SenseTime Bets on a "Task Economy": AI Token Prices Are Bound to Fall

Chinese AI company SenseTime says the commercialization of artificial intelligence is moving from a "token economy" to a "task economy": as token prices fall over the next few years, industry revenue will shift from selling compute to charging for results. SenseTime CEO Xu Li said this in an interview with the South China Morning Post in July 2026.

What SenseTime Predicts

SenseTime believes the current model, in which customers pay for tokens (the units of text that language models process), is running out of steam. Once foundation models and compute become ordinary infrastructure, the company argues, real commercial value will move to the tasks AI solves for businesses, not the number of tokens it burns doing so.

SenseTime's position is a bet by one of China's leading AI companies on how the market will be structured in a few years. Key points from Xu Li's interview:

• SenseTime is one of China's largest AI companies
• Forecast: token prices will "inevitably" fall in the coming years
• The analogy: the drop in telecom data costs about 20 years ago
• The shift in model: from a "token economy" to a "task economy"
• The condition for the shift: models and compute becoming basic infrastructure

Why Token Prices Will Fall

Token prices will keep dropping because compute and access to models are turning into mass-market commodities, just as happened with mobile internet. Xu Li explicitly compares today's AI market to the telecom industry of two decades ago, when gigabytes of traffic went from a premium resource to something cheap and ubiquitous.

"Token pricing will inevitably decline, much like the cost of telecom data fell two decades ago," said

Xu Li, CEO of SenseTime, in an interview with the South China Morning Post.

For developers, this means that competing on a low per-token price alone is no longer a sustainable strategy: if the cost of the "raw material" trends toward zero, it becomes impossible to make money from it directly.

What Is the "Task Economy"?

The "task economy" is a model in which the customer pays not for the volume of tokens consumed but for specific work completed: a processed document, a closed ticket, an operation carried through to the end. In SenseTime's logic, once models and compute become basic infrastructure, the ability to deliver a task to a finished result, rather than to generate tokens, will determine who earns money from AI.

Such a shift also changes how products are built: the value comes from agentic systems and vertical solutions that take over an entire business process rather than simply returning an answer to a query. This echoes the industry's broader move toward AI agents capable of carrying out multi-step tasks.

What This Means

SenseTime's thesis fits a broader trend: as models get cheaper, the industry's profit center is shifting from selling "fuel" (tokens and compute) to selling outcomes. The winners will be those who learn to package AI into complete tasks rather than resell API access.

Frequently Asked Questions

Why will AI token prices fall?

According to SenseTime CEO Xu Li, the decline is inevitable because models and compute are becoming mass infrastructure. He compares it to the drop in telecom data costs about 20 years ago.

What does SenseTime mean by the "task economy"?

It is a monetization model in which the customer pays for a task completed by AI rather than for the number of tokens consumed. SenseTime expects this transition as foundation models become the market's standard infrastructure.