11/16 2026
370

In the Double 11 shopping festival that just concluded, an unprecedented item appeared in shopping carts: tokens.
A host of major model vendors—including Zhipu, MiniMax, Qianwen Office, and Alibaba Cloud—all jumped into the fray, rolling out memberships and token packages at 30–50% off. In livestreams, AI memberships priced at one cent sold nearly 100,000 units in half an hour, and trillion-level token quotas were snapped up in an instant. What was once a billing unit confined to developers' backends is now sitting on e-commerce shelves right alongside video subscriptions and phone credit top-ups.
Behind the bustling promotions, a soul-searching question has been put on the table: Is this round of low-price frenzy in AI large models a prelude to AI truly reaching the masses, or just another internet-style cash-burning race to the bottom?
01 Computing Power Enters the Shopping Cart: Tokens Become a Double 11 New Product
A year ago, the vast majority of ordinary users had no idea what a token was. It is the smallest unit of large-model computation. In the past, only developers who logged into the console and configured API keys would encounter it, and ordinary consumers had no purchase entry point at all.
But this Double 11, tokens were placed on e-commerce shelves for the first time. Tmall launched a token recharge center, JD opened an AI token service station, and the recharge process was simplified to entering a phone number, paying, and binding an account—all done in half a minute. As a result, you saw:
Qianwen Office memberships dropped to as low as one cent for the first month, with nearly 100,000 units sold in 30 minutes in livestreams, corresponding to about 2 trillion tokens of computing quota; MiniMax and Zhipu monthly and weekly cards were 50% off, Alibaba Cloud token packages started at 50% off, and the monthly card for the AI programming tool Qoder was discounted to 33%—just 19.9 yuan.
Data shows that on the first day of Double 11 sales, transactions in Tmall's token category surged 450% month-on-month. This signals a landmark shift: AI computing power has officially transformed from a B-end developer tool into a commodity that ordinary users can directly stock up on and consume. Ordinary users no longer need to understand APIs; they can buy AI's "fuel" with money.
On the surface, AI is replicating the early path of cloud computing: lowering the threshold through price cuts to attract massive numbers of users to try it out. More and more workplace professionals and students are beginning to pay for AI to write documents, make summaries, and generate content, and user education for paid AI seems to have been completed overnight. But beneath the prosperous data, hidden risks are equally prominent.
02 Low Prices Can Buy Sales, But Not Retention
Most of the memberships that sold explosively for one cent in livestreams were limited to one order per new user. In essence, this is large-scale subsidized customer acquisition and does not mean users are willing to repurchase. A common reality in the industry is that a large number of low-price orders come from bargain hunters, and many users stop opening the product once the novelty wears off after claiming low-price memberships.
From my observation, a considerable portion of the free/low-price quota consumption of many large models comes from test users without real demand, who are just joining the excitement to experience it and have not formed stable usage habits. Promotions can quickly push up sales figures, but they struggle to solve the three core propositions of retention, paid conversion, and real usage scenarios.
More realistic pressure comes from costs. Large models belong to a heavy-asset industry, and inference, computing power, and servers all require continuous cash burn. The low-price and subsidy model is difficult to sustain over the long term. If there is only a simple competition over discounts and scale is exchanged through subsidies, it is easy to slide into the kind of race to the bottom seen in the past in the group-buying and food delivery wars: high orders, meager revenue, growth bought by burning cash, and user churn once subsidies recede.
At the same time, ordinary users have begun to learn to compare token package prices, choosing AI memberships just as they compare tissues and daily necessities. But the problem is that many ordinary consumers find it difficult to distinguish the actual capability gaps between different models and are easily swept along by low-price marketing, looking only at price while ignoring actual experience.

03 Two Sides of One Coin: Popularization and Race to the Bottom Happen Simultaneously
This round of Double 11 large-model promotions can neither be simply praised wholesale as the dawn of the industry nor directly denied wholesale as ineffective, self-destructive competition. In fact, popularization and consumption are two sides of the same coin.
The positive side lies in completing user education and accelerating the popularization of AI. The core value of a major promotion is not how many tokens are sold, but completing mass cognitive popularization. In the past, many people thought large models were esoteric technology. Now, through e-commerce livestreams and promotional events, tens of millions of users know that AI can be paid-subscribed and used for office efficiency, lowering the public psychological threshold and cultivating C-end payment habits.
At the same time, major promotions help force vendors to optimize C-end product experience, packaging complex API capabilities into membership packages that ordinary people can understand and afford, objectively pushing AI out of the tech circle.
The risky side lies in being vigilant against falling into meaningless price wars. If competition remains only at the level of discounts and subsidies, without polishing products and implementing real usage scenarios, it is purely self-destructive consumption. AI's moat is not cheaper tokens, but whether it can solve users' real pain points: office efficiency, content creation, and business processing.
When the price war erases profit margins, companies do not have enough funds to invest in model iteration and product polishing, and ultimately what suffers is still user experience. Therefore, the industry should form a consensus as soon as possible: the pure price war has reached its end, and the next step is to shift from competing on low prices to competing on value and on scenario implementation capability.
04 After Subsidies Recede, AI Commercialization Ushers in the Real Major Test
The bustling Double 11 is only a traffic-generation exercise. The real major test comes after the promotion ends, and it requires directly facing three realistic challenges:
① After subsidies fade, which users are willing to pay out of their own pockets to continue using it? Distinguish bargain-hunting traffic from paying users with real demand;
② Can the product match the price? After low-price traffic generation, model stability, generation quality, and scenario functions must keep up, and it cannot rely only on discounts to attract users;
③ Break out of C-end membership price wars; B-end scenarios are a larger battlefield. In catering, supply chains, government affairs, and enterprise office, the real value of large models lies in solving the actual problems of business operations, not just selling individual users membership packages.
Conclusion
Double 11 pushing tokens onto consumer shelves is a noteworthy milestone in the commercialization of domestic large models. It brings AI into the view of more ordinary people, but low prices are only a knock on the door, not the final answer. Clearly, the AI industry will not mature through discounts. After the carnival recedes, what truly remains will never be the cheapest computing power, but products that can truly help users solve problems. What do you think?