The "Next Stop" for Hashrate Leasing: Token Operations

As computing power gradually becomes an infrastructure, what will be the next high-return potential segment in the industry chain?

According to the latest report from Guolian Minsheng Securities, the industry chain is shifting from "selling computing power resources" to the higher-value model of "pricing by intelligent utility." The Token operation track is expected to become the next high-prosperity investment main line.

With the continuous, beyond-expectations expansion in AI computing power demand, GPU rental prices are generally rising, and the three major operators are accelerating their layout in the Token operating ecosystem. The essence of Token operation is to transform non-standard intelligent services into standardized, tradeable utility goods, upgrading the industry from "selling computing power resources" and "selling model APIs" to "pricing by intelligent utility," providing a unified value measurement and profit-sharing framework for all parties in the industry chain.

From a value transmission logic perspective: AI Agent drives an explosion in Token demand → computing power prices continue to rise → computing power commoditization → the value focus shifts from "selling GPUs" to "selling Tokens" → ultimately extended to "Token operation." The core competitiveness of Token operators lies in resource integration and ecosystem synergy; their long-term value is expected to surpass the traditional computing power rental model. Operators and platforms are likely to become new centers of value in the AI era.

Token "Inflation" Continues: Comprehensive Increase in GPU Rental Prices

The tight supply and demand structure for computing power is being repeatedly confirmed by a series of hard data.

On Nvidia’s Q1 FY2027 earnings call, CFO Colette Kress explicitly stated that since the beginning of this year, the rental price of H100 chips has increased by 20%, the cloud pricing of A100 chips has risen nearly 15%, and even beyond the GPU depreciation period, customers can still generate profitable income. Cloud computing service provider Nebius announced that its on-demand rental price for H100 GPUs would be raised from $2.95 per hour to $3.85 per hour.

The demand-side data is also remarkable. According to the report quoting People’s Daily, the daily average Token calls in China have surged from 100 billion at the start of 2024 to 100 trillion by the end of 2025, and reached 140 trillion by March 2026—an increase of more than a thousandfold in just over two years. The explosion of AI Agent applications is the core driver of this nonlinear growth.

The “price hike wave” among China’s leading cloud providers further confirms the demand-side boom:

• Alibaba Cloud: Multiple price adjustments since March this year; computing card services increased by 5%–34%, CPFS intelligent storage up 30%, monthly 95-mode DDoS protection generally up 20%–50%, and large model MU unit services up about 2%–7%;

• Tencent Cloud: As of May 9, 2026, prices for AI computing power-related products and services increased by 5%;

• Baidu Smart Cloud: As of April 18, 2026, prices for AI computing power-related products and services increased by about 5%–30%, parallel file storage and other products up 30%, and some models even faced throttled query limits.

Token Operation: Three-fold Logic as the Next Stop for Computing Power Leasing

Token operation is not simply a secondary sale of computing power, but a fundamental upgrade of the business model, supported by three dimensions:

Supply-side standardization: Tokenized operation can unify heterogeneous computing power across different architectures, vendors, and nodes into standardized Token supply. Users no longer need to worry about underlying hardware models or cluster deployment details—just purchase the required number of Tokens to access large model services, thus significantly lowering the threshold for usage, and raising computing power resource utilization through dynamic scheduling.

Commercial value upgrade: The business model evolves from "one-time hardware resource sale" to "continuous operation of intelligent services." Intelligent matching of computing power and models reduces users’ Token consumption per unit, while adding diverse profit models like tiered pricing and subscription packages, converting one-off computing power revenue into high-retention operating income streams.

Industry trend shift: The industry has shifted from the infrastructure building phase to the efficient scheduling operation phase of computing power. Token, as the core medium for AI value transfer, is gradually being recognized as a unified standard within the industry. Computing power rental companies extending into Token operation services can not only monetize idle power resources, but also cater to the rapidly growing AI inference and training demand.

The long-term competitive moat of Token operation lies in the closed-loop capability of "industry data accumulation—large model training optimization—high-quality Token output—user feedback iteration." Firms that possess key datasets and can consistently produce Tokens with low hallucination and high professional value are expected to build their own barriers and enhance long-term profit margins.

Driven by Both Policy and Demand, Token Operation Enters a Window of Rapid Development

On the policy side, the nation has incorporated the computing power network into its key planning alongside the water grid, new power grid, communications network, urban underground pipe network, and logistics network, listing them together as the "Six Networks." According to National Development and Reform Commission estimates, investment in the "Six Networks" and related sectors will exceed 7 trillion yuan this year. The Ministry of Industry and Information Technology and the National Data Administration have also launched the "Model-Digital Resonance" initiative, clearly supporting Token technology in empowering 20 key industrial sectors including steel and industrial mother machines.

The National Data Administration has also publicly solicited opinions on the "Implementation Plan for Promoting Industry High-Quality Dataset Construction (Draft for Comment)," clearly proposing to explore new dataset trading models such as token-based transactions, and to build a dataset value system based on tokens that is quantifiable and priceable.

Three Major Operators Fully Enter the Field, Token Ecosystem Takes Shape

The three major operators have completed the intelligent computing infrastructure layout and become core participants in the Token industry, each adopting distinctive competitive strategies:

China Telecom has made clear its "All in Token" strategy, released Token bundles, and completed the first 10-billion-level Token factory group procurement among the three major operators (ex-tax scale about 16.451 billion yuan), with a total intelligent computing scale of 91 EFLOPS and an inter-hub round-trip latency below 12 milliseconds. The Token packages cover proprietary Xingchen large models, DeepSeek, GLM, and other mainstream domestic models. For the consumer market, a minimum of 1 yuan can purchase 400,000 Tokens, and a minimum monthly package of 9.9 yuan includes 10 million Tokens. For SMEs, AI application startup costs drop from the traditional 500,000–1 million yuan to the thousand-yuan level. China Telecom has implemented more than 250 internal AI applications, saving over 2 billion yuan in costs.

China Mobile centers on computing infrastructure, releasing Token operation ecosystem and application alliance, pledging to offer a trillion-level Token trial bundle. Its core platform MoMA pioneered a token intensive operation model, supporting three modes of "cost priority/effect priority/balanced priority" to intelligently match the optimal model, with end-to-end latency under one minute. China Mobile has clearly set a goal to double its computing power and intelligent service revenue by the end of the 15th Five-Year Plan period.

China Unicom features the "Agent+Token+AI Cloud" model, with an intelligent computing scale of 45 EFLOPS, and has built seven 100-megawatt level AIDC parks.

In 2026, all three operators plan, as total capital expenditure declines, to increase the share of investment in computing power, and completely shift resources to intelligent computing services.