A 0.01-yuan AI membership monthly card sold 50,000 copies in 10 minutes, representing roughly 2 trillion Tokens; sales of JD.com AI PCs bundled with Token packages grew 100% month over month. We note that AI computing power is turning from a technical bill into a mass-market consumer product, but promotions are moving faster than the rules governing them.
What this is
A Token is the basic billing unit for a large model’s information processing and can be compared to electricity consumption. As daily Token calls in China exceed 175 trillion, AI usage has shifted from training models toward handling real-world tasks. Tokens have therefore entered chat, writing, analysis, and office devices much like electricity bills do.
This change is not confined to membership plans. AI PCs combine local computing power with cloud-based Tokens, creating a hybrid model in which everyday tasks run locally while complex tasks are charged by usage. Consumers are no longer buying only a computer; they are also buying a capacity allowance for continued AI use.
Industry view
We believe the commercialization of Tokens shows that industry competition is shifting from model parameter counts toward delivery costs and call reliability. On-device computing can reduce spending on high-frequency tasks, while cloud computing handles complex work. Combining the two can help enterprises control both budgets and response times.
The promotional wave has also exposed three problems. First, different models use inconsistent Token measurement standards, so their allowances cannot be compared directly on price. Second, low-cost relay services may access users’ input, creating risks involving data leaks, malicious content, and cross-border data transfers. Third, rules covering allowance expiration, refund liability, and privacy remain opaque. China’s National Information Center has called for standardized measurement and billing disclosure. If the industry focuses only on low prices without answering whose servers process users’ data, users may ultimately bear the cost.
Impact on regular people
For enterprise IT: AI computing power will gradually become procurable, allocable infrastructure. Enterprises need to account simultaneously for local device investment, cloud usage, and sensitive-data boundaries, preventing costs from spiraling out of control as usage expands.
For individual professionals: When buying AI memberships or AI PCs, users should compare effective allowances, model capabilities, speed, and data terms rather than relying on advertised Token counts alone. Tasks involving customer information or company documents should not automatically be assigned to relay services of unknown provenance.
For the consumer market: AI devices may shift from competing on hardware specifications to competing on total service costs. Package renewals, call limits, and after-sales experiences will determine actual value. Cheap Tokens will broaden AI adoption, but safety and metering standards will determine whether that expansion is sustainable.