Observer | From lipstick to AI tokens: China's new edge in the AI race

A Chinese livestreamer famous for selling lipstick has found a new product to pitch: AI tokens.

It's an odd-looking shopping trend that raises a bigger question: Who can make artificial intelligence genuinely useful and accessible in everyday life?

The Singles' Day shopping festival, also known as Double 11 because it falls on Nov. 11, is China's annual online shopping carnival and one of the world's biggest retail events.

In the run-up, Li Jiaqi sold 100,000 monthly memberships for an AI office service in half an hour, Reuters reported on November 3. That equals roughly 2 trillion tokens, the units AI services use to measure usage and set prices.

China's largest e-commerce company Alibaba said sales of its AI token plans rose 450 percent from October 1 to October 20 compared with the previous month.

Why is this happening in China?

Part of the answer is a vast digital consumer market and a strong willingness to try new technology.

Daily token use rose from 100 billion in early 2024 to more than 140 trillion by March 2026, an increase of more than a thousandfold, according to China's National Data Administration.

A survey by the University of Melbourne and the accounting firm KPMG found that 93 percent of employees surveyed in China said they use AI at work, against a global average of 58 percent.

Chinese consumers have grown used to AI productivity tools, Lian Jye Su, chief analyst at Omdia, said in comments to Reuters.

Demand pushes companies to innovate. More products, tougher competition and established online shopping channels reinforce each other, drawing in more users and more businesses.

Demand also needs supply at scale.

The Horinger data center cluster in Inner Mongolia had more than 126,700 petaflops of computing capacity in operation, more than 96.4 percent of it for AI, Xinhua News Agency reported in June. A petaflop is a quadrillion calculations a second.

A token-trading platform launched in May connects computing power, model access and payments.

Energy and computing power are now sold as AI services. That capacity lets providers serve users at scale.

Overseas companies are also testing Chinese models on cost. British financial technology company Revolut made a Chinese model called GLM the default coding assistant for roughly 2,000 engineers. It runs the model on its own servers.

Its AI infrastructure chief said spending on running the model came to about one-tenth of previous bills for commercial models, with an average cost of about $0.30 per million tokens.

Engineers then used AI more freely for debugging, documentation and routine work. Cheaper AI does more than cut bills. It makes more work worth automating.

Much of the media coverage of the global AI race focuses on who has the most advanced chips, the most powerful models and the deepest pockets.

These matter, but they are not the only measures of progress.

China's strength lies in linking a large market with computing power and increasingly competitive open models.

For smaller businesses and developing economies, that can mean more choice.

The next AI advantage may belong not only to those who build the smartest models but also to those who make useful intelligence affordable at scale.

A shopping festival promotion is of course no proof of a mature market. Some memberships in Li's livestream cost just 0.01 yuan, a price clearly intended to attract new users.

Comparisons are also difficult. Buyers risk prepaying for AI products that change fast. Token accounting is inconsistent. And some offers are sold at a loss to win customers.

Making AI accessible is not about persuading people to stockpile cheap tokens they may never use.

It is about reliable and transparently priced services that create lasting value. That is the next challenge for China's AI market and a more meaningful test of AI commercialization everywhere.