Token consumption reaches the hundreds-of-trillions level as the computing power industry begins to reorder itself around "delivery" and "billing"

date
11:51 19/11/2026
avatar
GMT Eight
Yuegangwan Intelligent Computing summarizes its delivery system as the "integration of three flows"facility flow, equipment flow, and technology flowcorresponding to the three stages of intelligent computing centers, servers, and networking tuning.
On November 18, two pieces of news are worth looking at together. One comes from the application side: Tencent disclosed its third-quarter report, with WorkBuddy's daily active users exceeding 10 million, average daily Token consumption exceeding 100 trillion, and AI demand driving cloud service revenue up 35% year-on-year. The other comes from the infrastructure side: the 2026 Computing Power Internet Conference announced for the first time the national computing power interconnection "1+M+N" industry nodes and launched the Hubei regional node. One is about "models being truly put to use," the other about "computing power beginning to connect across entities." Both point to the same judgment: once calls enter real workflows and computing power begins to be scheduled across parties, the industry's competitive yardstick is no longer "how much was built" but "whether resources can be reliably turned into usable services." Parameter leaderboards exit, consumption takes the stage In the first half of the large-model industry, what mattered was parameter scale, leaderboard rankings, and release cadence. After entering office, programming, marketing, and enterprise services, call volume, active users, and enterprise payments are beginning to provide harder observation dimensions. Tencent's 100 trillion Tokens is representative data of this shift. A Token is the unit by which a model processes information; consumption growth shows that models are being used to execute tasks, not merely displayed and discussed. But a bucket of cold water must be thrown here: there is no simple conversion relationship between Token consumption and computing power companies' revenue. Model efficiency, service pricing, task types, and infrastructure costs each affect the numbers that ultimately land on financial statements. A tenfold increase in consumption does not mean a tenfold increase in revenue. Heat on the demand side also does not equal applause from the capital markets. Nvidia's data center revenue in the third quarter of fiscal 2027 was US$107.6 billion, up 110% year-on-year, with guidance for the next fiscal quarter pointing to continued growth; yet in trading on November 18 Eastern Time, Nvidia's stock fell 2.89%, and the Philadelphia Semiconductor Index rose only 0.10%. Strong results and market divergence appearing at the same time show that, beyond demand, investors are also scrutinizing three things: delivery, financing, and cash flow quality. For computing power service providers, the stage of telling stories based on industry prosperity has passed. Computing power begins to interconnect, scheduling capability pushed to the fore Infrastructure is moving from "single-point expansion" to "resource connection." According to information released by China News Service and the Hubei Provincial Data Bureau on November 18, as of the end of September 2026, China's intelligent computing power scale stood at 2,592 EFLOPS, up 146% year-on-year; the national computing power internet service platform had allocated 11.98 million computing power identifiers, aggregated 362 EFLOPS of resources, and built more than 70 computing power transmission channels. Resource interconnection increases the possibility of computing power flow while pushing cross-region, cross-chip, and cross-cloud scheduling to the fore. Equipment from different sources, adaptation needs of different models, and task loads of different customers need to be matched within the same service system. Computing power services are therefore no longer equivalent to "equipment rental." Platform scheduling, networking optimization, and operations and maintenance capabilities are beginning to directly determine actual output. GBA AI COMP (01396) is one of the few companies in this segment with publicly disclosed sample significance. Its business covers integrated supply of multi-source heterogeneous computing power, "QuantumPai" computing power cloud services, domestic and overseas AIDC construction and operation, and AI vertical application development and investment incubation. The company summarizes its delivery system as the "three flows in one" of facility flow, equipment flow, and technology flow, corresponding to the three links of intelligent computing centers, servers, and networking optimization. The core of the model is not how long the business list is, but whether each link can form continuous delivery. Many participants in the industry cover only one of these links, and as customers shift from "buying resources" to "buying usable computing power," the weight of end-to-end coordination rises. From "how much was built" to "how much is billed" To look at a computing power service provider, orders must be viewed on two levels. Level one: delivered and stably billed. This is real capacity that has already landed on financial statementsas of June 30, GBA AI COMP had delivered computing power orders exceeding RMB20 billion that had entered stable billing scale. This level means "money is already being collected," the hardest validation. Level two: orders in hand. In the same period, the company had intended computing power technical service orders in hand exceeding RMB37 billion. This level is "capacity reserves for the next 35 years," with more than 95% being 35 year long-term contracts and more than 95% carrying prepayments. A key reminder: RMB37 billion is the figure as of June 30, not the current scale. Since then, the company has continued to add new orders and continue delivery and commissioningin other words, these two numbers are staged snapshots formed during rapid expansion, not a ceiling. Three operating indicators also come from the end of June: the computing power rack-up rate is stable above 95%; more than 95% of newly signed orders are 35 year long-term contracts; and more than 95% carry prepayments. Rack-up rate, long-term contracts, and prepayments cannot eliminate market risk, technology risk, and collection risk. But they do one important thing: they turn a heavy-asset company's operating state from "invisible" into "continuously trackable." For a computing power service provider with such large capital investment, this is more worth watching than any scale number. After market close the same day, GBA AI COMP issued a voluntary announcement: the board intends, pursuant to existing authorization, to repurchase from time to time in the open market shares with an aggregate value of up to HK$200 million, funded from internal resources. The announcement is clear: as of the announcement date, the company had not repurchased any shares, and there is no guarantee as to the timing, quantity, price, or whether the repurchase will be implemented. So "up to HK$200 million" is a planned amount, not a completed repurchase amount. Actual subsequent execution remains subject to company announcements and Hong Kong Stock Exchange disclosures. From an industry perspective, capital actions and operating actions must be viewed separately. The repurchase plan conveys the board's attitude toward long-term development, but it cannot replace order delivery, asset operation, and cash recovery. Conclusion After the computing power industry enters a new stage in which "real consumption" and "interconnected scheduling" proceed in parallel, the market will ultimately return to the same set of basic questions: Whether services are usable, whether delivery is stable, and whether revenue and profit can be continuously generated. Token consumption in the hundreds of trillions is the answer on the demand side; the aggregation of 2,592 EFLOPS and 362 EFLOPS is the answer on the supply side; and the RMB20 billion billed scale, 95% rack-up rate, and 95% long-term contracts are the answer given by this company. Whether the three answers align will be told by delivery and billing data in the coming quarters.