On the evening of September 2, Moonshot AI — the company behind Kimi — confidentially filed its A1 listing application with the Hong Kong Stock Exchange, becoming the second of China's "AI six little tigers" to enter the Hong Kong IPO process. CICC and Goldman Sachs are the joint sponsors. The company's only public response: "No comment."
Most people read this as good news: another Chinese large-model company is going public, and the capitalization of AI assets has taken another step forward.
But reading it more closely reveals another layer — this is a positioning battle: keep pace with the peers that have already listed in terms of funding capability, and get ahead of the rivals still behind, so as not to be left out.
1. For large-model companies, going public is competitive positioning, not the finish line
Compared with traditional companies, listing is usually the endpoint of "making it." But for a large-model company, listing is a competitive checkpoint: a company must stay on the same level as its peers across two tracks — compute and funding — and must not fall behind.
Why? Because competition among large models is essentially a dual arms race of "compute + capital." Every order of magnitude in model parameters pushes training cost, inference cost, and compute procurement cost up exponentially. Whoever keeps raising money and buying compute stays in the first tier.
A telling signal lies in K3.
On July 17, Moonshot released K3 — a 2.8-trillion-parameter ultra-sparse MoE model that topped the Frontend Code Arena leaderboard, the first time a Chinese large model surpassed OpenAI and Anthropic on that benchmark. Yet just 48 hours after launch, the company paused new user subscriptions, prioritizing limited compute for paying users; API pricing jumped from $4 per million tokens for K2.6 to $15.
The stronger the technology, the more compute it consumes, and expansion gets squeezed to the limit by funding and supply. At this point, whoever opens a secondary-market funding pipeline faster can fill the compute gap sooner and hold onto a leading position.
2. Tenfold in nine months: keep pace with the funding rhythm, don't fall behind
Look again at the valuation line, and you'll see a story more real than "racing ahead."
At the end of 2025, public reports put Moonshot AI's valuation at about $4.3 billion, and founder Yang Zhilin said in an all-hands letter that the company was "in no rush to go public." Nine months later, it filed confidentially, with a pre-IPO valuation of $50 billion — a tenfold rise.
On the surface it's market enthusiasm; underneath, it's a race for funding rhythm: in July it had just closed a $3.5 billion Series F (over three times oversubscribed and closed early), then immediately filed confidentially — the funding close and the filing only days apart.
Why is funding and listing squeezed so tightly together? Because peers already got ahead — Zhipu and MiniMax listed in Hong Kong via the 18C channel early this year, completing the "index inclusion, Stock Connect inclusion, listing" path and cashing in on secondary-market capital first. If Moonshot fails to keep pace, it will gradually be left behind in funding capability and valuation level.
At a $300 million ARR, a $50 billion valuation implies a static multiple of over 160x. That pricing itself is also a "positioning" move — locking in the market's perception of its technological lead to claim a higher perch for future funding, talent, and compute competition.
In other words: listing pulls the "funding capability" gap between it and its peers back onto the same starting line.
3. Two different financial models in China and the US: why Chinese large models need funding more urgently
To understand why Moonshot "must accelerate its listing," look first at a more fundamental difference — Chinese and American large-model companies run on two entirely different financial models.
In terms of model capability, China and the US are already level, even ahead in places: Kimi K3 topped Frontend Code Arena with 1,679 points, pushing OpenAI and Anthropic behind for the first time; DeepSeek and Zhipu GLM also sit in the first tier of major leaderboards. In other words, on the top few models, China's technological gap has narrowed considerably.
But on valuation, the gap is still an order of magnitude. Zhipu's market cap is under $63 billion, and Moonshot's pre-IPO valuation is $50 billion; OpenAI is valued at $852 billion, and Anthropic surged to $965 billion at the end of May, taking the crown as the world's most valuable AI company. $50 billion against $850 billion is a gap of more than 17x.
Capability nearly level on one side, valuation 17 to 19 times apart on the other — this "scissor gap" is the key yardstick for understanding how Chinese AI assets are priced in this cycle. It explains both why Moonshot must hurry to list and re-price itself in the secondary market, and why American leaders' valuations carry a large "front-runner premium." Whoever converges this scissor gap first takes the initiative in the next round of pricing.
The US runs on a "closed-source + premium pricing" path. OpenAI, Anthropic, and Google keep their models mostly closed-source, charging high prices through subscriptions and APIs, earning back the cost of inference and R&D directly from revenue. This path has already produced scale: OpenAI's annualized revenue hit about $24 billion (roughly $2 billion a month) with about 900 million weekly active users on ChatGPT; Anthropic's annualized revenue broke $30 billion in April 2026, overtaking OpenAI, and it is expected to post about $560 million in Q2 operating profit — genuinely turning a profit.
Of course, that doesn't mean American companies are flush with cash. OpenAI posted about $13 billion in 2025 revenue but still over $20 billion in operating loss, and raised $122 billion this March (at an $852 billion valuation). But the key difference is this: it can already cover a meaningful portion of its compute spending with revenue — funding is an "accelerator," not its "only water source." The market's understanding of their financial model is already very clear.
China, by contrast and based on reality, runs on an "open-source + low price" path. DeepSeek, Alibaba's Qwen, and the models various players have open-sourced one after another have driven domestic API prices to an extremely low level, plunging the industry into a price war. Open source and low prices buy ecosystem and visibility, but they also essentially shut the door on "making money by charging high prices."
This gap shows directly in the revenue scale: Moonshot's ARR only broke $300 million in June, Zhipu's August ARR is about $1.6 billion, and MiniMax about $800 million — nearly two orders of magnitude behind the $24 billion and $30 billion annualized revenue of their American peers. For Moonshot, this gap is decisive. A $50 billion valuation against a $300 million ARR is a static multiple of over 160x — the money it earns from its own operations falls far short of covering the cost of continuing to train the K series and procure compute. It's not a matter of "whether it wants to live on revenue," but "revenue can't sustain it for now."
So Chinese large-model companies are destined to rely on external funding more than their American peers. Listing, for them, is not the icing on the cake — it is moving the lifeline of "funding" from the primary market to a broader, more stable secondary market. This also explains why Moonshot must complete its listing at this moment and ahead of these rivals.
4. The earlier it lists, the more it must get ahead of rivals
This judgment is corroborated from three directions.
The first is the reality of the compute gap. Pausing new subscriptions exposes inference compute failing to keep up with growth. Large-model revenue is not zero marginal cost — the stronger the model, the more calls, the higher the inference cost. Quadrupling K3's API price partly passes on compute cost, but it cannot fully cover the resource consumption of new users. The larger a company's compute gap, the faster and wider a funding pipeline it needs to fill it.
The second is competition in time. Zhipu's first batch of locked-up shares unlocks in January 2027; once concentrated selling occurs, the overall valuation of the AI sector may be suppressed, and Moonshot needs to complete its listing before then. DeepSeek is rumored to start its listing in the first half of 2026, and overseas peers are also advancing — Moonshot must get ahead of them to avoid a "head-on collision" in the secondary market that dilutes attention and capital.
The third is the homework it is doing itself. In May, the company dismantled its red-chip VIE structure, dissolved its Cayman entity, switched to an onshore equity structure, and had to fold the interests of early overseas investors back in — this is the required question on the 18C listing exam, and it must be passed no matter how hard, because the listing window waits for no one.
Three facts point to the same conclusion: the positioning of "keeping pace with peers and getting ahead of rivals" is the key to securing funding and keeping a seat at the table.
5. The cost of positioning: putting fuzzy accounts on the table
Positioning has its costs.
The most fatal one is the metric. Large-model companies' ARR still uses inconsistent definitions — Zhipu's first-half revenue was RMB 954 million, yet its August ARR was reported at about $1.6 billion; MiniMax's first-half revenue was about $117 million, yet its August ARR exceeded $800 million. Some count "subscription recurring revenue," while others simply annualize the most recent week or month. Cross-comparison almost loses meaning.
Listing means these numbers will be put under the microscope of an audit. K3 has proven a technological generational gap, but whether that gap can translate into auditable revenue and profit has not yet been verified. The $50 billion valuation buys "capitalization within the window of technological leadership," not a financial report that has already been delivered.
In other words, it uses secondary-market funding to buy the chips to keep up with competition, at the cost of laying bare everything that could be blurred in the primary-market era. The sooner it lists, the sooner this account comes due.
6. What to make of it
Finally, a judgment from an industry and investment perspective: what's really worth noting about Moonshot's filing is not that "another large-model company has arrived," but the competitive law it reveals:
For large-model companies, going public is essentially a positioning battle — using secondary-market funding capability to keep pace with peers and complete capitalization ahead of rivals, ensuring it isn't left behind in the compute-and-capital arms race.
Under the two different financial models of China and the US, this logic is especially sharp in the Chinese market: American peers can rely on "closed-source + premium pricing" to build revenue and gradually cover compute, while Chinese open-source players, for now, can only rely on funding to fill the gap. The earlier and more urgently a company lists, the more it shows that competition in this track waits for no one.
For this industry, this is both a sign of prosperity and a footnote to white-hot competition. The one who has the last laugh is not necessarily the fastest, but the one who, while maintaining its lead in funding and compute, ultimately aligns the speed of burning cash with the speed of making money.
The technology window is counting down, peers' listing rhythm is counting down, and rivals' entry is counting down. When three countdowns ring at once, listing becomes a race you cannot afford to lose.
This article does not constitute investment advice.