The article takes the explosive popularity of the OpenClaw AI agent framework as a starting point, and systematically elaborates and analyzes the emerging "Token Economy".

Article author and source: Decision Magazine

This moment is just like that moment back then.

OpenClaw, the cartoon lobster that has become a sensation across China, is just as popular as the spicy crayfish, another gourmet delicacy that has traveled from overseas. This massive "crayfish farming movement" is reminiscent of the time when DeepSeek r1 was launched, and local deployments were in extremely short supply.

As an open-source, local-first, and self-hostable AI agent framework, OpenClaw has propelled AI applications to new heights, heralding the official arrival of the era of AI workers who can "get things done".

Putting aside the application barriers and maintenance costs, the only price users need to pay to enjoy the services provided by AI workers is a token.

If traffic is the barometer of the internet age, then tokens are the hard currency of the smart age.

Compared to the tech frenzy of "499 for installation" and "299 for uninstallation," a deeper, more intense battle over computing power consumption and cost control is rewriting business logic in an unprecedented way.

OpenClaw, like a lobster, is more like an appetizer in a grand feast of intelligence. It allows the public to taste the benefits of AI and subtly shapes token consumption habits. Its waving red claws have ushered in the era of the token economy.

1. Are you raising "crayfish" or a money-devouring beast?

In terms of monetization, OpenClaw currently primarily targets more specialized B-end users, including OPCs (one-person companies). For ordinary people who lack high-frequency application scenarios, installing OpenClaw is either like "using a cannon to kill a mosquito," with a poor cost-performance ratio, or it's more like a useless tool, requiring them to spend money again to uninstall it.

A private equity researcher used OpenClaw to automate research report processing, consuming over 12 million tokens in a week at a cost of nearly 1,000 yuan. The underlying cost structure determines OpenClaw's user segmentation. Light users complete 10 tasks daily, consuming approximately 30 million tokens monthly at a cost of 100-300 yuan; self-media creators complete 50 tasks daily, consuming 150 million tokens monthly at a cost of 600-1,500 yuan; and automated teams complete 200 tasks daily, consuming 600 million tokens monthly at a cost of 3,000-10,000 yuan.

Even more critically, OpenClaw employs a perpetual motion model, meaning it can work 24/7, excluding network and power outages. This is fundamentally different from the traditional conversational AI's "answer only when asked, end when answered" response mode. This "always-on" or "constantly flowing" computing power consumption transforms token costs from relatively controllable "pulse-like expenditures" into a continuous and ongoing expense.

For professional users and enterprises, the efficiency leap brought by OpenClaw is enough to cover the high costs. However, for ordinary users, without high-value tasks to support them, it's more like spending tokens to raise a pet shrimp and experiencing the emotional value of being served by a "digital employee".

II. What is Token Economy?

Token economics refers to an economic operating model in the intelligent era that uses tokens for pricing. Tokens can be understood as a "universal pricing benchmark + value circulation carrier" in the intelligent world.

This means that the token itself has a dual attribute:

Semantic metrics: It is the smallest semantic unit for AI to process information—every conversation between you and AI (input) and every piece of content generated by AI (output) is quantified in terms of the amount of information processed.

Computational power mapping: It is also the basic unit of measurement for the computing power consumption that supports AI operation. Every step of AI thinking and reasoning requires computing power. Computational efficiency determines the token output per unit of computing power (FLOPs). Therefore, computing power resources with different performance can be mapped to a unified measurement standard in units of tokens.

Today, the daily token consumption of large-scale models globally has reached 30 trillion, with China's model usage surpassing that of the United States for the first time, accounting for over 60% of the global total. From a simple AI question to enterprise-level model training, everything is settled using tokens. This is not just a technical parameter, but a core indicator of the scale and vitality of the intelligent economy.

When every intelligent interaction is priced with tokens, and every service call is settled with tokens, the thousands of industries empowered by artificial intelligence have quietly completed a reconstruction at the level of economic infrastructure.

III. Andy-Bill's Law is reborn

In the PC era, we upgraded our computer configurations to run more demanding software or games; in the smart era, we need to upgrade our token plans to provide more powerful agent services.

Andy-Beer's Law, the causal law that dominated the PC era for 30 years, is being reborn in the intelligent era.

The traditional narrative is that hardware manufacturers increase performance, and software vendors then use more complex functions and thicker abstraction layers to quickly consume the increased computing power. However, the perceived "speed increase" for users is far less than the doubling of underlying performance.

In the age of intelligence, this law is applied even more radically. Intelligent applications like OpenClaw, as their capabilities increase, consume ever more computing power. Traditional conversational AI might only call the model once, while OpenClaw performs multiple rounds of thought and calls when executing a task. It's quite common for a complex task to consume tens or even hundreds of thousands of tokens.

The deeper change lies in the fact that before OpenClaw, Scaling Law only involved scaling up models and computing power on the B-end—enterprises training models and deploying applications, with computing power consumption concentrated in the R&D and production stages. After OpenClaw, Andy-Beer's Law began to shift the competition between applications and computing power to the C-end.

As the complexity of intelligent agent tasks increases, computing power consumption also skyrockets, with token consumption growing at a staggering compound annual growth rate of 3000%. Heavy OpenClaw users consume between 30 million and 100 million tokens daily. Based on the cost of using top international models, the daily cost ranges from $900 to $3000. Even using more cost-effective domestic models, the cost is still $40 to $140 per day. In comparison, the subscription fee for traditional conversational AI is negligible, and an active OpenClaw user's daily consumption can be several times, or even tens of times, higher.

This means that the "inflation" in computing power consumption is being transmitted from the enterprise side to the consumer side. Previously, it was chip manufacturers and cloud providers who were playing a game of strategy; now, every ordinary user is directly facing the cost decision of "to use or not to use."

The experience of the PC era also reminds us that true competitiveness lies not only in computing power, but also in how to design more restrained and efficient systems. The explosive growth in token consumption, if not offset by technological innovation and efficiency improvements, will ultimately become an obstacle to the development of the intelligent era.

IV. Everything is a Token

Musk proposed a five-layer world theory of AI development, with energy (electricity) at the bottom, followed by chips, data centers, models, and applications. This framework reveals a fundamental truth: innovation at all levels ultimately comes down to the physical constraints of energy.

In the age of intelligence, when electricity is priced uniformly using tokens, this framework can be summarized as: everything is a token.

Based on tokens, we are able to reconstruct the relationship between the physical and digital worlds.

Data is the raw material of tokens. Every training and inference process is a process of transforming massive amounts of data into tokens. The quality, scale, and diversity of data determine the value density of the token. High-quality, specialized data produces scarcer and more expensive tokens; massive amounts of generalized data produce universal but inexpensive tokens.

Algorithms are the refineries of tokens. With the same investment of electricity and computing power, a more efficient algorithm architecture can produce more tokens. DeepSeek V3, with one-tenth the computing power cost, rivals top-tier models, achieving a higher token output rate through architectural innovation. Every improvement in algorithm efficiency translates into an increase in token production per unit of electricity.

Services form the network for token circulation. From large-scale API calls to A2A (Agent to Agent) collaboration and end-to-end application scenarios, tokens flow efficiently between different levels and entities, enabling a collaborative ecosystem for seamless exchange of tokens and services.

More profoundly, tokens will become a cross-modal "hard currency." Text, images, audio, video, 3D models, and sensor data can all be converted into tokens and enter the same computing and pricing system.

The underlying economic logic is clear and profound. In the industrial economy era, electricity was measured in kilowatt-hours; in the digital economy era, data traffic was measured in gigabytes (GB); and in the intelligent economy era, intelligence is measured in tokens. Energy in the physical world, computing power in the digital world, and ubiquitous intelligent services all ultimately converge on the unified unit of value: tokens.

V. The Anchor of Currency

Taking the H100 chip inference scenario as an example: 1 kWh (3.6 million joules) can theoretically produce about 9.23 million tokens, but after deducting heat dissipation losses (PUE≈1.2), the actual output is about 5.5 million tokens.

When electricity costs (energy costs) denominated in tokens become the ultimate unit for measuring the economy, currency will eventually be pegged to electricity in the form of tokens.

This may sound radical, but it's not unfounded. From the perspective of the nature of money, an anchor must meet three conditions: scarcity, stability, and liquidity. Electricity perfectly meets these conditions: it is the basic currency of the universe, convertible into any form of work; it is bound by the laws of physics, cannot be created out of thin air, and is naturally resistant to inflation; moreover, with the popularization of green energy, the electricity supply is becoming increasingly stable and sustainable.

As a financial representation of electricity, tokens are building a complete value chain of "energy → computing power → token → value". In western China, green electricity costing 0.2 yuan per kilowatt-hour can be transformed into AI services at several times the price after computing power conversion. In contrast, electricity prices in Europe and America, ranging from 0.8 to 1.2 yuan per kilowatt-hour, make their token costs 3 to 5 times higher than the Chinese model. This cost difference directly translates into price competitiveness for AI services.

A deeper impact lies in the restructuring of cross-border trade models. Traditional electricity exports require physical power grid transmission, facing 5%-10% line losses, high infrastructure investment, and complex geopolitical barriers. Token exports, however, can achieve instant delivery through increasingly advanced communication networks. The electricity remains within the national grid, but its value is consumed by global users through tokens. Based on a domestic model pricing of approximately 2 yuan per 1 million tokens, one kilowatt-hour of electricity can be sold for 11 yuan through tokens—a value-added effect unattainable by traditional electricity exports.

Within this framework, the anchor of currency value is undergoing a subtle but profound change. As AI services become the absolute engine of global economic growth, and as tokens become the universal unit of account for intelligent interactions, the physical attributes of energy will gradually give way to the digital attributes of tokens. The future monetary system may no longer be based on fiat currency pegged to oil, but rather on tokens pegged to electricity.

VI. Freedom of Calculation

With the widespread penetration of smart applications, the token economy will enable society to achieve computational freedom.

This freedom is first and foremost manifested in the borderless access humans have to intelligent services. Through a token-based pricing system, anyone with an internet connection can access the world's most advanced AI. E-commerce companies in Southeast Asia, medical teams in South America, and sovereign wealth funds in the Middle East can all purchase tokens to obtain intelligent services at the same level as world-class AI companies. This could allow developing countries to skip the arduous traditional IT infrastructure development phase and leap directly into the intelligent economy.

A broader freedom lies in the autonomous collaboration between agents. Based on the A2A protocol, intelligent agents built on different frameworks and from different vendors can discover each other, exchange information, and coordinate actions under a unified standard. An agent skilled in data analysis can pass results to an agent skilled in visualization; an agent focused on financial risk control can invoke the services of an agent skilled in compliance auditing. This multi-agent collaborative network is building a distributed intelligent ecosystem that transcends a single centralized platform.

A further step is the economic autonomy of agents. When intelligent agents possess independent token accounts, enabling them to autonomously trade services, purchase computing power, and optimize profits in the market, humans may become the "initiators" rather than the "managers" of this intelligent economic network. The open-source nature of autonomous intelligent agents such as OpenClaw is the first step in this direction—they possess execution rights, memory systems, and the ability to invoke tools, essentially representing the prototype of "digital citizens."

The end of computing power is electricity, and the future of electricity is tokenization. This is the true main course of the token economy. In the next decade, we may talk about tokens as naturally as we talk about water, electricity, and gas today—it is a unit of measurement for intelligence, a carrier of value, and the basic currency of a society where humans and intelligent agents coexist.

(Original title: "Lobster is just an appetizer in the token economy." Author Tao Heshan is a smart economy worker specializing in policy planning in the field of artificial intelligence.)