2026 AI Listing Wave: Amid the Carnival of Trillion-Dollar Valuations, the Commercialization Test Has Only Just Begun
If January 8, 2026 marks the "coming-of-age ceremony" of China's AI industry, then a new record was born on that day — Zhipu AI was listed on the Main Board of the Hong Kong Stock Exchange under the stock code 2513, becoming the world's "first large model stock".
One day later, on January 9, another large model company MiniMax followed suit to get listed, with the stock code 00100. In just 24 hours, two Chinese AI unicorns successively rang the listing bell on the Hong Kong Stock Exchange. Among them, by June, Zhipu's share price surged intraday, and its market cap once exceeded HK$1 trillion — from an IPO market cap of HK$51.1 billion to the trillion mark, its market cap closed at HK$557.3 billion as of September 1, which is still far higher than the HK$57.9 billion on the listing day.
This is the first year of capitalization belonging to Chinese AI. While Zhipu and MiniMax have already "reached the shore", Moonshot AI (Kimi) is rushing to the Hong Kong stock market at a rocket speed with its valuation skyrocketing seven times in half a year, while DeepSeek plays a thought-provoking opposite role — Liang Wenfeng, the richest AI billionaire in the world, holds an estimated valuation of about 400 billion yuan, and is still preparing for its listing at its own pace so far.
Zhipu: HK$557.3 Billion Market Cap, RMB 950 Million Revenue, RMB 2 Billion Loss
Zhipu is the first enterprise among the "Six Little Tigers" of China's large model sector to launch an IPO, and is also one of the two listed core large model companies at present (the other one is MiniMax).
Its listing path is clean and efficient: it passed the hearing of the Hong Kong Stock Exchange in December 2025, issued shares at HK$116.2 per share on January 8, 2026, raising about HK$43.48 billion, with an IPO market cap of about HK$51.1 billion to HK$57.9 billion; it closed at HK$131.5 on the first day of listing, up 13.17%. After that, its share price kept rising, the intraday market cap exceeded HK$1 trillion in June, and fell back to HK$557.3 billion as of September 1.
In terms of finance, Zhipu just released its semi-annual report. In the first half of 2026, the company achieved operating revenue of RMB 954 million, a year-on-year increase of 399.7% — the revenue in only half a year has exceeded the total RMB 724 million of 2025. Among them, the revenue from the MaaS open platform and API business reached RMB 825 million, a year-on-year surge of 2735.7%, and its proportion in total revenue jumped from 26.3% at the end of 2025 to 86.5%; while the previously dominant localized deployment revenue dropped to RMB 67.04 million. This structural reversal marks that Zhipu has officially switched from the "project selling" mode to the MaaS mode of "calling service selling".
The profitability side is improving. The net loss for the period in the first half of the year was RMB 2.072 billion, narrowing by 12.1% year on year; the adjusted net loss was RMB 1.964 billion. The gross profit margin of the core open platform API business has turned positive to 24.6% from -0.4% in the same period of last year. After the release of GLM-5 in February 2026, Zhipu carried out structural price increase for some APIs, the average selling price of APIs in the first half of the year increased by about 101% compared with the same period of last year, while the unit Token inference cost decreased by about 80% compared with the beginning of the year. With both volume and price rising, the Token call volume of the MaaS platform increased by more than 40 times compared with the beginning of the year, and the number of paying daily active users increased by 603%. As of August 31, the annualized revenue (ARR) of the company's MaaS platform reached USD 1.6 billion calculated on a monthly basis, and USD 2 billion calculated on a weekly basis.
R&D investment is still increasing. The R&D expenditure in the first half of the year was RMB 2.131 billion, a year-on-year increase of 33.6%. Although its proportion in revenue decreased due to the rapid revenue growth, the absolute value continued to expand. Zhang Peng, CEO of Zhipu, reiterated the core logic of "strong model + high pricing power + rigid demand from government and enterprises" at the performance meeting, and pointed out that the improvement of API pricing power did not affect the call volume, but verified the value of technical barriers instead.
Just five months after listing, Zhipu announced on June 1, 2026 that it had applied for listing on the STAR Market of the A-share market, planning to raise RMB 150 billion. The A+H dual capital platform has been launched. JPMorgan Chase recently predicted that with the API gross profit margin turning positive and the release of scale effect, Zhipu is expected to turn a profit in 2029-2030 — and this semi-annual report obviously makes it possible to advance this schedule.
Moonshot AI: The Rocket with Valuation Septupled in Half a Year
If Zhipu is described as having "already reached the shore", then Moonshot AI (Kimi) is in the middle of "sprinting for listing".
Founded in April 2023 by Yang Zhilin, this company's founder is one of the core inventors of Transformer-XL and XLNet, known as the "post-90s AI genius" in the industry. Yang Zhilin holds 51.83% of the shares personally and is the controlling shareholder of the company. Before 2026, the valuation of Moonshot AI was only USD 4.3 billion; entering 2026, it raised funds almost every month, and its valuation soared all the way: USD 10 billion in January, USD 18 billion in February, USD 20 billion in May, and USD 31.5 billion in June. After completing the over USD 3.5 billion Series F financing on July 29, the post-money valuation of the company has reached USD 35 billion. The target pre-money valuation of the Pre-IPO round originally scheduled to start in August further rose to USD 50 billion — from USD 4.3 billion to USD 50 billion in less than a year.
What pushes the valuation to rise like a rocket is Kimi K3. Released in July this year, this model has a total parameter of 2.8 trillion, 1 million token context, native visual understanding support, and adopts the MoE architecture (only 16 experts are activated each time among 896 experts), making it the open source model with the largest parameter scale in the world. In the Frontend Code Arena front-end code evaluation, K3 topped the global ranking with 1679 points, surpassing Claude Fable 5; it ranked third globally in the Artificial Analysis comprehensive intelligence index, second only to Anthropic's Claude Fable 5 and OpenAI's GPT-5.6.
The company's commercialization is also accelerating: as of June 2026, the ARR (Annual Recurring Revenue) of Moonshot AI has steadily reached USD 300 million, breaking USD 100 million for the first time in March, exceeding USD 200 million in April, and reaching USD 300 million in June, of which API revenue accounts for more than 70%. According to Bloomberg, the company may land on the Hong Kong capital market as soon as within 6 months, and has sent listing proposals to investors.
The other side of this growth curve is the sentence in Yang Zhilin's internal letter at the end of 2025 that "we are not in a hurry to go public in the short term, nor do we take listing as our goal" — only seven months later, the release of Kimi K3 completely changed the pace. From "not in a hurry" to sprinting for IPO, Moonshot AI only spent the time of one technological breakthrough.
MiniMax: The Second One to Land, Valuation Roller Coaster
MiniMax is the second core large model company listed on the Hong Kong Stock Exchange after Zhipu, but it has taken a more turbulent capital curve.
It was listed on January 9, 2026, with an issue price range of HK$151 to HK$165. On the first day of listing, it surged 109.09% to close at HK$345, and its market cap quickly exceeded HK$100 billion. By May 29, its share price rose 4.09 times compared with the issue price, with a market cap of HK$263.5 billion; after being included in the Stock Connect on August 6, its market cap fluctuated in the range of HK$100 billion.
In sharp contrast to its high market cap is its meager revenue. In the whole year of 2025, MiniMax's revenue was USD 79.038 million (about RMB 543 million), a year-on-year increase of 158.9%; the adjusted net loss was USD 250 million (about RMB 1.75 billion). If the change in fair value of financial liabilities is included, the net profit attributable to shareholders is -RMB 12.76 billion, and after excluding this impact, it is -RMB 1.921 billion. The founder and CEO Yan Junjie (born in 1989, postdoctoral fellow of Tsinghua University, former vice president of SenseTime) holds more than 20% of the shares; Alibaba is the second largest shareholder with a 12.52% stake.
The unique value of MiniMax lies in globalization: more than 70% of its revenue comes from overseas, its products cover more than 200 countries, and it has about 300 million global users. At the end of May, the company signed a coaching agreement with CITIC Securities to officially launch its A-share IPO — after Zhipu, the "two leading large model companies" will meet again on the A-share market.
DeepSeek: The Last Entrant Giant
Among all Chinese AI companies rushing to the capital market, DeepSeek is the one that started IPO preparation the latest, but also the most valuable one.
The three principles of "no financing, no listing, no commercialization" set by Liang Wenfeng in 2023 were broken one by one in 2026: in June, it completed the first round of external financing of about RMB 500 billion, with a post-money valuation of about USD 52 billion (about RMB 3.5 trillion); entering August, the company is advancing the second round of financing with a pre-money valuation of about RMB 5 trillion, planning to raise RMB 500 billion.
In terms of finance, according to a report by The Information on August 26 citing people familiar with the matter, DeepSeek achieved revenue of about RMB 475 million in the first 7 months of 2026, which is about 10 times the total revenue of 2025; the net loss in the same period narrowed to RMB 715 million, lower than the total RMB 935 million of 2025. The overall gross profit margin is 44.6%, of which the gross profit margin of API business is as high as 82.9%, which mainly benefits from the significant reduction of inference cost brought by AI infrastructure optimization. According to previous reports, the company's annualized revenue is close to the range of USD 400 million to USD 500 million. In terms of expenditure, the AI infrastructure investment in the first 7 months of 2026 was about RMB 110 billion, mainly used for renting AI servers, purchasing chips and other computing power equipment, while the total expenditure of this item in 2025 was only about RMB 12 billion.
In terms of IPO, DeepSeek was exposed in July to have launched the preparation for A-share listing, planning to submit the listing application as soon as the end of 2026, and target to be listed in 2027. However, as of the end of August, there is no record of DeepSeek's listing application, and the company's official has never officially announced the listing plan.
While Zhipu and MiniMax have been listed, and Moonshot AI is sprinting for listing, DeepSeek chooses to hold the initiative in its own hands with absolute controlling stake (about 78%) and a cautious pace. Liang Wenfeng's transformation from "preparation to wait-and-see" and Yang Zhilin's "rocket sprint" constitute two completely different capital philosophies at the same gaming table.
The Gap Between Valuation and Revenue is the Underlying Tone of the Whole Carnival
Putting the figures of several companies together, a clear industry commonality emerges: high growth, high investment, extremely high valuation, and no profit yet.
Zhipu: RMB 954 million revenue, RMB 2 billion loss (first half of 2026), corresponding to a market cap of about HK$557.3 billion;
MiniMax: RMB 543 million revenue, RMB 1.75 billion loss (2025), corresponding to a market cap of hundreds of billions of Hong Kong dollars;
Moonshot AI: USD 300 million ARR, corresponding to a valuation of about USD 50 billion;
DeepSeek: about RMB 475 million revenue, RMB 715 million loss (first 7 months of 2026), corresponding to a pre-money valuation of about RMB 5 trillion (for the second round of financing).
The price-to-sales ratio is so high that it cannot be explained by the traditional framework. Is it the comprehensive pricing of the market for "technical barriers, growth slope, globalization space, and ecological discourse power", or a bubble destined to return to the fundamentals? Regulators and investment banks are answering with different calibers. JPMorgan Chase predicts that Zhipu is expected to turn a profit in 2029-2030; Yan Junjie, founder of MiniMax, calls 2026 a year of "very large leapfrog development" for the industry.
The deeper pressure comes from two ends: on one end, giants including Alibaba, Tencent, ByteDance suppress comprehensively with traffic, computing power and capital; on the other end, open source models (DeepSeek, Qwen, Kimi K3) continuously reduce the inference cost and erode the pricing moat of closed-source models. In the first half of 2026, the total financing amount of China's AI track has exceeded RMB 3 trillion, of which the large model sector alone accounts for about RMB 159.8 billion — capital is still pouring in, but the exam for commercialization closed loop has just begun.
In the story of the United States, the IPOs of Anthropic and OpenAI answer the question "Is it a new king sought after by the market, or a valuation illusion in the private equity circle?"; in the story of China, the same question is placed in front of all players — what exactly does HK$1 trillion buy, a ticket to AGI, or a forward check that takes several years to cash?
2026 is just the first page of this open-book exam.
This article is from the WeChat official account "Tech Business", author: Tech Business, published with authorization from 36Kr.