AI token economy revolution creates new technology winners for China

China’s cheap artificial intelligence models are rapidly attracting users around the world and creating new winners in China’s stock market.

As AI development increasingly focuses on performing diverse tasks, an economy is emerging around the demand for “tokens,” the basic units of data in large-scale language models. A surge in global token consumption has given China a distinct advantage, thanks to an abundant supply of low-cost electricity and a large number of competing AI model developers.

Niche startups are rapidly gaining investor support over established technology leaders as the pure field of AI has greater growth potential. Recently listed development companies MiniMax and Knowledge Atlas Technology Joint Stock, also known as Zhipu, have already had a market value of more than US$40 billion, surpassing Baidu Inc and Kuaishou Technology.

The big stock prices of major companies, including Alibaba Group Holding Ltd., are defying losses and operating erratically exposed businesses to the broader economy. Investors such as Victoria Mio, portfolio manager and head of Greater China equities at Janus Henderson Group, see big changes beginning.

“This increase reflects a structural rerating rather than a short-term transaction,” Mio said. “Investors are reacting to visible token consumption, early pricing power, and evidence that AI inference is becoming a monetizable activity rather than just a story.”

Similar to last year’s deep-seeking boom, investors are buying into the assumption that China will be the long-term winner due to its cost efficiency. The Chinese government has seized on the latest trend to boost the global competitiveness of its local technology industry, with “token exports” becoming a buzzword in state media.

AI economy

A token is a chunk of data that is approximately 4 characters worth. AI companies offer free and fixed monthly plans, but token limits are low. Users whose demands exceed these limits, such as application developers, pay a fee for using LLM based on the amount of tokens used. They pay for both input and output, with the latter token costing up to 10 times the former.

Tokens have become the basic unit of the AI ​​economy. They are used not only in transactions between companies and customers, but also by technology employers to measure developer productivity.

According to data from LLM distribution platform OpenRouter, the three most used models in token consumption this month were all Chinese brands: Xiaomi’s MiMo, Alibaba’s Qwen, and DeepSeek. Among small companies, MiniMax has two models in the top 10, and Chinese startup StepFun has one model.

“China can compete in terms of offering tokens and models to global users,” said Bush Chew, investment manager at Aberdeen Standard Investments in Hong Kong. “The total cost of building a data center in China is much cheaper than in the United States.”

Perhaps more impressively, MiniMax ranks fourth in overall market share by company as of April 19, according to OpenRouter data. This was followed by only US giants Alphabet, Anthropic PBC and OpenAI. MiniMax’s most popular models cost around US$1 per million output tokens, while some models from Google cost US$3 and Anthropic’s Claude more than US$15, and that cheapness helps increase its appeal.

Goldman Sachs views MiniMax as one of the best-suited Chinese AI model companies to capture growth potential “on the back of comprehensive multimodal services, strong commercialization capabilities, and cost advantages of per-token AI models.”

It and Zhipu are two of the “Six Tigers,” startups offering more specialized AI services, often backed by larger companies. Alibaba still owns a stake in MiniMax.

Both companies have been listed in Hong Kong since January, and Zhipu shares have soared about 670% since then, while MiniMax is up nearly 440%. By contrast, Alibaba is down 5% since the beginning of the year, and fellow giant Tencent Holdings is down 15%.

“Maximize tokens”

More companies are set to make their market debut as investors rush to acquire new startups, with StepFun, Moonshot AI and Baichuan among other Six Tigers expected to launch IPOs. Indeed, the rapid rise in the share prices of new entrants that remain unprofitable has raised red flags for some investors.

“The valuation expansion in some areas has happened quickly over a short period of time, probably faster than actual earnings growth,” said Nicholas Chui, a fund manager at Franklin Templeton. “The introduction of the model will be important in the future. If companies and individuals in Japan and overseas continue to introduce it, it will provide some support.”

The market is constantly undergoing intense competition from new entrants, and even large companies with deep pockets are not staying silent. Tencent is seen as having a huge potential advantage in the AI ​​space given that it owns the ubiquitous WeChat app, but its latest efforts have been unimpressive.

On the other hand, cost remains an important factor for users. The immense popularity of OpenClaw, which helps users automate all sorts of everyday functions, has led to a ferocious demand for tokens offered by cheap niche models. The same goes for “token maximization,” a trend in the tech industry where engineers maximize the consumption of tokens to demonstrate how much work they’re doing.

While the US still maintains a technological advantage, China’s low-cost pure AI applications are scooping up much of the global demand. China’s average daily token consumption exceeded 140 trillion in March, compared to 100 billion at the beginning of 2024, according to state media.

“For investors, China’s AI should be evaluated based on utilization rates, cost curves, and monetization efficiency, rather than the quality of headline models,” said Janus Henderson’s Mio. “China is industrializing AI services at low marginal costs and rapidly incorporating them into real-world workflows.” Bloomberg