it surpassed $100 million in March 2026, $200 million in May, and $300 million by mid-June. Tripling in three months is not linear growth—it’s a tipping-point explosion.
Source: Shanghai Securities News, Wall Street Journal (June 30, 2026)
More importantly, consider the structure: API revenue accounts for over 70%. What does this mean? It means Kimi isn’t buying user growth through consumer-side spending—it’s enterprise customers voting with real money. Developers, startups, and large corporations are integrating Kimi’s API into their own products and services, generating genuine usage volumes and demonstrated willingness to pay.
Overseas data is even more significant. Overseas paying users have grown by 400%, API revenue has increased by approximately 400%, and services now cover more than 200 countries. Within 20 days of the K2.5 release, revenue surpassed the entire year of 2025. Moonshot is transforming from a Chinese company into a global one—and its internationalization is product-driven, not marketing-driven: developers know immediately whether the API is good or not.
This path is highly similar to Anthropic’s. From the outset, Claude bet on APIs and enterprise customers, taking a relatively restrained approach to consumer-facing products. As a result, Anthropic achieved approximately $1 billion in ARR by the end of 2024, with a valuation of around $18 billion. Moonshot’s current ARR is about one-third of Anthropic’s at the same stage, yet its valuation has reached $31.5 billion—1.75 times Anthropic’s valuation at that time.
The capital market is assigning a premium to Moonshot AI. The underlying logic is this: if Kimi can prove that the “API-first” approach can succeed in China, it won’t be a follower of Anthropic—but a leader in a parallel赛道.
K3’s pricing strategy further supports this assessment. Priced the same as Sonnet 5, it indicates that Kimi’s target customers are global developers, not just those in China. Its goal is to compete with Anthropic and OpenAI, not to engage in price wars with domestic competitors.
The shift from “funding-driven” to “product-driven” holds profound significance for a Chinese large model company. Previously, all Chinese AI startups essentially told the same story: “I can build China’s version of ChatGPT.” Now, Kimi is telling a different story: “I can build an API that global developers will pay for.” These two stories have entirely different valuation models.
One detail is noteworthy: K3’s API pricing is identical to Sonnet 5’s, yet its cost is less than one-third. This means Moonshot holds significant room to lower prices—if competition intensifies, it can cut prices to gain market share while still maintaining healthy margins. If it keeps current pricing, each additional API call directly boosts profit. This flexible pricing power—the true moat that capital markets value most.
Why now?
Three windows are open simultaneously—miss them, and they won’t return. First, the ARR inflection point—$300 million in ARR is a psychological threshold; capital markets view this figure as a signal that the business model has been validated. Second, the K3 launch—with 2.8 trillion parameters and Code Arena reaching the top, it has delivered the strongest technical narrative for Moonshot in nearly a year. Third, competitor benchmarks—Zhipu is set to list on the Hong Kong Stock Exchange in January 2026, and MiniMax has already gone public; Hong Kong’s Chapter 18C provides a fast-track route for tech companies, and the precedent has already been set.
On July 19, Moonshot sent a listing proposal to investors, with the Hong Kong IPO expected to be completed within as little as six months. The company is currently dismantling its VIE and red-chip structures and has been maintaining close communication with investment banks such as CICC and Goldman Sachs. Moonshot has not responded to this—standard practice during the pre-IPO quiet period.
The valuation curve offers another perspective: $4.3 billion by end-2025, $10 billion in February 2026, $20 billion in May, $31.5 billion in July—sixfold growth in six months. Capital is rushing in with real money, yet the capacity of the private market is limited. After $31.5 billion, who will take the next round of funding? Rather than continuing to tell stories in the private market, it’s better to bring those narratives to the public market via Hong Kong’s stock exchange.
Source: Shanghai Science and Technology Daily, Every Day Economic News
Hong Kong Stock Exchange Chapter 18C is a tailored pathway. Introduced in 2023, this listing rule allows unprofitable tech companies to go public with lower barriers, specifically greenlighting “hard tech” firms. Zhipu has already paved this path, and MiniMax has followed. Yuezhai Anmian does not need to rebuild the road—only to follow it.
But what about after listing? Several hidden concerns cannot be ignored.
A team of 300 people. While this is lean for an AI startup, is it sufficient to handle the pressures of quarterly earnings reports post-IPO, compliance requirements, and investor relations? Anthropic had over 1,000 employees at the same time, and OpenAI had more than 5,000. Moonshot AI achieved nearly half of its competitors’ ARR with only one-third of the workforce—this reflects an efficiency advantage, but also carries management risks.
The computing power bottleneck has genuinely constrained growth. Suspending new user registrations is merely a symptom; the root cause is supply failing to keep up with demand. After going public, every quarter’s revenue growth will be scrutinized under a magnifying glass. If the pace of computing power expansion lags behind, the stock price will react immediately.
The double-edged sword of open-source models. Releasing the full weights of K3 benefits ecosystem expansion but poses challenges to commercialization. If enterprise customers can directly download and deploy the model themselves, why would they purchase your API? Moonshot AI needs to find a sustainable balance between an open ecosystem and commercial monetization.
Another less favorable reference: After Zhipu’s Hong Kong listing, its stock price dropped significantly from a high of HK$2,980. Market enthusiasm for AI-related stocks is limited, and valuations are slashed mercilessly whenever earnings fall short of expectations.
The competitive landscape is also evolving at a deeper level. Anthropic’s latest valuation stands at $965 billion, while OpenAI’s target valuation is $2–3 trillion—pushing the ceiling for top U.S. players into the trillion-dollar range. At $31.5 billion, Moonshot AI’s valuation is merely a midpoint on the global spectrum. To move higher, it must prove it is not just “China’s Anthropic,” but has its own differentiated path.
Going public in Hong Kong also involves a geopolitical variable. The uncertainty in Sino-U.S. relations has led to a real valuation discount for Chinese tech stocks. The significant price volatility of Zhipu’s stock after its listing largely reflects this discount. Moonshot’s choice of Hong Kong over the U.S. market is, to some extent, a pragmatic response to geopolitical pressures—the 18C chapter offers speed, but it also implies a valuation ceiling potentially lower than that of the U.S. market.
But conversely, if Kimi can generate a globally significant API revenue story within the Hong Kong market framework, it would demonstrate that Chinese large model companies can earn international capital recognition without relying on U.S. markets—this itself is an even stronger narrative.
Yang Zhilin’s shift in attitude is the most intriguing underlying thread in this story.
By the end of 2025, he still said, “There’s no rush to go public.” Seven months later, in July 2026, the listing proposal had already been sent to investors. What changed? Not his attitude—but the facts on the ground: ARR grew from under $100 million to $300 million, K3 moved from the lab to become number one globally, and the valuation surged from $4.3 billion to $31.5 billion.
The founder, born in 1992, has a resume steeped in legend: top of his class at Tsinghua University, earned a PhD from CMU in four years instead of the standard six, first author of Transformer-XL and XLNet, with over 22,000 paper citations, and co-authored papers with Turing Award winners Yoshua Bengio and Yann LeCun. In March 2023, he founded Moonshot AI, and in just two and a half years, brought the company to the brink of IPO.
He once said, “We shared this precious journey toward the truth, and we believe that this simple joy can help us navigate through cycles amidst the noise.”
Now, the cycle has arrived—but not the one he imagined. Capital cycles are shorter and harsher than technology cycles. Venture capitalists can wait five years; public markets wait only one quarter. K3’s rush to go public isn’t because Yang Zhilin changed; it’s because he knows the window won’t stay open forever. The current $300 million ARR is just a number—by the time of the IPO six months from now, the capital markets will be looking at the number six months ahead.
Going public isn’t the end—it’s the beginning of a new game. The $31.5 billion valuation from the capital market is both a check and an IOU—every dollar must be repaid with growth.
Can Kimi continue to make this story work after the IPO? The answer isn’t in the prospectus—it’s in the next quarterly earnings report.
(Source of technical data: Moonshot AI official, Artificial Analysis, Caixin; source of business data: Shanghai Securities News, Wall Street Journal China, Sci-Tech Daily, Economic Daily; source of listing information: Sci-Tech Daily, Bloomberg.)