AI Token Goes Global: Cheap Electricity and Impassable Walls

China accounts for 61% of the world’s AI Token call volume, but the money earned may be just a fraction of what Anthropic makes.

On a late night in February 2026, in the Mission District of San Francisco, an Indian developer named Arjun stared blankly at the bill on his screen. He was running an automated code review agent workflow with Claude. Over a dozen sub-tasks running in parallel, repeatedly calling context, the consumption of AI Tokens was not linear but exponential. That night, he burned through dozens of dollars.

The next day, he switched to MiniMax’s M2.5 model on OpenRouter—the world’s largest AI model aggregation platform. With the same workflow, the bill shrank by an order of magnitude. The code ran the same, but the cost was hardly different.

What Arjun didn’t know was that each of his requests was originating from California, traveling across the Pacific Ocean via submarine cable, arriving at a data center in northwest China. GPU clusters powered up, electricity flowed from the national grid into the chips, inference was completed, and results were returned in less than two seconds.

Electricity never left China’s power grid, but the value of that electricity, carried by large models and AI Tokens, has gone overseas.

This is not just a developer’s choice; all of Silicon Valley is engaged in “full token-maxxing.” Arjun is just a microcosm of this global AI Token consumption frenzy.

On the other side of the Pacific, the scene is completely different.

In spring of the same year, on the Gobi Desert in northwest China, several computing centers that had been bustling two years prior had fallen silent. An old-timer who had been building computing centers for years told me during a period of FOMO:

“AI Token going overseas isn’t as simple as many media reports suggest.”

He wasn’t talking about technical issues but digestion issues. Self-built data centers by large model companies are the optimal path; telecom operators are second, with channels to backstop; the most awkward are the private capital chasing the AI wave—seemingly sunny and prosperous, but without anyone to backstop.

The wall of cross-border restrictions is insurmountable—those abandoned or semi-idle data centers in northwest China aren’t because they can’t be built, but because after building them, they realize: the wall is even higher than the power station.

On one side, Silicon Valley is engaged in full token-maxxing; on the other, the idle data centers in northwest China. China’s cheapest electricity has produced the world’s cheapest AI Tokens. But how much value can Chinese AI companies actually capture from this?

1. A decade ago, the same power plants hosted another wave of visitors

Around 2015, managers of power plants in Sichuan, Yunnan, and Xinjiang began hosting a strange group of visitors. They rented abandoned factories, packed them with dense machines, and ran them 24/7. The machines produced nothing but kept solving a mathematical problem. Occasionally, a Bitcoin would pop out from endless calculations.

That was the 1.0 version of electricity going overseas. Cheap hydropower was used by miners to hash and convert into globally circulating digital assets, then realized into dollars on exchanges. Electricity didn’t cross borders, but its value, carried by Bitcoin, flowed globally. At its peak, China accounted for over 70% of the world’s Bitcoin hash power.

Bitcoin’s value capture path was extremely short—electricity turned into computation, computation into BTC, BTC into dollars. No reliance on anyone in between. Bitcoin itself is a terminal product, a form of energy-backed digital gold, which can be cashed out directly from the mine. It can go offshore naked.

But a short path means shallow roots.

In 2021, a regulatory crackdown scattered miners, and the value capture instantly dropped to zero. Hash power migrated to Kazakhstan, Texas, and Canada. The power plants continued to produce electricity, but that electricity no longer translated into dollars.

The logic of electricity didn’t disappear; it was just waiting for a new shell.

After ChatGPT’s emergence, the same power plants, the same factories, and some even the same power contracts, transformed into AI data centers. Mining machines replaced with GPUs, Bitcoin replaced with AI Tokens.

But AI Tokens are not Bitcoin.

Bitcoin is a terminal product; AI Tokens are semi-finished. AI Tokens must go through layers of modeling, products, and workflows to become something customers are willing to pay for. They cannot go offshore naked. The industry chain built around this semi-finished product is far more complex—ranging from underlying green electricity and liquid-cooled data centers, to AI chips and servers, to upper-layer large model APIs, aggregation platforms, and cross-border compliance, with seven interconnected layers.

The chain has lengthened. Every layer is vulnerable to hijacking.

What remains unchanged are electricity and one unaltered question: how much value can actually stay?

2. The journey of a kilowatt-hour and its broken bridge

To answer this, let’s follow a kilowatt-hour.

Every time Arjun presses Enter, five links pass the baton: electricity → compute power → model training → model inference → AI Token delivered to his screen.

The first three steps are long completed. A kilowatt-hour from Sichuan hydropower flows to a data center in Inner Mongolia, powering GPU clusters for months, feeding trillions of data points, training a large model. After burning out, the model’s “recipe” is fixed. China’s electricity advantage is fully encoded here—lower training costs, more efficient architecture design, engineering optimizations driven by fierce competition among a dozen companies. The energy of a kilowatt-hour is compressed into a few hundred GBs of a model file.

The last two steps are happening right now. Every time Arjun presses Enter, a data center must start inference: load the recipe, consume compute and electricity, produce a batch of AI Tokens on-site, and send back results. Each AI Token’s birth requires real-time electricity.

The question is: which data center?

2.1 Two completely different paths for AI Token going overseas

Arjun is currently following the first— inference is completed in China, and AI Tokens are delivered across borders via API. Requests cross the Pacific to Guizhou, then are transmitted back to San Francisco. The domestic green electricity at 0.38 yuan/kWh directly lowers the marginal cost of each AI Token. The low price Arjun enjoys is essentially China’s hydropower paying the bill, with the electricity advantage fully realized.

The second—moving the recipe overseas, doing inference in Virginia or Singapore. Companies like Zhipu, DeepSeek deploy on Microsoft Azure; MiniMax on Amazon AWS. They use local electricity and local GPUs, and the cost of each AI Token bears no relation to Chinese electricity prices.

2.2 Coca-Cola never exports bottled water

The second model is like Coca-Cola—never shipping bottled water from U.S. factories worldwide, but exporting the formula, which is then made locally with local water and bottling lines.

In Alibaba Cloud’s Baolian backend, there’s a dropdown menu: Service Deployment Scope—“Mainland China,” “International,” “Global.” Selecting “International,” inference runs in Virginia, compliance disappears, and electricity prices become U.S. prices. This dropdown is the switch from the first to the second model.

But Coca-Cola’s global success is because no one can copy its formula. Large models are different—DeepSeek, Tongyi Qianwen are open source, weights files are downloadable by anyone. The formula is public, but the bottling lines are others’, and the electricity is others’. The only remaining anchor for value capture is: speed of releasing the formula.

Guotai Junan’s research reports that the global large model iteration cycle has shrunk from half a year to a few months. Once chip sanctions slow down iteration, the window of freshness shortens further.

Arjun doesn’t care which country the formula comes from. He only cares about which is cheapest this week, and whether it will still be cheap next week.

The overseas journey of a kilowatt-hour is broken between training and inference. Not just the cost chain, but the value capture as well.

The first model preserves the electricity advantage but loses the market—enterprise clients won’t accept data passing through China, and latency is a hard constraint. The second opens the market but loses the cost advantage—you’re no longer selling cheap electricity, but the formula itself.

Time is not on China’s side either. Deloitte predicts that the global AI compute focus is shifting from training to inference— inference will account for two-thirds of the total, possibly over 80% in the future. Training costs are a one-time expense; inference bills are daily. As inference becomes more important, electricity prices matter less.

This is the fundamental difference between AI Token going overseas and photovoltaic going overseas. PV’s cost advantages from silicon to modules to shipping are fully transmitted without breaks. AI Token’s cost chain cracks open between training and inference—if the electricity dividend can’t be transferred, the real cross-border transmission is the engineering capability embedded in that hundreds-of-GB formula.

Two models, two dead ends. How far can the path of selling AI Tokens directly with cheap electricity go?

3. Cheap electricity, an insurmountable wall

Not far. Three walls are narrowing this path.

First is the physical wall.

Arjun’s requests cross the Pacific, with a round-trip latency of 150 to 300 milliseconds. It’s barely noticeable in a single exchange. But an agent workflow involves dozens of continuous calls, and machines don’t wait—they accumulate seconds of delay, and the workflow stalls. Not a political issue, not a regulatory issue—speed of light is the limit.

Either stand in the light, or the light stands there.

Second is the regulatory wall.

When U.S. companies procure AI services, technical leaders must answer five questions: Does the data pass into China? Where are logs stored? Are inputs and outputs used for training? Is it compliant with local laws? Who is responsible if something goes wrong?

If they can’t answer these five questions, the procurement process stalls. It’s not that the models are bad, but compliance departments won’t sign off.

Morgan Stanley’s chief economist Xing Ziqiang cited a sharper precedent: Huawei’s 5G equipment also had technical and price advantages, yet after 2018, it was still kicked out of Western telecom networks. 5G base stations, like AI Tokens, involve data passing through whose devices and servers. His words:

“Don’t overhype the electricity advantage of AI Token going overseas, and ignore geopolitical and security considerations.”

Technical advantage can’t fix the trust deficit.

Third is the political wall.

Chip bans block training; model review blocks listing. The most unpredictable variables.

These three walls narrow the scope: relying on cheap electricity to sell AI Tokens overseas can only reach a tail of developers who are insensitive to compliance, tolerant of latency, and highly price-sensitive. The route of exporting the formula, meanwhile, can’t leverage the electricity advantage.

Two models, two dead ends. Where does China’s AI enterprise stand in the global value chain?

4. Champions on the small stage, spectators on the value chain

On February 24, 2026, OpenRouter’s data revealed: the top ten models on the platform consumed a total of 8.7 trillion AI Tokens, with Chinese models accounting for 5.3 trillion, or 61%.

“China surpasses the US for the first time”—this news spread across Chinese internet.

OpenRouter’s COO Chris Clark described what he saw in a podcast: “Chinese open-source models have an unusually high share in U.S. enterprise agent workflows.” A complex coding task costs Claude $50 to $100, while DeepSeek V3.2 costs about $0.50. A hundredfold price difference— for startups running dozens of agents, it’s a matter of life and death.

But behind the 61% figure lie two truths.

First: this is just a small stage.

OpenRouter’s statistics cover only about 3% of global AI Token consumption. The big stage is elsewhere, with a huge disparity.

By April 2026, Anthropic’s annual revenue surpassed $30 billion. Just 15 months earlier, it was only $1 billion— a 30-fold increase. Claude Code, a coding tool, reached $1 billion in six months. Over 1,000 companies pay over a million dollars annually, with 80% of revenue from enterprise clients. OpenAI’s annual revenue is about $25 billion.

On China’s side: MiniMax burned $500 million in nine months, with revenue of $79 million. Caijing magazine is even colder—some Chinese models’ API gross margins may be negative, losing money on every request.

China produces the most AI Tokens, but is almost invisible in the value chain.

Second: not all AI Tokens are equally valuable.

The same AI Token, valued at $0.00001 per million for casual chat, but $200 per million for coding, and $1,000 per million for legal review. A difference of hundreds of thousands of times. Industry estimates show that less than 5% of consumption creates over 80% of commercial value.

In the same week as this article was written, Anthropic and Blackstone, Goldman Sachs established a $1.5 billion joint venture, deploying engineers directly into PE portfolio companies. They also launched 10 financial agents—pitchbook generation, KYC review, credit memos, month-end closing, financial audits. Jamie Dimon and Dario Amodei appeared together. A KYC agent consumes dozens of dollars of AI Tokens per run, saving thousands of dollars in compliance manpower.

This is the real picture of high-value AI Tokens.

Goldman Sachs’ Marc Nachmann said: “Having a model alone won’t change your workflow. You need people who can integrate technology with actual business.”

This sentence sharply highlights the value gap between China and the US. Chinese companies compete on who has cheaper AI Tokens; Anthropic competes on how to embed AI Tokens into Goldman’s every business line. The former sells raw materials, the latter sells solutions. Chinese models are tools in the global developer toolbox, not chief architects—they are pieceworkers.

It’s very much like China’s photovoltaic industry in 2008—leading in global shipment volume, but the most profitable components are in the hands of others. Pricing power, brand premium, high-end market share—all in others’ hands. That year, China’s PV hadn’t yet truly dominated globally; it would take more than a decade.

But the PV story of 2008 didn’t stop at “volume without profit.” The Chinese companies that ultimately won globally weren’t just cheap silicon—they controlled the entire stack from silicon to modules to power stations.

The hope for AI Token going overseas isn’t in electricity prices—it’s in whether it can transform from “cheapest producer” into the infrastructure embedded in enterprise workflows.

Some are trying.

XunCe Technology is using AI Tokens for billing in finance and energy verticals, deeply binding customer data and business processes. Token billing share increased from 5% to 20-30%, turning profitable in late 2025. They’re not selling cheap AI Tokens but “re-running your business with AI Tokens.” The logic is similar to Anthropic’s on-site delivery to Goldman—just in a Chinese vertical industry, with a $1.5 billion and Jamie Dimon’s backing.

The gap is huge, but the direction is right.

5. Electricity remains unchanged, bills are changing

In May 2026, Doubao launched a paid version at 68 yuan/month. With 345 million monthly active users, it started charging.

In the same week, Tencent Cloud AI services increased prices by 5%, and Zhipu’s GLM-5 pricing rose 50% over the previous generation.

Someone analyzed Doubao’s inference costs—58% hardware depreciation, 29% electricity. Each additional user means more GPU cluster consumption. ByteDance internally voiced: “No clear path to commercialization; the inference cost for large DAU puts pressure on profits.”

The days of losing money to chase growth are over.

Price hikes reveal a more fundamental issue than the three walls: electricity advantage reduces AI Token production costs, but AI Token’s value has never depended on production costs; it depends on what it’s used for. A kilowatt-hour turned AI Token, whether for casual chat or enterprise decision-making, can be worth ten thousand times more.

China has the cheapest electricity in the world, trained the cheapest AI Tokens, and holds 61% of OpenRouter’s volume, but Anthropic’s annual revenue may surpass the combined total of all Chinese large model companies.

The water plants and data centers in Guizhou are still turning. Ten years ago, the same electricity powered mining farms, which were dispersed. Now, the same electricity powers AI Token factories. The factories are still there, but the output of AI Tokens is increasing, while the remaining value diminishes.

Arjun again saved dozens of dollars tonight. But every dollar he saves is precisely the part of value that Chinese AI companies have failed to capture.