China Packages AI Tokens as Credit Cards, Phone Plans, Loans

China turns AI tokens into retail products

Chinese financial institutions and telecommunications companies have begun packaging AI model tokens as consumer products, marking a shift from enterprise-focused compute sales to retail offerings. Moonshot AI and Agricultural Bank of China launched what Moonshot describes as an "AI-native credit card" available exclusively in mainland China, with promotional campaigns running through September 30.

First-time cardholders who spend 5,888 yuan within three months receive a co-branded plush charm and two months of premium membership, limited to 1,000 participants. Moonshot, the company behind the Kimi AI assistant, has been pursuing a $50 billion valuation ahead of a planned Hong Kong listing, according to earlier reporting by The Next Web.

China Telecom launched trial token packages on May 17, structured in three tiers targeting developers, small businesses, and households. Individual customers receive access to the carrier's Xingchen model and DeepSeek V3.2, while developers gain additional access to GLM5. Industry sources place the consumer entry tier at 9.9 yuan monthly for 10 million tokens. The carrier plans to introduce a Tianyi Token system allowing customers to redeem loyalty points for token packages through a management platform called TokenHub.

District government backs token-based lending

The Haizhu district of Guangzhou has taken the most aggressive approach, releasing a Token Loan program in August alongside eight supporting measures. Rather than evaluating AI startups based on traditional assets like plant and equipment, participating banks now incorporate token consumption, platform qualifications, and payment-collection progress into credit reviews. Bank of China's Guangzhou branch can establish credit limits based on contracts or a company's token usage patterns.

Haizhu offers subsidies up to 2 million yuan annually toward companies' token expenses and up to 1.5 million yuan for improving token output per chip.

Why it matters

The commodification of AI tokens as consumer products represents a fundamental shift in how compute resources reach end users. While Western markets focus on infrastructure investment—Europe has allocated roughly €30 billion toward AI gigafactories with only about €1 billion committed from Brussels—China is subsidizing consumption rather than production capacity. This approach could accelerate AI adoption among small businesses and individual users while creating new revenue streams for carriers and financial institutions. However, the model faces inherent challenges: research across 25 European languages shows tokenizer efficiency ranges from 1.23 tokens per word in English to approximately 3.1 in Greek and Maltese, meaning identical token packages deliver vastly different service levels depending on language.

Language disparities complicate token pricing

Tokens present an unusual unit for consumer pricing due to dramatic variations in tokenizer efficiency across languages. Romance languages require between 1.5 and 1.7 tokens per word, Germanic languages between 1.7 and 1.9, Slavic languages between 2.2 and 2.5, and Uralic and Baltic languages between 2.7 and 3.0. A monthly token plan sold across the European Union would deliver approximately 2.5 times less service in Athens than in Dublin for the same price.

These details were first reported by The Next Web.