In most countries, AI is something people access through an app, a chatbot, or a subscription service. In China, it is increasingly something people receive with their coffee, their dumplings, or their credit card statement. A quiet but significant shift is underway: AI tokens, the basic unit of computation that powers large language models, are being packaged, marketed, and distributed to ordinary consumers in ways that have no real parallel elsewhere in the world.
What a Token Actually Is, and Why Its Price Now Matters
Before understanding what is happening in China, it helps to understand what a token is. When a user types a question into an AI model, the text is broken down into small units called tokens. The model processes these units to generate a response. Tokens are, in effect, the currency of AI computation: every interaction costs some number of them, and the more capable the model, the more expensive each token tends to be.
Token prices have become a competitive battleground in the global AI industry. Chinese AI models, many of them open-source or open-weight, can cost between 60% and 90% less than comparable offerings from companies like OpenAI and Anthropic. That price gap is not just a business detail. It is the structural condition that makes everything else in this story possible. Cheap tokens can be bundled, resold, and given away in ways that expensive tokens cannot.
Daily token consumption in China reflects the scale of what is happening. In early 2024, the figure stood at 100 billion tokens per day. By mid-2026, it had reached 500 trillion. That is not a gradual increase. It is a transformation in how a population relates to AI.
Five Ways Tokens Are Reaching Ordinary People
The distribution mechanisms are varied, and each one reveals something about how Chinese businesses are thinking about AI adoption.
Credit cards are one channel. In July, AI startup Moonshot AI launched a card named after its model Kimi, in partnership with the Agricultural Bank of China and American Express. Cardholders earn AI tokens on their spending, much as other cards offer airline miles or cashback. China Merchants Bank launched a similar card in June, offering new users up to 1.8 billion tokens for use on MiniMax’s models. Shanghai Pudong Development Bank followed with a card offering subsidies worth up to 3 billion tokens for Alibaba’s Qwen models.
Telecom operators have taken a different approach, packaging tokens like mobile data plans. China Telecom offers plans starting at 9.9 yuan, roughly $1.40 per month, for 10 million tokens, with access to 142 large language models through its TokenHub platform. China Mobile offered 400,000 tokens for 1 yuan in Shanghai, billed through phone accounts. China Unicom introduced monthly plans for business users ranging from 15 yuan to 45 yuan.
Physical venues are experimenting too. Beijing’s AGI Bar gives customers who buy a drink unlimited access to DeepSeek V4 Flash, running the model locally on a Nvidia DGX Spark workstation. A dumpling restaurant in Beijing hands out 10 yuan worth of computing credits after a meal, with table cards that frame the offer as feeding your AI agent after you have fed yourself. The restaurant’s two locations distribute more than 100 vouchers a day. A coffee shop in Changsha is preparing a similar model.
Tokens have also reached secondhand markets. On Xianyu, an Alibaba-owned resale platform, sellers offer access to AI models in packages measured by tokens, targeting students, independent developers, and small businesses. Some sellers arbitrage the gap between official prices and discounted access obtained elsewhere.
Finally, banks in Guangzhou are using token data in a genuinely novel way. The Haizhu district launched a financial product called a “token loan,” allowing banks including Bank of China, China CITIC Bank, and Bank of Guangzhou to assess a startup’s creditworthiness based on its token production and consumption rather than traditional metrics like physical assets.
What This Signals About AI Adoption as a Design Problem
Here is what most coverage of this story misses. The distribution of tokens to consumers is not primarily a story about AI capability. It is a story about adoption strategy.
Beijing-based analyst Poe Zhao, founder of the newsletter Hello China Tech, described these token packages as a “supply-led experiment,” noting that most ordinary users still encounter AI through an app or feature without ever seeing a token balance. The infrastructure is being built ahead of clear consumer demand, in the hope that familiarity will follow access.
This matters beyond China. The question of how AI moves from developer tools to everyday behavior is one every technology ecosystem is navigating. China’s approach, distributing tokens through credit cards, phone bills, restaurants, and loan products, treats adoption as a logistics and marketing problem rather than a purely technical one. Whether that approach generates genuine behavioral change or simply inflates consumption numbers is a question worth watching.
In Short
AI tokens, the basic units of computation behind every AI interaction, are being distributed in China through credit cards, telecom plans, restaurants, and even bank loans. Daily token consumption grew from 100 billion to 500 trillion between early 2024 and mid-2026. The driving force is a significant cost advantage for Chinese models relative to US competitors. The deeper question is whether making tokens abundant and cheap translates into meaningful AI adoption, or simply into a new kind of promotional currency.
Based on reporting from Rest of World.