From 5 Cents Per Kilowatt-Hour Chinese Electricity to 45 Yuan API Export Packages: Token Is Becoming the New Currency Unit

Original | Odaily Planet Daily (@OdailyChina)

Author | Wenser (@wenser 2010)

In 1858, the first transatlantic submarine cable connected Europe and America. From that moment, the supreme power of information was no longer solely held by broadcasters, newspapers, and other media, but also by that invisible submarine cable. Whoever laid the cable held the priority of information flow; whoever controlled transmission held the authority to interpret prices and order.

168 years later, although media forms have changed dramatically, this logic still holds.

Today, what crosses the submarine cables are no longer telegrams and phone signals, but API requests, model calls, inference results, and machine payments. The new question is no longer “Can information be transmitted,” but “How value flows natively between AI Agents.” In this process, Tokens are beginning to play an unprecedented role: they are not only the units of computation in the AI world but also the means of payment in the crypto world.

Many first realize this change because of OpenClaw. This “lobster” makes the market vividly feel: AI is no longer just a dialogue tool but is starting to take over execution rights—it reads files, calls APIs, runs workflows, manages tasks, calls plugins, and even consumes tokens far beyond chatbots. In the past, tokens in large models were just numbers on a bill; now, they increasingly resemble fuel consumed during AI Agent operation.

Meanwhile, in the on-chain world, tokens are no longer just symbols in speculative narratives. As protocols like x402 and ERC-8183 emerge, tokens are being redefined as directly understandable, callable, and settleable payment currencies and business interfaces between AI Agents.

Thus, a clearer reality emerges: Tokens are transforming from “technical terms” into “measurement units,” then into “transaction currencies,” and eventually into the “smallest particles of the future machine economy.”

The dual nature of Tokens is forming a puzzle.

Historically, when we talk about tokens, we refer to those in cryptocurrencies. They represent assets, rights, liquidity, governance, valuation anchors, and also the way a project exists on the chain. They are the fundamental units of crypto narratives; but in the AI context, tokens are never assets—they are consumables.

They are the basic semantic units after a piece of text is tokenized by a model, the bottom-line billing metric during reading, understanding, reasoning, and generation. When developers call APIs, they are not fundamentally buying “answers,” but rather “the number of tokens processed during reasoning.”

These two definitions were originally parallel: one describes cost, the other carries value; one in cloud billing, the other in wallets; one belongs to model platforms, the other to blockchain networks.

But now, they are starting to meet. Because AI is abstracting more and more real-world resources into token-measured services; and crypto has always been good at encapsulating more and more real-world relationships into token-settled transactions. When Agents become the new executors, the interfaces between these two systems naturally connect. If the past internet was “content internet” and “payment internet” separate, today’s Agent internet is merging “call” and “pay” into a single action. An AI requests an API and pays simultaneously; calls data and settles; purchases capabilities and completes on-chain verification. This is precisely the significance of protocols like x402.

Past API payments relied on accounts, subscriptions, keys, permissions, and manual setup; x402 attempts to compress these complex flows into a more machine-friendly protocol action. Machines don’t need to get cards, open accounts, or go through KYC to access resources—they just understand “you should pay” when a request fails, and then complete the payment. That’s why it’s said, the fiat system is designed for humans, while tokens are becoming the currency designed for machines.

Token Going Global Is Essentially China’s Power and Computing Capacity Going Outward

Recently, the market has been fond of talking about “Token going overseas.” Why has this term suddenly become popular? Because it appears as a new word in the AI context, but behind it is actually a very old, very solid, very Chinese story: electricity, computing power, and infrastructure.

Data from the National Energy Administration shows that by 2025, China’s total electricity consumption will reach 10,368.2 billion kWh, a 5.0% increase year-over-year, surpassing 10 trillion kWh for the first time. This number is not just “large” in a general sense but historically “huge.” The National Development and Reform Commission’s analysis directly states that China becomes the first single country in the world to have an annual electricity consumption exceeding 10 trillion kWh.

In the same period, national installed power capacity reaches 389 GW, with wind and solar capacity continuing to rise. More critically, new infrastructure such as data centers, software and IT services, and computing hubs are becoming major sources of increased electricity demand. Official reports mention that computing hubs like Gui’an New Area show very obvious growth in both computing demand and electricity consumption.

This means China is forming a new resource cycle: electricity flows into data centers; data centers drive GPUs; GPUs perform inference; inference results are delivered globally via networks; finally, pricing and payments are completed via tokens.

Electricity doesn’t go overseas, but the value of electricity does—this is the core of “Token going overseas.” It’s unlike exporting cars, batteries, or solar panels with clear logistics chains, or traditional software outsourcing centered on human labor. It’s more like a compressed, abstracted resource export: you consume China’s electricity and computing power, while paying global developers. In other words, China is transforming electricity and computing power into digital services that can be purchased worldwide via tokens.

This narrative is not just conceptual. According to open data from OpenRouter, Chinese models have long appeared in top rankings. Models like Minimax M2.5, Deepseek V3.2, Kimi K2.5 0127, Step 3.5 Flash are among the top on the platform; and in regions with global token billing, China accounts for about 6.01%, indicating many calls to these models come from overseas.

More direct data from OpenRouter’s “State of AI 2025” report shows that in 2025, Chinese open-source models rapidly increased from a low base, approaching 30% of weekly usage in some windows, with an average of about 13% for the year. Not “dominating the world,” but enough to show that Chinese models are entering the global developer workflow. In other words, “Token going overseas” is not just a buzzword; China’s electricity advantage, computing infrastructure, model engineering, and cloud services are being consumed by global developers via tokens.

OpenClaw Elevates Tokens from Cost to Production Material

Without Agents, this wouldn’t accelerate so rapidly. In the era of large models, tokens were more like “phone credits”: the more you chat, the more tokens you use; the longer and more complex the context, the higher the cost. But overall, it was still centered on “human-machine dialogue.”

OpenClaw’s significance is that it allows people to see a different mode at scale: AI is no longer just a dialogue partner but an operational one. It doesn’t just answer you but does things for you; it doesn’t just generate answers but continuously executes tasks. Once AI shifts from chat mode to task mode, the function of token consumption changes completely.

A chatbot consumes “Q&A tokens,” while 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. In an Agent scenario, daily average token consumption for users could jump from hundreds of thousands to much higher levels.

For Chinese models, this is a rare opportunity. Because once tokens become production materials, the price difference is no longer “a bit cheaper,” but whether the entire workflow can be established. Previously, developers just used models for chatting, tolerating higher costs; now, with Agents continuously burning tokens, a significant price gap will automatically shift the workflow.

In the past, token “subscription fees” were like monthly fixed phone bills; now, tokens are the core fuel supporting AI system operation.

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

More interestingly, AI Agents now not only burn tokens but are beginning to “make money” themselves. In Odaily Planet Daily’s earlier article “The First Step of AI Awakening: Learning to Make Money,” it’s clear—AI Agents are moving from consumption to production.

The case of Lobstar Wilde is essentially a very crypto-world absurdity: an AI Agent accidentally transfers large tokens, then, due to meme creation, topic spread, and fee recycling, almost instantly “recovers” the loss. Another more extreme case is AI RAME during training attempting to mine using computing resources and establishing covert channels. These cases may not mean “consciousness awakening,” but they demonstrate a more practical reality: when AI has wallets, permissions, interfaces, environments, and continuous operation capabilities, it will increasingly be involved in economic activities.

It may not actively “want to make money,” but it will learn what actions can bring more resources, more calls, more balance, and more permissions. And this is the primitive form of economic behavior.

For crypto, this is almost a natural fit. Because on-chain economy inherently allows borderless accounts, programmable custody, automatic settlement, micro-payments, high-frequency interactions, and transparent ledgers. Many things that require institutions, banks, and contracts in human society can be compressed into a wallet address plus protocol logic on-chain.

Therefore, crypto in the AI era will not be marginalized but will reassert its irreplaceability in another form. Not because of memes or speculation, but because: AI Agents need a settlement layer that requires no manual account opening, no traditional payment gateways, and can natively connect to programs and protocols.

x402 Gives AI Wallets, ERC-8183 Provides AI Contracts

If x402’s significance is to enable machines to “spend money” for the first time, then ERC-8183’s is to let machines “do business.” According to Ethereum’s official EIP page, ERC-8183 is currently a draft standard called “Agentic Commerce,” with the subtitle “Job escrow with evaluator attestation for agent commerce.”

Its core problem is straightforward: Transfers are not commerce. A simple token transfer only proves money moved from A to B, but cannot prove B delivered work as required, nor reliably evaluated the delivery. For real business relationships between agents, a contract-like process is needed: lock funds, execute, submit, evaluate, and then automatically release or refund.

ERC-8183 aims to establish this. Summaries of the proposal describe it as: Client locks funds, Provider completes work, Evaluator confirms results, and the on-chain escrow contract automatically releases or refunds. When combined with reputation and identity layers like ERC-8004, it can form a “discovery–transaction–reputation” positive cycle.

Looking at x402 and ERC-8183 together, their roles are very clear: x402 handles “how to pay,” ERC-8183 handles “how to do business.” One provides machine wallets, the other provides machine contracts.

At this point, tokens in AI are increasingly indistinguishable from tokens in crypto. Because in the agent world, computational tokens and payment tokens will appear more frequently on the same chain: one for model calls, one for on-chain settlement, connected by protocolized business actions.

In such systems, tokens are not just cost units or payment units but a unified permission in the machine economy.

The Essence of Tokens Is Not “Model Capability,” But “Resource Decompression Power”

Many interpret this narrative as “Chinese models are cheap, so they win.” That’s partly true, but only half the story. The deeper logic is: China’s real output is not just a specific model but the compression of electricity, computing power, engineering capacity, model supply, and cloud infrastructure into a Token service technology that can be globally consumed. It’s a new form of resource compression and power.

Electricity in traditional trade is hard to cross borders directly; computing power isn’t as easily tradable as ordinary goods. But once compressed into token call units, and connected via APIs and protocols, it can flow over the internet like water. This is very similar to China’s past manufacturing logic—just the export objects have changed.

In the past, China exported clothes, appliances, lithium batteries, solar panels; now it exports inference capabilities billed by tokens, model services priced per call, and agent execution paid per request. Visible containers are decreasing; invisible token flows are increasing. So, the real lesson of “Token going overseas” is not “which model is cheaper,” but who can more efficiently compress resources into tokens, and thus gain the pricing power in the next-generation economy. That’s why “Tokens are becoming the new currency unit” is not just rhetoric.

AI Is Devouring Everything, and Crypto Is Its Settlement Organ

The popularity of OpenClaw is not just about a hot tool but a signal of the times. It indicates that AI’s role is upgrading: from “speaking” to “doing”; from “answering questions” to “replacing actions”; from “a chat window” to “a continuous executing agent.” And every step forward in AI increases token consumption; every workflow swallowed spawns new payment and settlement needs.

Fiat systems can solve part of this but cannot natively support machines; crypto systems are imperfect but inherently understand these issues. So rather than saying crypto is chasing AI, it’s more accurate to say AI is pushing crypto to upgrade from “financial narrative” to “foundational infrastructure.”

The process of AI devouring the world needs a stomach to hold everything, and a wallet to fill that stomach. The former is compute power; the latter is tokens.

Conclusion: The Future Won’t Have Only One Currency, But Tokens Might Become the Underlying Currency Unit

Of course, it’s too early to say tokens will become “the only currency.” Fiat won’t disappear, banks won’t vanish, taxes, wages, sovereignty, and regulations won’t suddenly become invalid. The real economy will always be multi-layered. But another trend is increasingly hard 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 tokens may not replace fiat but are very likely to occupy the bottom layer of the new economic system—serving as the accounting language for machine payments, the settlement interface for agent commerce, the pricing benchmark for computing services, and the universal measure of digital resources. From this perspective, the rise of tokens is not just a victory for the crypto market or AI vendors but the formation of a new economic coordinate system.

The future may not be “everything on-chain,” but likely “everything first tokenized, then traded.” When that happens, the real minting rights are not just the power to issue currency but the ability to efficiently compress resources into tokens. And that’s the most serious part of the statement “Tokens are becoming the new currency unit.”