Author:Black Lobster, Deep Tide TechFlow
In the summer of 1858, a copper-core cable was laid across the Atlantic seabed, connecting London and New York.
The significance of this event was never about transmission speed, but about the power structure. Whoever laid the submarine cable could extract value from the flow of information. The British Empire, with this global telegraph network, held in its grasp intelligence from its colonies, cotton prices, and news of wars.
The empire's strength lay not only in its fleet but also in that cable.
Over 160 years later, this logic is replaying itself in an unexpected way.
In 2026, Chinese large language models are quietly devouring the global developer market. The latest data from OpenRouter shows that among the top ten models on the platform by Token consumption, Chinese models account for 61%, with the top three all来自中国。来自中国。Developers in San Francisco, Berlin, and Singapore send API requests daily that traverse Pacific submarine cables to data centers in China, where computing power is consumed, electricity flows, and results are sent back.
The electricity never leaves China's power grid, but its value is delivered cross-border through Tokens.
The Great AI Model Migration
On February 24, 2026, OpenRouter released weekly data: the total Token consumption of the top ten models on the platform was about 8.7 trillion, with Chinese models独占 5.3 trillion, accounting for 61%. MiniMax M2.5 topped the list with 2.45 trillion Tokens, followed by Kimi K2.5 and Zhipu GLM-5, the top three all来自中国。来自中国。
Latest data on February 26
This is no accident; a fuse has been lit.
Earlier this year, OpenClaw emerged, an open-source tool that truly allows AI to "work," capable of directly controlling computers, executing commands, and并行完成复杂工作流。Its GitHub stars surpassed 210,000 within weeks.
John, a finance professional, installed OpenClaw immediately and connected it to the Anthropic API to automatically monitor stock market information and provide trading signals. A few hours later, he stared at his account balance in disbelief: dozens of dollars, gone.
This is the new reality brought by OpenClaw. In the past, chatting with AI consumed a few thousand Tokens per conversation, with negligible cost. After integrating OpenClaw, the AI runs dozens of subtasks in the background, repeatedly calling context and iterating循环。Token consumption isn't linear; it's exponential. The bill is like a car accelerating with the hood open, the fuel gauge dropping, unstoppable.
A "clever trick"随即流传在 developer communities: use OAuth tokens to directly connect Anthropic or Google subscription accounts into OpenClaw, turning the monthly "unlimited" quota into free fuel for AI Agents. This is the method many developers adopted.
Official countermeasures随即到来。
On February 19, Anthropic updated its agreement, explicitly prohibiting the use of Claude subscription credentials for third-party tools like OpenClaw. To access Claude's functionality, one must use the API billing channel. Google更是大面积封禁了 subscription accounts connected to Antigravity and Gemini AI Ultra via OpenClaw.
"The world has long suffered under Qin," John随即投入了国产大模型的怀抱。
On OpenRouter, the国产大模型 MiniMax M2.5 scores 80.2% on software engineering tasks, while Claude Opus 4.6 scores 80.8%, a negligible difference. But the prices are worlds apart: the former costs $0.3 per million input Tokens, the latter $5, a difference of about 17 times.
John switched over. The workflow still ran, but the bill shrunk by an order of magnitude. This migration is happening simultaneously worldwide.
OpenRouter's COO, Chris Clark, put it bluntly: Chinese open-source models have captured a large market share because they account for an unusually high proportion of the agent workflows run by US developers.
Electricity Going Global
To understand the essence of Token出海, one must first understand the cost structure of a Token.
It seems light; one Token is roughly equal to 0.75 English words. An ordinary conversation with AI consumes only a few thousand Tokens. But when these Tokens stack up in trillions, the underlying physical reality becomes heavy.
Breaking down the cost of a Token, there are only two core components: computing power and electricity.
Computing power is the depreciation of GPUs. You buy an NVIDIA H100 for about thirty thousand dollars; its lifespan amortized over each inference is the depreciation cost. Electricity is the fuel for the continuous operation of data centers. A GPU at full load consumes about 700 watts, and加上冷却系统的开销, the annual electricity bill for a large AI data center can easily exceed hundreds of millions of dollars.
Now, map this physical process.
A US developer in San Francisco sends an API request. The data travels from California, via Pacific submarine cables, to a data center somewhere in China. The GPU cluster starts working; electricity flows from China's power grid to those chips. The inference is completed, and the results are sent back. The entire process也许只用了一两秒。
The electricity never leaves China's power grid, but the value of the electricity is delivered cross-border through Tokens.
There is something magical here that ordinary trade cannot achieve: Tokens have no physical form, do not need to go through customs, are not subject to tariffs, and are almost invisible in any current trade statistics. China exports a large amount of computing power and electricity services, but it is almost隐形 on official merchandise trade data.
Tokens have become derivatives of electricity. Token出海 is essentially electricity出海。
This also benefits from China's relatively low electricity prices, about 40% lower than the US on average. This is a physical cost difference that competitors cannot easily replicate.
Additionally, Chinese AI models have algorithmic and "involution" advantages.
DeepSeek V3's MoE architecture activates only a portion of parameters during inference. Independent tests show its inference cost is about 36 times lower than GPT-4o. MiniMax M2.5 similarly activates only 10B out of 229B total parameters.
On top of this is involution: Alibaba, ByteDance, Baidu, Tencent, Moonshot AI, Zhipu, MiniMax... over a dozen companies are trampling on each other on the same track. Prices have long fallen below reasonable profit margins; losing money to gain market share is already the industry norm.
Look closely, this is just like Chinese manufacturing going global, utilizing supply chain advantages and industry involution to drive down Token prices hard.
From Bitcoin to Token
Before Tokens, there was another form of electricity出海。
Around 2015, power station managers in Sichuan, Yunnan, and Xinjiang began to receive some strange visitors.
These people rented abandoned factory buildings, filled them with dense rows of machines, and powered them 24/7. The machines produced nothing but continuously solved a mathematical problem, occasionally calculating a Bitcoin from this endless equation.
This was the first-generation form of electricity出海: converting cheap hydropower and wind power, via the hash calculations of mining rigs, into globally circulating digital assets, which were then cashed out into dollars on exchanges.
The electricity never crossed any border, but the value of the electricity, embodied in Bitcoin, flowed to the global market.
During those years, Chinese computing power once accounted for over 70% of global Bitcoin mining算力。China's hydropower and coal power, in this roundabout way, participated in a global redistribution of capital.
In 2021, this came to an abrupt halt. Regulatory hammers fell, miners scattered, and算力 migrated to Kazakhstan, Texas, USA, and Canada.
But the logic itself never disappeared; it was just waiting for a new shell. Until ChatGPT emerged, large models became the new battlefield, the former Bitcoin mining farms transformed into AI data centers, mining rigs became computing GPUs, the once-produced Bitcoin became Tokens. The only constant is the electricity.
Bitcoin出海 and Token出海 are isomorphic in their underlying logic, but Tokens have more commercial value today.
Mining with rigs is pure mathematical calculation; the output, Bitcoin, is a financial asset whose value comes from scarcity and market consensus, having nothing to do with "what was calculated." The computing power itself is not productive; it's more like a byproduct of a trust mechanism.
Large model inference is different. GPUs consume electricity and output real cognitive services: code, analysis, translation, creativity. The value of a Token comes directly from its utility to the user. This is a deeper embedding. Once a developer's workflow relies on a certain model, the cost of switching increases over time.
Of course, there is another key difference: Bitcoin mining was驱逐出去的 by China, while Token出海 is being主动选择的 by global developers.
Token War
The submarine cable laid in 1858 represented the British Empire's sovereignty over the information highway. Whoever owned the infrastructure could define the rules of the game.
Token出海 is also an undeclared war, fraught with resistance.
Data sovereignty is the first wall. An API request from a US developer is processed by a Chinese data center; the data physically flows through China. For individual developers and small applications, this is not a problem. But for scenarios involving sensitive corporate data, financial information, or government compliance, this is a hard stop. This is why the penetration rate of Chinese models is highest in development tools and personal applications, with almost no presence in core enterprise systems.
The chip ban is the second wall. China's AI development faces export controls on NVIDIA's high-end GPUs. MoE architecture and algorithm optimization can only partially offset this disadvantage; the ceiling still exists.
But the current resistance is just the prologue. A larger battlefield is taking shape.
Tokens and AI models have become a new strategic dimension of博弈 between China and the US, no less important than semiconductors and the internet in the 20th century, or even closer to an older analogy: the space race.
In 1957, the Soviet Union launched Sputnik 1, shocking the United States, which随即启动 the Apollo program, pouring resources equivalent to hundreds of billions of dollars today to绝不落败 in the space race.
The logic of the AI hegemony race is strikingly similar, but its intensity will far exceed the space race. Space is, after all, physical; ordinary people don't feel it. AI渗透的是经济的毛细血管; behind every line of code, every contract, every government decision-making system, a certain country's large model could be running. Whose model becomes the default infrastructure option for global developers gains structural influence over the global digital economy无形中。
This is what truly makes Washington uneasy about China's Token出海。
When a developer's codebase, Agent workflow, and product logic are all built around the API of a certain Chinese model, the migration cost increases exponentially over time. By then, even if the US legislates restrictions, developers will resist with their feet, just as no programmer can abandon GitHub today.
Today's Token出海 might just be the opening chapter of this long game. Chinese large models don't claim to颠覆什么; they simply deliver services to every global developer with an API Key at a lower price.
This time, those laying the cables are the engineering teams writing code in Hangzhou, Beijing, and Shanghai, and the GPU clusters operating day and night in some southern province.
This hegemony has no countdown; it happens 24 hours a day, measured in Tokens, on the battlefield of every developer's terminal.
