To The Moon

On September 2, 2026, Z.AI (02513.HK), hailed as the "world's first large model stock," quietly launched an official flagship store on Tmall. The only products listed are four tiers of GLM Coding Plan Token packages at varying price points. Shoppers can now add AI computing power to their carts and check out with the same ease as topping up a video streaming subscription.

This is no simple channel expansion. When an AI company worth over HK$500 billion packages its core API calling capabilities as standardized goods on an e-commerce shelf, it signals a pivotal shift in the commercialization path of the large model industry—from "selling models" to "selling Tokens," from project-based to subscription-based, and from B2B customization to C2C retail.

From "Selling Projects" to "Selling Tokens": A Dramatic Reshuffling of Revenue Structure

To understand why Z.AI turned to Tmall, one must first examine its freshly released semi-annual report. In the first half of 2026, Z.AI generated revenue of RMB 954 million, a 399.7% year-on-year surge, with half-year results already surpassing its full-year 2025 figures. But the real headline is the drastic revenue mix transformation: open platform and API business revenue hit RMB 825 million, an explosive 2,735.7% increase, jumping from 15.2% of total revenue a year ago to 86.5% now. Meanwhile, localized deployment revenue fell from RMB 148 million to RMB 67.04 million, down 54.6% year-on-year.

Behind these numbers lies a complete business model switch. A year ago, Z.AI was a government and enterprise project-based company that earned revenue by deploying models into client data centers and recognizing income per contract. Today, it operates as a Token economy firm that places models on the cloud and charges continuously based on usage volume. This transformation is underpinned by staggering user growth: by the end of August, Z.AI's MaaS platform had surpassed 7.4 million users, up 144% from the start of the year, with paid daily active users growing 603% and Token call volumes soaring over 40-fold. Notably, despite the surge in volume, the average API selling price actually increased by roughly 101%. With both volume and price rising, the API business gross margin swung from -0.4% to positive 24.6%.

During the earnings call, Z.AI's management unveiled a formula: "AGI commercial value = intelligence ceiling × Token consumption scale." In plain terms, the stronger the model, the more complex tasks it can handle, making each Token more valuable; the more calls made, the greater the total revenue. Launching on Tmall is the natural extension of this formula on the sales side—lowering the barrier for developers, programmers, and even individual users to become Token consumers.

But the flip side is equally stark. While revenue nearly quadrupled, Z.AI still recorded a net loss of RMB 2.072 billion in the first half. Overall gross margin plunged from 50% to 26.4%, and sales costs skyrocketed 635.4% year-on-year. Compounding market concerns, Z.AI's share price closed at HK$1,098 on September 3, down over 60% from its June peak of HK$2,980, erasing more than HK$500 billion in market value. The day after the earnings release, the stock fell 1.34% instead of rallying.

Explosive growth, persistent losses, and a pressured stock price—this is the real context behind Z.AI selling Tokens on Tmall. At a time when the capital market is voting with its feet, any move that can boost revenue and improve unit economics is worth attempting. The Tmall flagship store offers three key commercialization levers.

First, it lowers customer acquisition costs. Compared to organic website traffic, Tmall provides faster access to high-frequency users like independent programmers and developers. Within 48 hours of opening, brand searches surged 50-fold month-on-month, driving a 160% jump in AI Token subscription product transactions across Taobao and Tmall. While the absolute initial order volume may be limited, the amplification effect of Tmall's traffic cannot be underestimated.

Second, it reinforces subscription mindshare. By turning Token packages into standardized retail goods akin to phone plans, users can activate services directly in their Z.AI accounts after purchase. Subscriptions naturally align with e-commerce—auto-renewal, promotional discounts, and price comparisons. In the future, users may buy Tokens as easily as topping up their phone bill, even enjoying "Double 11 AI Service Sub-venue" discounts.

Third, it creates a scenario for price increases. The Coding Plan packages listed on Tmall have seen substantial price hikes from earlier this year: the Lite version rose from RMB 49 to RMB 118, the Pro from RMB 149 to RMB 538 (a 3.6-fold increase), and the Max from RMB 469 to RMB 1,078. Z.AI justifies this by citing the new points-based system, 50% point offsets during off-peak hours, and the model upgrade from GLM-5 to GLM-5.3. Undeniably, the standardized display and comparison logic of e-commerce provide a relatively moderate environment for transparent price adjustments.

Tokens on Retail Shelves: The "iPhone Moment" for Large Model Commercialization or Just Marketing Hype?

On the day after Z.AI's store launch, Tmall officially introduced the "AI Space Station" (Token top-up center) on September 3, initially integrating leading large model providers including Alibaba Cloud, Z.AI, Kimi, MiniMax, and DeepSeek. Overnight, buying Tokens became as routine as recharging a phone plan. This evokes the famous 2007 iPhone launch line: "Apple reinvented the phone." So, does putting Tokens on e-commerce shelves herald an "iPhone moment" for the large model industry?

Optimists see a paradigm revolution. Tokens are evolving from mere measurement units in technical documents into purchasable, priced, and tradable digital goods. Chen Xiaohua, Executive Dean of the Education Science Academy under the China Mobile Communications Federation, notes that Tmall's value lies in leveraging mature e-commerce transaction and traffic systems to lower AI consumption barriers, pushing the large model business model from "model competition" toward "Token consumption and ecosystem competition." Senior telecom analyst Ma Jihua observes that Alibaba has formed a relatively complete business loop—Alibaba Cloud provides computing power, Qianwen offers intelligent services, and Tmall handles Token subscription distribution.

Industry data suggests the Token economy is indeed exploding. Domestic daily Token consumption has exceeded 140 trillion, a more than 1,000-fold increase from 100 billion in early 2024. Zhongtai Securities research indicates that the Token economy is comprehensively reshaping cloud computing and AI infrastructure, shifting computing power competition toward per-watt Token efficiency and business models from subscriptions to pay-as-you-go metering.

Yet skeptics see a different picture. First, the ceiling for C-end Token payments may be far lower than expected. One analysis bluntly states that Z.AI's Tmall store "selling Tokens feels more like advertising." The reasoning is straightforward: AI applications are tools, not pastimes, putting them at a natural disadvantage in the attention economy. Even ChatGPT, a phenomenon-level product, boasts a paid penetration rate of under 5%. Twelve hours after Z.AI's flagship store launched, not a single package had been sold. Programmers buying Coding Plans and white-collar workers purchasing Workbuddy credits are essentially "smallest production units"—Tokens should be borne by companies, not individual pockets.

Second, the symbolic value of e-commerce channels far outweighs short-term sales volume. Reports have acknowledged that "the symbolic value of this 'store opening' far exceeds short-term sales." Z.AI Chairman Liu Debing stated at the earnings briefing that the first half's keyword was "capability realization." The Tmall launch seems more like telling the market, "We are aggressively expanding our API business." Amid the capital market's repricing of Z.AI's valuation, a high-frequency C-end narrative offers more imaginative upside than a purely B-end API platform story.

The deeper question lies in the essence of the Token business. Barclays' AI industry research reveals a harsh statistic: for every US$100 in revenue model companies earn, roughly US$35-40 flows to the three major cloud giants in the form of inference computing costs. In other words, large model vendors selling Tokens are earning a "hard-earned wage"—paying upstream for cloud computing, facing fierce price wars downstream, and bearing massive R&D expenses in between. Z.AI is not blind to this. The company reduced per-Token inference costs by about 80% in the first half and has achieved large-scale, low-cost inference on 100,000-class domestic chips. GLM-5.3-Flash marks the first time Z.AI has fully deployed domestic chip clusters in real-world, ultra-large-scale traffic, with per-Token domestic chip inference costs now comparable to mainstream imported cards. Even so, R&D spending still reached RMB 2.13 billion, up 33% year-on-year. The more Tokens sold, the greater the computing cost pressure—a paradox of diseconomies of scale.

So what exactly is Z.AI after with its Tmall move? The answer may lie in a set of figures: as of August 31, Z.AI's MaaS platform annualized recurring revenue (ARR), calculated on a monthly basis, reached US$1.6 billion, and on a weekly basis, US$2 billion. The company has raised its year-end ARR target to US$2.4 billion. From government projects to cloud APIs to subscription packages on Tmall, Z.AI's commercialization path is steadily extending toward end users. The Tmall store is essentially "channel sinking"—converting API calling capabilities from enterprise exclusives into digital goods anyone can order. It doesn't chase short-term sales spikes; it is laying infrastructure for the long-term scale-up of the Token economy.

As Z.AI states in its own earnings report: "Each time the model completes a capability leap, what customers buy moves one step closer to results, and the company's revenue structure is rewritten accordingly." When Tokens become everyday consumer goods like phone bills, and when large model calls are metered like water and electricity, AI will truly complete the risky leap from "technology" to "commodity." Z.AI's Tmall debut is exactly that step. Whether it leads to a trillion-dollar market cap or merely serves as a stopgap amid commercialization anxiety, time will tell. What is certain is that the era of large model companies "seriously making money" has irreversibly begun.

This article was compiled with the assistance of AI tools for market data collection and industry information gathering, combined with supplementary analysis and writing.