On March 23, Liu Liehong, Director of the National Data Bureau, announced a shocking set of data at the China Development Forum: China’s average daily AI Token calls have skyrocketed from 100 billion at the beginning of 2024 to 100 trillion by the end of 2025, and surpassed 140 trillion in March 2026, a more than thousandfold increase in two years. Meanwhile, data from OpenRouter, the world’s largest AI model API aggregation platform, shows that China’s large models’ weekly call volume has consecutively exceeded the US for several weeks, with the top three global call volumes all held by Chinese models. An industry revolution driven by Tokens is reshaping the global tech competition landscape, business models, and even national core competitiveness at an unprecedented speed.
In early 2026, multiple industry developments in Silicon Valley have attracted global tech attention. OpenAI is gradually abandoning the nearly 20-year-old core internet metric DAU (Daily Active Users), shifting to TPD (Tokens Per Day) as its core operational indicator. This shift is no coincidence. At GTC 2026, NVIDIA CEO Jensen Huang redefined data centers as “Token factories,” emphasizing that future competition will focus on “Tokens per Watt.” This is not an isolated phenomenon but signals that a new paradigm of intelligent economy centered on Tokens has fully arrived.
- The Value and Measurement of AI Tokens
1. AI Tokens as the Value Benchmark of the Intelligent Era
From a computer science perspective, a Token is the basic unit of processing various information by AI models. When a piece of text is input into a model, it is broken down into words or subwords; an image is decomposed into pixel patches; an audio clip is segmented into time slices. These indivisible basic units are all called Tokens.
In practical applications, Token measurement follows certain rules. For English text, a short word may count as one Token, while longer words are split into multiple Tokens; a common rule of thumb is approximately 1 Token ≈ 4 English characters. For Chinese text, typically 1-2 Characters correspond to 1 Token. Whether during data processing in model training or in functional output during model service calls, every core action of AI is measured in Tokens. The scale of Token consumption directly reflects the workload and value output of the model, aligning with Marx’s labor value theory.
Tokens provide a quantifiable, comparable measure of value for the development of the intelligent economy. As AI technology evolves from text modalities to multimodal applications, and from application scenarios to programming, video, and scientific research, the strategic positioning of Tokens as a “unified measurement standard” becomes increasingly prominent. This is not arbitrary but an inevitable result of industry development: the industrial age used “kilowatt-hours” to measure electricity consumption; the internet age used “GB” to measure data flow; the AI era naturally requires Tokens to measure intelligent output. At the economic and commercial level, Tokens have become the core value units that are measurable, priced, and tradable in the intelligent age. They connect underlying energy, computing power, data, and top-level intelligent services, serving as a universal metric for AI productivity, cost accounting, and service settlement.
The value chain of Tokens covers hardware manufacturing, infrastructure construction, computing power supply, platform operation, and application development. In its cost structure, electricity and depreciation of computing hardware account for 70-80%, becoming the key factors influencing the international competitiveness of Tokens. “Tokens per Watt” has become a core indicator of AI enterprise competitiveness. This means that under a fixed electricity budget, those who can produce more Tokens with higher energy efficiency will have the lowest production costs and strongest market competitiveness.
2. Factors Influencing AI Token Measurement
As application scenarios become extremely diverse, Token measurement methods have evolved from early simple counts to complex multi-dimensional, dynamically weighted systems.
(1) Input and output differentiation. The most basic measurement still follows the binary structure of “input Tokens” and “output Tokens.” Input Tokens represent the amount of information provided by the user to the model (including prompts, uploaded documents, historical dialogues, etc.), while output Tokens are the responses generated by the model. In commercial billing, because generation consumes significant GPU bandwidth and computational cycles, the cost of output Tokens is usually 3 to 5 times that of input Tokens. This price difference reflects the fundamental difference between “creative labor” and “information reading” in computational resource consumption.
(2) Context measurement and memory costs. From 2024 to 2025, large models’ context windows expanded from 8K, 32K to 128K and even 1 million Tokens. By 2026, handling ultra-long contexts has become routine. However, long contexts are not free. The attention mechanism based on Transformer architecture causes the computational complexity of processing long sequences to grow quadratically or linearly. Therefore, modern measurement systems introduce “context weighting coefficients.” When a user asks a question in a session with a 1 million Token context, even if only 10 Tokens are generated as a reply, the system must scan or retrieve extensive historical memory, incurring hidden costs counted as “active context Tokens.” This makes measurement more precise in reflecting the resource costs of maintaining long-term memory.
(3) Multimodal data tokenization. With the maturity of multimodal large models (LMM), images, videos, and audio are also incorporated into Token measurement. A high-resolution image is no longer viewed as a single file but sliced into hundreds of visual patches, each encoded as one or more visual Tokens. A one-minute video may convert into tens of thousands of temporal visual Tokens. This unified measurement breaks modal barriers, enabling tasks like image captioning, video understanding, and speech interaction to be calculated under the same economic model. For example, generating a 10-second high-definition video may consume a number of Tokens equivalent to writing a thousand-word article, visually demonstrating the information density differences across modalities.
(4) The implicit value of Tokens. As AI Agents become widespread, models no longer just produce single responses but engage in complex autonomous planning, coding, self-reflection, and multi-round searches. These processes generate many intermediate reasoning Tokens that are not directly shown to users but are essential for high-quality outputs. New measurement standards distinguish “surface output Tokens” from “internal reasoning Tokens.” For complex scientific calculations or logical reasoning, internal reasoning Tokens can be dozens of times the final output. Some advanced platforms are beginning to charge differently based on effective reasoning steps or chain-of-thought depth, marking a fundamental shift from “word count” to “intelligence measurement.”
- Trends in AI Token Development
In recent years, AI Token development exhibits three core trends: exponential growth in total volume, extreme compression per unit, and layered value solidification.
Trend 1: Explosive growth in consumption. Statistics show that in 2024, the global daily Token consumption was about 100 billion, which by Q1 2026 surged to 180 trillion, nearly 1,800 times increase. This growth is not linear but driven by paradigm shifts. Early Token consumption mainly came from human-computer dialogues (chatbots), which are low-frequency, shallow interactions; by 2026, mainstream applications are autonomous agents. An agent decomposes tasks, calls tools, writes and debugs code, and verifies results, generating tens of thousands or even hundreds of thousands of Tokens per cycle. With embodied AI (robots) becoming reality, perception and decision-making in real time will produce massive Token streams, with daily global consumption expected to reach 10^16 Tokens by 2030.
Trend 2: Unit cost follows Moore’s Law-like decline. Thanks to hardware iteration (e.g., NVIDIA Blackwell and subsequent Rubin architectures), software optimization (e.g., MoE, quantization, speculative sampling), and improved cluster scheduling, the cost to generate a high-quality Token in 2026 has decreased by about two orders of magnitude compared to 2023. This “Jevons paradox” effect is prominent: efficiency gains do not reduce total resource consumption but stimulate unprecedented demand. Future disruptive technologies like photonic computing and neuromorphic chips could further lower energy per Token, making “infinite intelligence” theoretically possible.
Trend 3: Layered value and specialization. The future Token market will feature clear “value stratification.” General-purpose large models produce “standard Tokens” that are cheap and homogeneous, mainly used for Q&A, basic translation, and simple classification; while high-end Tokens, fine-tuned for specific domains, with proprietary data, and capable of deep reasoning, will be expensive and scarce. For example, diagnostic suggestion Tokens generated by top medical models will be worth far more than casual chat Tokens. This stratification will give rise to “Token futures markets” and “quality certification systems,” with users paying premiums for specific QoS levels.
II. Comparison of the AI Token Industry in China and the US
1. Production and Consumption Scale: China’s Total Surpasses the US
The US’s core advantages in AI lie in chip design and model capability. NVIDIA, dominating the global GPU market, saw its market cap soar from about $300 billion at the end of 2022 to over $4 trillion, a 14-fold increase. Behind this is the US’s sustained leadership in advanced process chip design. Meanwhile, closed-source models like Claude and GPT remain the most capable, with high prices above $5 per million Tokens. This pricing reflects both technological leadership and high-end market dominance.
However, the US’s lead faces structural challenges. On one hand, grid bottlenecks constrain further AI compute expansion, with high electricity costs; on the other, dense model architectures lead to low utilization rates, making unit Token production costs slow to decline.
In contrast, China’s competitive edge mainly lies in cost control and open-source ecosystems. Chinese models like DeepSeek price at about $0.028 per million Tokens, just 1/180 of GPT. This extreme cost-performance attracts global developers—between February 16-22, 2026, Chinese models on OpenRouter consumed 51.6 trillion Tokens, up 127% in three weeks, while US models only consumed 27 trillion and declined. Among the top five models worldwide, four are Chinese, accounting for 85.7%. China’s weekly call volume first surpassed the US in February 2026 and has maintained the lead, with models like MiniMax, DeepSeek, Kimi ranking high for a long time. China’s share of global Token consumption once exceeded 60%.
It’s important to note that China’s overtaking mainly occurred on the inference side, not training. Inference requires less single-card performance, and domestically optimized chips can support massive inference demands; training still relies on a small number of high-end cards, requiring distributed architectures and MoE techniques. This structural feature means China has significant advantages in AI application deployment and value realization, but still has room to catch up in the underlying innovation of foundational models.
- China’s Energy and Engineering Cost Advantages
China’s cost advantages stem from multiple coordinated factors. Electricity costs are the most fundamental, often accounting for over 30% of compute costs. Since training and inference are energy-intensive, a country’s grid stability and green electricity prices directly influence Token production costs. China’s “East Data West Computing” project and unified grid enable green power prices as low as 0.2 yuan/kWh (~$0.028/kWh), compared to 0.08-0.12 USD/kWh in Europe and the US.
Chip costs include hardware procurement, depreciation, and maintenance. The US, led by NVIDIA, has advantages in high-end chip supply but at higher procurement costs. China’s strategy is to rely on fewer high-end chips for training and large-scale use of domestically produced chips for inference, optimizing to minimize unit compute costs. Deep integration of models, cloud services, and chips by Chinese vendors maximizes utilization; US vendors often depend on third-party clouds and chips, with higher adaptation costs.
Engineering efficiency is a key variable affecting Token costs. Chinese firms widely adopt MoE (Mixture of Experts) architectures—splitting large models into multiple experts, activating only relevant ones. For the same $1,000 investment in compute, different technical routes can produce over ten times more Tokens. MoE models outperform dense models in Token output per compute unit. Deep integration across the stack—when model, cloud, and chip design teams cooperate—further boosts utilization beyond expectations.
Global AI competition has shifted from a pure “model performance race” to a comprehensive national strength contest centered on “Token production efficiency” and “unit Token cost.” China’s low-cost, stable energy supply, large unified market, and efficient deployment capabilities have established a huge advantage in large-scale, low-cost Token production, making it a “cost-effective factory” for AI compute. The US relies on technological innovation, high-end ecosystems, and financial capital, occupying the high-value segments of the value chain. The essence of this competition is a comprehensive contest over energy pricing, industrial organization, and digital ecosystem influence. In the near future, China may convert its domestic energy advantage into an international trade advantage—creating a new competitive product: AI Tokens. In this rapidly growing field, China runs a trade surplus with all countries except the US, reshaping global economic and strategic patterns.
III. Will AI Tokens Become a New Global Currency Asset?
1. The Preconditions for Monetization and the Reality Gap
To consider whether AI Tokens can become a global circulating currency, we must clarify the core attributes of money. Economics states that an asset must fulfill three main functions: a measure of value, a medium of exchange, and a store of value. Additionally, it needs widespread acceptance, stability, and sovereign credit backing. Comparing these standards, AI Tokens are unlikely to become true currency in the foreseeable future.
The biggest obstacle is value instability. Over the past two years, the price of a single Token has plummeted over 99%. Such extreme volatility makes merchants unwilling to accept a “currency” that could halve in value within a week. Even if prices stabilize, AI Token value remains tightly linked to compute costs, which are affected by chip technology cycles, energy prices, geopolitical conflicts, and more—making long-term stability difficult.
Limited acceptance is another key constraint. Currently, AI Tokens are only accepted within API calls and AI applications, not for purchasing everyday goods and services. Money is a universal equivalent for all commodities in society, but AI Tokens’ network is confined to AI services. To achieve widespread acceptance, a global transaction network for goods and services must be built—a long-term, massive infrastructure project.
Compared to becoming a currency, AI Tokens are more likely to evolve into a new type of bulk commodity, similar to oil, gold, or copper. This judgment is based on several observations:
First, AI Tokens possess core features of bulk commodities: standardization, tradability, and broad demand. As Jensen Huang pointed out, “In the future, data centers will become nonstop factories producing not traditional products but the most valuable bulk commodity of the digital world: Tokens.” Just as the industrial age relied on oil as fuel, the smart age will rely on Tokens as “intelligent fuel.”
Second, Token pricing mechanisms are increasingly market-oriented: prices rise when supply tightens, fall when demand weakens. This is very similar to traditional commodity markets. As trading scales and standardization improve, Token derivatives like futures and options may emerge, providing risk management tools for producers and consumers.
Third, the supply-demand structure of Tokens exhibits typical bulk commodity features. Supply is constrained by chip capacity, energy supply, and infrastructure, with long adjustment cycles; demand grows rapidly with AI adoption, showing clear pro-cyclicality. This structure causes Token prices to fluctuate periodically rather than decline linearly. The 2026 Token price surge already demonstrated this—despite a long-term downward trend, short-term supply-demand imbalances can trigger sharp price increases.
Fourth, Tokens are becoming potential strategic reserves for nations. As AI capabilities penetrate defense, finance, energy, and other critical sectors, compute security rises to national security levels. Some countries may start strategic reserves of compute resources, with Tokens serving as the measurement unit—leading to a “compute standard” akin to the gold standard.
2. Stablecoins as a New Solution for AI Token Monetization
Given the difficulty of AI Tokens becoming currency, a notable trend is that stablecoins are emerging as innovative monetary forms within the AI Agent economy. When AI Agents need autonomous decision-making and transactions, traditional financial systems reveal limitations: banks do not open accounts for AI, credit cards are not designed for algorithms, and credit systems are human-centric. For AI, money is not wealth but an interface; not a store of value but a path for execution logic. In this context, blockchain-based stablecoins offer unique advantages—permissionless global transactions, instant settlement, and low-cost collaboration—perfectly fitting AI Agent economic needs.
Data shows rapid growth of stablecoins in the AI Agent economy. By March 2026, the total transaction count on the x402 ecosystem exceeded 163 million, with total volume surpassing $45 million, and over 435,000 AI Agents acting as buyers, 90,000 as sellers. USDC dominates the transaction layer on the x402 protocol, accounting for 98.6% of EVM chain volume and 99.7% on Solana.
- Three Possible Future Paths
Based on the above analysis, the future evolution of AI Tokens may follow three paths:
Path 1: Maintain their role as measurement units, not becoming independent assets. In this scenario, AI Tokens always serve as the pricing unit for AI services but lack independent asset attributes. Users pay for AI capabilities, not for the Token itself; Tokens are merely billing tools, not investment targets. This is the most conservative forecast and reflects the current situation.
Path 2: Evolve into bulk assets, forming compute futures markets. As trading volume and standardization increase, Tokens could become tradable commodities like oil or copper. Exchanges might launch Token futures and options, providing price discovery and risk management. Under this path, Token prices will be more volatile but also more financialized.
Path 3: Serve as the measurement basis for a compute-backed monetary system. This is the most revolutionary path: compute becomes the value anchor of money, similar to gold in the gold standard. In this system, sovereign digital currencies (CBDCs) could be issued with compute as the backing, with each unit representing a standardized amount of Tokens. While technically and institutionally challenging, realization would fundamentally reshape the global monetary system.
IV. Strategies for the AI Token Era
1. National Level: Strengthen compute sovereignty and strategic infrastructure
In response to the rise of the Token economy, countries need to incorporate compute resources into strategic infrastructure planning and proactively address governance issues. Specific measures include:
Building compute infrastructure systems. Drawing on successful projects like “East Data West Computing,” plan nationwide compute networks, optimize resource allocation—such as deploying large AI centers in energy-rich western regions utilizing green power; establishing edge computing nodes in demand-dense eastern areas; creating a unified national compute scheduling platform for on-demand, elastic allocation.
Standardizing Token measurement. Currently, different platforms use varied Token measurement methods, complicating developer choices and enterprise cost accounting, and constraining scale. The government can guide industry associations and leading enterprises to develop unified Token standards, clarify conversion rules across modalities, and establish transparent, fair cost accounting mechanisms. This will promote efficient domestic markets and enhance China’s influence in global Token economy.
Improving Token economic governance. Rapid development raises new governance challenges: legal classification of Tokens (service units, digital assets, securities), cross-border regulation, financial risks from price volatility, balancing user rights and innovation. Policymakers, technologists, industry, and academia must collaborate to build governance systems suited to the smart economy.
Participating in international rule-making. China should actively shape international standards for Token measurement, promote cross-border cooperation, and include Token trade rules in trade agreements. Leading the rules will ensure a favorable position in future global Token economy.
2. Enterprise Level: Reconstruct Token efficiency mindset and business models
For enterprises, Token strategy is no longer just a technical choice but a top-level design affecting competitiveness and value. Key actions include:
Establishing Token efficiency awareness. When selecting AI tech, evaluate Token efficiency as a core metric—aligning compute resources with Token consumption. From prompt design, model invocation, to output optimization, every step should optimize for efficiency and cost. Precise prompt engineering reduces wasteful Token use; strategic invocation improves utilization. Borrowing from telecom “good-put,” focus on “how many Tokens truly advance user goals,” shifting from “how many Tokens used” to “how much value created.”
Reconstructing business models and pricing. The industry is shifting from “traffic subsidies” to “value-based pricing.” Early low prices attracted trial users, but caused inefficient resource use—some estimates show 40% of free calls are non-business testing. Raising prices appropriately filters non-core demand and stabilizes service for high-value clients. This “value-for-volume” approach marks a move from internet-scale growth to software industry pricing.
Talent and incentive reform. Jensen Huang proposed giving engineers Token budgets worth half their annual salary as an incentive—“If you hire a $500K/year engineer and they don’t consume at least $250K in Tokens, I’d be worried.” This aligns talent incentives with Token efficiency and AI productivity.
3. Personal Level: Cultivate Token literacy and new human-AI collaboration skills
For individuals, the rise of Token economy presents both challenges and opportunities. To adapt, individuals should:
Build Token literacy. Most users lack understanding of Token consumption, model capabilities, and pricing, leading to misuse—some buy and sell stocks via AI agents and wake up to zero; others issue commands on social media that cause multiple AI agents to be exploited. Token literacy is becoming a fundamental digital skill.
Develop new work modes of human-AI collaboration. Huang Huang predicts AI agents will operate 24/7, continuously generating Tokens as they execute tasks. This shifts personal work from “doing oneself” to “commanding AI,” from “executor” to “supervisor.”
Embrace lifelong learning and skill iteration. The rapid development of Token economy shortens skill half-life. Today’s popular models will soon be replaced by more efficient architectures; current hot models will be surpassed. Maintaining learning agility and adaptability is more important than mastering specific skills. Individuals should cultivate continuous learning habits, stay updated on AI and Token trends, experiment with new tools and methods, and develop interdisciplinary knowledge to understand the economic and social implications—only then can they thrive in the Token economy wave.
