In a small apartment somewhere in Southeast Asia, a designer taps a phone screen and asks an artificial intelligence tool to sharpen a blurred product photo, soften the lighting and remove a messy background.

The request looks weightless. It travels as text and pixels, invisible and almost instant. But behind it is a new kind of Chinese export: not a toy, a solar panel, an electric vehicle or a container of electronics, but a string of AI "tokens" produced by servers running in China and sold as a digital service to overseas users.

In April, South China's port city of Shantou, Guangdong province, had completed China's first city-level closed-loop verification for token exports. It was a full process covering offshore data centers, token generation, overseas application programming interface requests and compliant cross-border data transmission and meant that overseas users in parts of Southeast Asia had begun calling AI services powered by tokens generated in Shantou.

The language is technical, but the ambition is straightforward: China wants to turn domestic electricity, data centers and large-language-model capability into a high-value export product.

Under this model, the power stays inside China. The servers stay inside China. The models run inside Chinese intelligent computing centers. What crosses borders is the AI service itself — inference results delivered through an API.

That makes "token exports" a new form of services trade. Traditional computing-power globalization often meant heavy investment abroad: building data centers, laying fiber, securing local energy supply and operating networks in foreign markets. Token exports promise something different: electricity does not leave the grid, computing power does not leave the country, but value flows globally.

Liu Xiaoge, deputy director of research at the development strategy and regional economy department of the Development Research Center of China's State Council, said: "Token exports have become the deep integration of artificial intelligence technology and digital trade in China. They have become a completely new form of digital services trade."

"At a time when China's traditional trade faces geopolitical and tariff challenges, token exports, with their purely digital and zero-logistics delivery model, provide a new way to offset the risks of traditional trade frictions and drive growth in services trade. They will reshape the position of China's digital services in the global value chain," Liu said.

The experiment also offers a glimpse of how China is trying to rewire its export economy for the AI age. For decades, the country's growth was powered by factories, ports and container ships. The next phase may depend as much on offshore cable landing stations, green electricity, GPUs, compliance zones and APIs.

For instance, Shantou's international submarine cable landing station gives it a direct route into Southeast Asia. Shantou carries a large share of China's international outbound bandwidth and provides a low-latency direct node to Singapore, making it one of the fastest Chinese gateways to the region.

The city also sits near large offshore wind resources, giving its token-export story a green-energy narrative that matters to global clients under pressure to meet environmental, social and governance standards.

That is why China's policymakers are paying attention. In the 2026 Government Work Report, Beijing called for new infrastructure projects involving large-scale smart computing clusters and the coordinated development of computing capacity and electricity supply. It also pledged integrated monitoring and distribution of computing capacity across the country and support for public cloud services.

The logic is already visible in China's northwestern regions. In Qingyang, Gansu province, white wind turbines turn slowly over ochre hills, and rows of blue-black photovoltaic panels catch the hard northwestern sun. Instead of being pushed entirely through ultra-high-voltage lines to distant factories and cities, part of that green electricity can be consumed nearby by data centers.

The current runs into racks of servers. GPUs turn electrons into matrix multiplication. The model produces tokens. A user thousands of miles away sees an edited image, a generated paragraph, a line of code or a customer-service answer.

From outside the China Energy Engineering Corp project area in Qingyang, data center buildings operated by Kingsoft Cloud, China Telecom and Chindata can be seen rising nearby. Qingyang has already built 11 data centers, created a domestic 10,000-card computing cluster and surpassed 140,000 petaflops of computing scale.

In the old export economy, a city needed factories. In the token economy, it needs power, fiber, servers, regulation and an ability to guarantee low-latency service.

Now, Shantou and Qingyang do not need to become the next great AI model laboratories. They can become AI trade infrastructure: the service desk for computing power, the customs broker for tokens, and the bonded zone for cross-border data flows.

Wang Peng, a researcher of Beijing Academy of Social Sciences, said: "Token exports are a concentrated expression of China's combined competitive advantages in energy, data, infrastructure and engineering talent. In the context of a deeply reshaped global trade landscape, supporting token exports will help extend China's trade competitiveness into intelligent computing power and has major, far-reaching significance for shaping new drivers of foreign trade."

The timing is important. Global AI spending is surging, and the most expensive layer is infrastructure. Gartner forecast worldwide AI spending would reach $2.52 trillion in 2026, up 44 percent from 2025. It estimated AI infrastructure spending alone would rise from about $965 billion in 2025 to $1.37 trillion in 2026.

That scale creates a massive opportunity for countries that can provide cheap, reliable and increasingly green computing capacity. China sees a chance to sell not merely applications, but the underlying intelligent service.

The demand signal is already visible in developer behavior. OpenRouter, a global AI model API aggregation platform, reported that Chinese-developed open-source models rose from a tiny share of weekly token volume in late 2024 to nearly 30 percent in some weeks in 2025, with growth concentrated in the second half of the year. Deep-Seek, Qwen, MiniMax and Moonshot AI were among the Chinese model families gaining traction in its dataset.

In early 2026, Chinese model token calls continued to climb with Chinese models overtaking US rivals in token usage for the first time.

For Chinese telecom operators and tech companies, that creates a new business line.

State-owned telecom carrier China Telecom announced in late April a series of trial token plans, offering consumers monthly AI usage packages starting at 9.9 yuan ($1.46) for 10 million tokens. Small businesses and developers can purchase enterprise packages with up to 150 million monthly tokens for 39.9 yuan.

Alibaba has announced a new Alibaba token hub business group, defining tokens as a strategic asset and setting out three missions: creating tokens, transmitting tokens and applying tokens.

It has said it will invest more than 380 billion yuan over three years to build a global cloud computing network, using full-chain capabilities in models, computing power and chips to support token exports.

Rao Shaoyang, director of the industry and enterprise strategy at China Telecom's research institute, said such efforts change the nature of Chinese digital globalization.

"In the past, China's digital industries went overseas through apps, software or complete systems. Token exports allow overseas customers to directly call the inference capability, interface capability and intelligent services of Chinese models," he said.

"The model could also alter the economics of China's renewable energy buildout. China has installed vast amounts of wind and solar power, especially in western and northern provinces," he added.

Wang Linwei, an engineer at China Energy Engineering Corp's Qingyang big data center, added that computing-power and electricity coordination has already formed several clear and feasible paths.

These include direct green-power purchases and direct green-power connections, integrated source-grid-load-storage dispatch, flexible peak-shifting of computing loads, and linked regulation between energy storage and computing power, he explained.

"Whether through one-way regulation or two-way interaction, computing-power and electricity coordination has a broad development space," he emphasized.

chengyu@chinadaily.com.cn