AI starts billing by the degree: 48 hours of Tokens placed on Double 11 shelves

At Double 11 2026, the shelves carried a new product that had never been there before: Tokens. On the evening of November 2, Li Jiaqi's livestream began selling Qwen Office Token monthly cards; 100,000 were sold in half an hour, totaling about 2 trillion Tokens. Almost the same night, the JD Double 11 livestream listed Token packages from more than ten major model providers including JD Cloud, DeepSeek, Z.AI, and Volcano Seed. In the first hour of the broadcast, over 1 billion Tokens were given away; in two hours, sales rose 20-fold from the previous period, and searches for "Token" rose 7-fold from the previous period. At Taobao's recharge center, Token now sits alongside phone bills and utilities in the same entry point. A unit of measurement that once belonged only to developer backends broke into mass-consumer shelves within 48 hours. This is not a simple marketing event but a signal: AI is moving from a "model race" into an "infrastructure race." Token is AI's unit of accountevery piece of text or image a model generates is measured in Tokens. Over the past two years, it was just a cold number in model vendors' API backends: Kimi charged RMB13 per million input Tokens and RMB68 per million output Tokens; Meta charged US$4 to US$20 per million Tokens; of Anthropic's US$16.286 billion in revenue in the first half, consumption-based income billed by Token accounted for 80%. Now the "selling location" of Tokens has changedfrom developer backends to livestream shopping carts. When a unit of measurement can be bought directly by ordinary consumers, it is no longer a technical metric but the smallest unit of a business. Explosive demand-side growth supports this shift. Figures from the National Data Administration and the CCID Research Institute show that nationwide average daily token calls rose from the hundreds of billions level in early 2024 to over 175 trillion by June 2026. CICC's "AI Ten-Year Outlook IIPersonal Agent," published on November 4, estimates that under high-frequency usage, a personal agent consumes about 360 million Tokens per person per month, with hardware costs of about US$23 per month; by 2030, China's Personal Agent market is expected to exceed RMB510 billion. A QuestMobile report the same day showed that Doubao's average monthly usage per person has reached 72 times. User habits have formed, and AI's "electricity consumption" is taking off. The commoditization of Token benefits more than just those who sell Tokens. Along the chain of "productionmass productiontransmissiontrading," clear positioned players have emerged. The production end is the model vendors. Alibaba's (09988) Qwen Office is the protagonist of this round of Token commoditization; the RMB0.01 monthly card is essentially using Tokens as a customer-acquisition tool, with the goal of keeping users inside Qwen's model ecosystem. For the big tech companies, Token is a traffic gateway. At the mass-production end, dedicated "token factories" have appeared. The Beijing No. 1 Token Factory built by iSoftStone Information Technology (301236.SZ) has been put into operation, producing 1.4 trillion Tokens per day; the token factory of Shijiazhuang ChangShan BeiMing Technology (000158.SZ) connected more than 100 customers within just over ten days of going online. Token factories package models, computing power, and inference systems and charge by Token outputToken has gone from a "model accessory" to a commodity that can be mass-produced independently. The transmission end is the cloud vendors. Established cloud service providers such as KINGSOFT CLOUD (03896) are responsible for reliably delivering Tokens to customers: it has more than 100,000 servers in stock, an EBITDA margin of about 32%, and a validated customer structure and delivery capability. The larger the Token call volume, the steadier the flow through this "pipeline." The scheduling and trading end is the layer that operates Tokens as a "standard commodity." GBA AI COMP (01396) is positioned at these two links: its self-developed "Quantum Pai" computing-power cloud platform supports unified scheduling of computing power across regions, chips, and cloud environments, solving the "production and transmission" of Tokens; at the same time, as the operator of the Guangdong Token Trading Center, it participates in Token trading and services, solving the "measurement and trading" of Tokens; and the tens-of-billions-level token production base it is building in Hohhot uses green power to lower Token production costs. On the operations side, its computing-resource rack rate has stayed above 95%; more than 95% of new orders signed this year are three-to-five-year long-term contracts, and over 95% carry prepayments. Put the four links together, and you get a complete Token commodity chain: model vendors set prices, token factories mass-produce, cloud vendors transmit, and operators organize trading. What Li Jiaqi and JD's livestreams sell is the front end; what the later links capture is the call volume of every day that follows. Looking at these developments together, the logic chain is clear: Tokens are put on shelves call volume is activated consumption continues to expand the value of each linkproduction, scheduling, tradingis repriced. Within 48 hours, Li Jiaqi sold 2 trillion Tokens, and JD gave away 1 billion Tokens. These numbers may carry the marketing flavor of Double 11, but the change they point to is real: when Tokens appear in the same recharge center as phone bills and utilities, AI's "electricity" has already begun to be billed by the degree. Tokens on shelves are the coming-of-age ceremony for AI as it moves from the tech altar to a livelihood infrastructure. And every business billed by the degree must ultimately answer the same questionwho produces, who transmits, who measures.