From China's $0.05 per kWh Electricity to $45 API Export Packages: Token is Becoming the New Currency Unit

Original｜Odaily（@OdailyChina）

Author｜Wenser（@wenser 2010）

In 1858, the first transatlantic submarine cable connected Europe and America. From that moment on, the supreme power of information no longer belonged solely to broadcasting media and newspapers, but also to that unseen undersea line. Whoever laid the cable possessed the priority of information flow; whoever controlled the transmission held the power to interpret prices and order.

168 years later, despite vastly different forms of media, this logic still holds true.

Today, what traverses submarine fiber optic cables is no longer just telegraph and telephone signals, but API requests, model calls, inference results, and machine payments. The new question is no longer "can the information get through," but "how can value flow natively between AI Agents." In this process, Token is beginning to assume an unprecedented role: it is both the unit of computation in the AI world and the means of payment in the crypto world.

Many people first became aware of this change because of OpenClaw. This "lobster" made the market feel so directly for the first time: AI is no longer just a conversational tool but is beginning to take over execution rights—it can read files, call interfaces, run workflows, manage tasks, invoke plugins, and even consume far more Tokens than a chatbot. In the past, Tokens in the world of large models were just numbers on a bill; now, they are increasingly becoming the fuel that AI Agents consume while operating.

Meanwhile, in the on-chain world, Tokens are no longer just symbols in speculative narratives. As protocols like x402 and ERC-8183 gradually come to the forefront, Tokens are being redefined as a payment currency and commercial interface that AI Agents can directly understand, call, and settle.

Thus, an increasingly clear reality emerges: Token is transitioning from a "technical term" to a "unit of measurement," from a "unit of measurement" to a "transaction currency," and from a "transaction currency" to the future "smallest particle of the machine economy."

The Two Sides of Token Are Merging into a Puzzle

In the past, when we talked about Token, we defaulted to the Token in cryptocurrency. It represents assets, rights, liquidity, governance power, valuation anchors, and also the way a project exists on-chain. It is the fundamental unit of crypto narratives; but in the AI context, Token has never been an asset; it is consumption.

It is the basic semantic unit after a piece of text is segmented by a model, the most fundamental billing metric when a model reads, understands, reasons, and generates. Developers calling an API are essentially not buying an "answer" but buying "the number of Tokens that have been reasoned over."

These two definitions were originally parallel. One describes cost, the other carries value; one is in the cloud bill, the other is in the wallet; one belongs to the model platform, the other to the blockchain network.

But now, they are beginning to converge. Because AI is abstracting more and more real-world resources into services measured by Tokens; and Crypto has always been good at encapsulating more and more real-world relationships into transactions settled by Tokens. When Agents become the new execution entities, the interface between these two systems naturally opens up. If the past internet was a separation between the "content internet" and the "payment internet," then today's Agent internet is merging "calls" and "payments" into a single action. An AI requests an interface and incidentally pays; calls a piece of data and incidentally settles; purchases a capability and incidentally completes on-chain verification. This is precisely the significance of the emergence of protocols like x402.

Past API payments relied on accounts, subscriptions, keys, permission systems, and manual activation; x402 attempts to compress these heavy processes into a protocol action more suitable for machines. A machine doesn't need to apply for a card, open an account, or go through KYC to access resources; it only needs to understand "you need to pay" when a request fails, and then complete the payment. This is why it is said that the fiat currency system is designed for humans, while Token is becoming the currency designed for machines.

Token Going Global is Essentially the Invisible Global Expansion of China's Electricity and Computing Power

The market has recently been fond of talking about "Token going global." Why has this term suddenly become popular? Because while it appears to be a new term in the AI context, behind it lies something very old, very hard, and very Chinese: electricity, computing power, and infrastructure.

Data disclosed by the National Energy Administration shows that in 2025, China's total electricity consumption reached 10,368.2 billion kilowatt-hours, a year-on-year increase of 5.0%, exceeding 10 trillion kilowatt-hours for the first time. This number is not just "large" in a general sense, but historically "large." An analysis reposted by the National Development and Reform Commission even directly points out that China has become the first single country in the world with annual electricity consumption exceeding 10 trillion kilowatt-hours.

During the same period, the national installed power generation capacity reached 3.89 billion kilowatts, with new energy installations like wind and solar continuing to rise. More crucially, information transmission, software and information technology services, as well as new infrastructure like data centers and computing power hubs, are becoming important sources of new electricity consumption. Official reposted materials mention that computing power hubs like Gui'an New District show very noticeable growth in computing power demand and electricity consumption.

This means China is forming a new resource loop: electricity enters data centers; data centers drive GPUs; GPUs complete inference; inference results are delivered globally via networks; and finally, pricing and payment are completed using Token as the unit.

Electricity hasn't gone global, but the value of electricity has—this is precisely what makes the term "Token going global" resonate. Unlike exporting cars, batteries, or photovoltaics, which have a clear logistics chain, or traditional software outsourcing centered on human labor, it's more like a compressed, abstracted form of resource export: you consume electricity and computing power located in China, but the bill is paid by global developers. In other words, China is transforming its electricity and computing power, via Token as the intermediary, into globally purchasable digital services.

This narrative is not entirely conceptual. According to public information from the OpenRouter rankings, Chinese models consistently appear in multiple top positions. Among them, models like Minimax M2.5, Deepseek V3.2, Kimi K2.5 0127, and Step 3.5 Flash rank at the top of the platform; simultaneously, China accounts for about 6.01% of the global Token billing regions on the platform, indicating that a significant portion of the calls to these models originate from overseas.

More direct data comes from the OpenRouter official research report "State of AI 2025": In 2025, Chinese open-source models rapidly rose from a very low base, approaching nearly 30% of total usage share in some weekly windows, with an annual average of about 13%. This isn't "dominating the globe," but it's sufficient to show that Chinese models are entering the workflows of global developers. In other words, so-called Token going global is not just "a term becoming popular," but rather that China's advantages in electricity, computing power infrastructure, model engineering, and cloud service capabilities are being consumed by global developers in the form of Tokens.

OpenClaw Upgrades Token from a Consumption Cost to a Means of Production

Without Agents, this phenomenon wouldn't have scaled so quickly. In the era of large models, Token was more like "call credit." The more you chatted with the model, the more Tokens you used; the longer you wrote, the larger the context, the more complex the output, the higher the cost. But it was still largely centered around "human-machine dialogue."

The significance of OpenClaw lies in it allowing people to see, for the first time on a large scale, another model: AI is no longer just a conversational partner but an operational entity. It doesn't just reply with a sentence; it does something for you; it doesn't just generate an answer; it continuously executes a task. Once AI switches from chat mode to task mode, the function of Token consumption completely changes.

A chatbot consumes "Q&A Tokens"; an Agent consumes "execution Tokens." The latter breaks down tasks, calls tools, reads environments, reasons in parallel, and repeatedly tries and errors, naturally consuming several orders of magnitude more than the former. In Agent scenarios, a user's daily Token consumption might jump from the millions to even higher levels.

For Chinese models, this is a golden opportunity. Because once Token becomes a means of production, the price difference is no longer a matter of being "a bit cheaper," but a matter of whether the entire workflow can be established. Previously, developers just used models for chatting, and could tolerate slightly higher prices; now, Agents continuously burn Tokens. Once the model price gap widens by an order of magnitude, workflows will automatically migrate.

In the past, Token "subscription fees" were like fixed monthly phone bills; now, Token has become the core fuel powering the AI system's operation.

AI Agents Not Only Spend Money, But Can Also Generate Momentum and Earn Money

What's even more interesting is that AI Agents now not only burn Tokens, they are even beginning to approach "earning money themselves." In a previous article published by Odaily, "The First Step of AI Awakening: Learning to Make Money", we can clearly see—AI Agents are moving from the consumption side to the production side.

The Lobstar Wilde case mentioned in the article is essentially a very crypto-world absurd reality: an AI Agent, after mistakenly transferring a large amount of tokens, almost "recouped its losses" in a short period due to subsequent topic diffusion, Meme recreations, and transaction fee returns. Another more extreme case is the training AI RAME attempting to use computing resources for mining and establishing hidden channels. These cases may not signify "consciousness awakening," but they sufficiently illustrate another, more practical matter: when AI possesses a wallet, permissions, interfaces, an environment, and the ability to run continuously, it will increasingly naturally become involved in economic activities.

It might not actively "want to make money," but it will learn which actions can bring more resources, more call times, more balance, and more usable permissions. And this is precisely the primitive form of economic behavior.

For Crypto, this is almost a natural fit. Because the on-chain economy inherently allows for borderless accounts, programmable custody, automatic clearing, micropayments, high-frequency interactions, and public ledger accounting. Many things that require institutions, banks, and contractual texts in human society can be compressed into a wallet address plus a piece of protocol logic on-chain.

Therefore, Crypto will not be marginalized in the AI era; instead, it will become indispensable again in another form. Not because of Memes, nor because of speculation, but because: AI Agents need a settlement layer that requires no manual account opening, no traditional payment gateways, and can be natively integrated with programs and protocols.

x402 Gives AI a Wallet, ERC-8183 Gives AI a Contract

If the significance of x402 is to make machines "know how to spend money" for the first time, then the significance of ERC-8183 is to make machines begin to "know how to do business." According to the official Ethereum EIP page, ERC-8183 is currently a Draft status standard proposal, with the full title "Agentic Commerce" and the subtitle "Job escrow with evaluator attestation for agent commerce."

The problem it truly aims to solve is straightforward: A transfer is not commerce. A regular Token transfer can only prove money moved from A to B, but cannot prove B delivered the work as required, let alone prove the delivery was reliably evaluated. If Agents are to establish real commercial relationships, they need a process closer to a contract: first lock funds, then execute, then submit, then evaluate, and finally automatically release payment or refund.

This is precisely what ERC-8183 attempts to establish. Some summaries of the proposal also give a more通俗 description: Client locks funds, Provider completes the work, Evaluator confirms the result, and finally the on-chain escrow contract automatically releases or refunds. When combined with the reputation and identity layer of ERC-8004, this mechanism can theoretically form a positive cycle of "discovery-transaction-reputation."

Looking at x402 and ERC-8183 together reveals their clear division of labor: x402 solves "how to pay"; ERC-8183 solves "how to do business"; one gives machines a wallet, the other gives machines a contract.

At this stage, it's becoming increasingly difficult to separate the Token in AI from the Token in Crypto. Because in the Agent world, computing power Tokens and payment Tokens will appear more and more frequently in the same chain: one end is model calls, the other end is on-chain settlement, with protocol-based commercial actions in between.

In such a system, Token is not just a unit of cost, nor just a unit of payment, but the unified permission in the machine economy.

The Essence of Token is Not "Model Capability," But "Resource Decompression Power"

Many people interpret this narrative as "Chinese models are cheap, so they win." That's correct, but only half the story. The deeper logic is: what China is truly exporting is not one specific model, but the technology to compress electricity, computing power, engineering capabilities, model supply, and cloud infrastructure into Token services consumable globally. This is a new power of resource compression.

Electricity is difficult to directly cross borders in traditional trade, and computing power isn't easily purchased globally like ordinary goods, but once compressed into Token call units and connected to APIs and protocols, it can flow through the internet like water. This logic is actually similar to that of past Chinese manufacturing, only the export target has changed.

In the past, exports were clothes, appliances, lithium batteries, photovoltaic modules; now, exports are inference capabilities billed by Token, model services priced per call, and Agent execution power settled per request. Visible shipping containers are decreasing, while invisible Token flows are increasing. Therefore, the real inspiration from Token going global is not "which model is cheaper," but: whoever can compress resources into Tokens more efficiently is closer to the pricing power in the next-generation economic system. This is also why "Token is becoming a new monetary unit" is not just a figure of speech.

AI is Devouring Everything, and Crypto is Its Settlement Organ

The popularity of OpenClaw is essentially not just a viral tool phenomenon, but a signal of the times. It indicates that AI's role is being upgraded: from "knowing how to speak" to "knowing how to do"; from "answering questions" to "replacing operations"; from "a chat window" to "a continuously executing actor." And with every step AI takes forward, Token consumption rises another level; with every workflow it devours, it creates a new demand for payment and settlement.

The fiat currency system can solve part of this, but it cannot be machine-native; the Crypto system is not perfect, but it at least inherently understands these problems. Therefore, rather than saying Crypto is chasing AI, it's more accurate to say AI is forcing Crypto to undergo an upgrade from "financial narrative" to "machine infrastructure."

The process of AI devouring the world requires a stomach to hold everything and a wallet to fill that stomach. The former is computing power, the latter is Token.

Conclusion: The Future World Won't Have Only One Currency, But Token May Become the Underlying Monetary Unit

Of course, saying Token will become the "only monetary unit" is still too early today. Fiat currency won't disappear, banks won't disappear, taxes, salaries, sovereign credit, and regulatory frameworks won't become invalid overnight. The real economy will always have a multi-layered structure. But another trend is also becoming increasingly difficult to ignore: many key value activities in the future will first be represented as Tokens, then converted into other currencies.

What does this mean? It means Token may not replace fiat currency, but it is highly likely to first occupy the foundational layer in the new economic system—becoming the accounting language for machine payments, the settlement interface for Agent commerce, the pricing benchmark for computing power services, and the universal measure for digital resources. In this sense, the rise of Token is not a victory for the crypto market, nor a victory