AI as an Export Hit Tokens Are China’s New Export Hit—And the Price Advantage Is Systematic

Related Vendors

Taiwan Machine Tool & Accessory Builders' Association (TMBA)

Wuxi InfiMotion Technology Co., Ltd.

Chinese AI models are gaining in importance internationally, and this development is primarily driven by low costs. Behind this is a complex interplay of energy prices, infrastructure and model architecture.

A computer science student in San Francisco types a prompt into his AI chat window. The answer comes back in less than two seconds. The student is happy because the Chinese LLM DeepSeek and its chatbot are much cheaper than ChatGPT or Claude.

His prompt was sent as a light pulse through a fiber optic cable under the Pacific Ocean to China, where it was processed by a Chinese AI model in a data center and then traveled back to San Francisco in the form of words on his screen.

The student paid a fraction of what an American model would have cost him. Many Chinese models are inexpensive, but still offer comparable performance that is acceptable for many computing tasks.

The Chinese tech portal Huxiu has compared the prices. "For example, processing one million input tokens with certain MiniMax and Zhipu models from China only costs 0.3 US dollars, while comparable products such as Claude Opus cost around 5 US dollars, i.e. around 16.7 times as much."

Because many Chinese AI models are also published as open source or open weight models, their use is also becoming increasingly popular with start-ups and larger companies in the US. Most AI models in the United States, on the other hand, are not freely accessible.

Export Hit Token

For these reasons, a mass export of tokens from China to the USA and other regions of the world is currently developing. It is a kind of indirect export of computing power and energy, as the prices for tokens are largely determined by the electricity prices in data centers.

"Every time a global developer calls up a Chinese model, they are effectively buying a token service that was created with China's electricity, computing infrastructure and algorithmic efficiency," says the Chinese business magazine Shidai Caijing, describing this new export hit from the People's Republic of China.

We are no longer talking about trivial data streams. In February 2026, Chinese AI models processed more tokens than American models on the international OpenRouter platform for the first time, reported the industry portal Shenchao TechFlow. In the week from February 9 to 15, 2026, Chinese models delivered 4.12 trillion tokens, while American models only delivered 2.94 trillion. Four of the five most widely used AI models worldwide now come from China.

The second major cost factor is the amortization costs for GPUs. These are also cheaper in China. In 2025, cross-border transactions in the area of computing power will have risen to the equivalent of 240 billion euros worldwide, 68% more than in the previous year, according to the American market research firm IDC.

Lower Energy Prices

The main reason for the cheaper tokens is the lower energy costs in China. These are partly dictated by the government because they also have an important impact on the competitiveness of the Chinese manufacturing industry. However, they can now also be kept so low because China is investing more than any other country in the world in the expansion of solar and wind power and other renewable energies.

In an international comparison, this creates a strong incentive to access Chinese tokens. "China's cost advantage is considerable. Industrial electricity prices in the eastern regions are 0.34 to 0.45 yuan per kilowatt hour (approx. USD 0.05 to 0.07 per kWh), while green electricity in the western regions only costs 0.15 to 0.28 yuan per kilowatt hour (approx. USD 0.02 to 0.04 per kWh)," writes the Chinese business portal.

Comparisons of electricity prices are notoriously difficult, it has to be said. We would first have to define what exactly we are talking about, and that would go beyond the scope of this article. They also fluctuate considerably in both China and Europe. However, according to the Federal Network Agency, industrial energy prices for German companies were around 17 to 18 cents per kilowatt hour without discounts (approx. USD 0.20 to 0.21 per kWh). It is no exaggeration to say that energy prices in the USA and Europe are usually many times higher than in China.

Electricity costs are just one of several factors that make Chinese tokens so cheap. According to experts such as Shi Yuxia from the China Academy of Information and Communications Technology, the synergies between low energy costs and inexpensive hardware for data centers, the cost efficiency of AI models and the well-developed power grid are the key factors.

AI Architecture as a Lever

One cost advantage lies in the architecture of the AI models themselves. Chinese developers like to use lightweight "Mixture of Experts" architectures, or MoE for short. Instead of activating the entire neural network for every query, MoE models only wake up the specific paths that are required for a particular task. This drastically reduces the computing costs per token.

Since the beginning of 2025, when DeepSeek shook up the entire AI industry worldwide with its open R1 model, a price war has broken out among Chinese providers. As a result, tokens have become cheaper and cheaper. A further cost advantage results from the fact that China's government is specifically promoting synergies between green electricity and AI computing power. New AI data centers are now being built right next to wind turbines in the Gobi Desert or solar parks in the Taklamakan Desert.

The green electricity, which is unbeatably cheap even by Chinese standards, is no longer fed into the grid and priced accordingly, but powers the AI computers directly. "The model of direct connection to green electricity connects data centers directly with nearby solar, wind and energy storage systems, thus enabling a point-to-point supply. This reduces transmission losses and lowers energy costs," writes the Chinese AI specialist portal AI Xintianxia. "As of February 2026, 84 such projects had been approved across the country, with a total renewable capacity of 32.59 million kilowatts," the specialist medium adds.

Ultra-High Voltage Lines

A third factor that in many cases makes China's combination of artificial intelligence and cheap green electricity possible in the first place is a network of many thousands of kilometers of ultra-high-voltage lines, known as UHV lines, which transport green electricity from the desert regions of China to the densely populated coastal provinces in the east and south of the country.

The government program "Dong Shu Xi Suan" (Eastern Data, Western Computing) has recently started using these lines to supply data centers with cheap green electricity. The Chinese government has ordered that new data centers must obtain at least 80 percent of their electricity from green sources. According to its latest government report, it is thus specifically promoting a self-reinforcing industrial cycle of cheap energy, which creates more demand for computing power, which in turn promotes the expansion of green infrastructure.

Inner Mongolia and the neighboring province of Ningxia, with their many wind and solar farms and brand new AI data centers now being built in the immediate vicinity, are developing into a cluster that Chinese media are calling the "Xishu Gu" or "Western Data Valley".

If Jensen Huang, CEO of Nvidia, is correct in his statement that the era of tokens has just begun for AI, then China has made a pretty good start to this new era. The prerequisite for rapid and meaningful scaling of AI in industry and society is a good infrastructure: energy, data centers, powerful models. And, if possible, the linking of these resources in a way that releases synergies. China is moving in the right direction along all these axes. Students in California or Freiburg don't have to think about any of this, but they still benefit from it. (sb)