SenseTime CEO Xu Li declared on Saturday, 20 July 2026, at the World AI Conference (WAIC) in Shanghai that artificial intelligence commercialisation is on the cusp of a fundamental transition — from a token economy to a task economy — as the unit cost of AI tokens heads toward a structural decline.

The Token Price Collapse Thesis

Xu Li drew a direct parallel to the telecommunications industry, stating: 'The pricing of tokens will inevitably drop, much like the cost of telecoms data did two decades ago.' His argument is that once foundational models and computing infrastructure become commoditised, the marginal value of raw token consumption will erode, mirroring the collapse of per-megabyte data pricing in the early 2000s. This positions the current token-centric revenue model as a transitional phase rather than a durable commercial architecture.

What the Task Economy Means

Under the emerging task economy model, according to Xu, users would pay for the completion of complex, high-value outcomes — such as reviewing architectural blueprints or generating product videos — rather than for the raw volume of compute consumed. This represents a shift from infrastructure pricing to outcome-based pricing, a distinction that carries significant implications for how AI companies structure their revenue and how enterprises budget for AI adoption.

Why It Matters: Scale and Market Expansion

China's daily AI token consumption has already surged more than 1,000-fold over the past two years, according to official data, underscoring the scale at which the current token economy operates. Xu acknowledged that AI coding is presently among the most commercially productive verticals, but argued the next breakthrough lies in work and design scenarios that extend well beyond developers. He projected that the total addressable market for these next-wave applications could expand by at least 10 times. 'Major shifts in industry conditions can produce tenfold changes when a critical inflection point is reached,' he said, adding that professionals would eventually represent only a small fraction of the total user population in these new scenarios.

The Competitive Backdrop

SenseTime's framing arrives as major Chinese AI companies — including rivals in the large language model space — are actively doubling down on enterprise token consumption as their primary growth engine. The implicit challenge in Xu's remarks is that companies anchored purely to token-volume monetisation may find their pricing power eroded as model costs commoditise. Firms that can pivot to task-layer products and outcome-based billing stand to capture disproportionate value in the next cycle.

What's Next

The transition Xu describes is not imminent but directional — the window for companies to reposition their commercial models toward task-based offerings is opening now. Investors and enterprise buyers alike will be watching whether SenseTime can translate this strategic thesis into product revenue, and whether the broader Chinese AI sector follows suit or remains committed to token-volume growth metrics.