AI usage is quantified in “tokens,” the small chunks of data that AI systems employ to comprehend, analyze and generate language. On average, one token corresponds to roughly three‑quarters of an English word.
In 2026, AI consumption surged dramatically, and the rise in token usage far outpaced the simultaneous decline in per‑token cost. Leading corporations that have adopted AI introduced “token leaderboards” to incentivize staff to extract maximal productivity from their most advanced models.
Over the past several months, trillions—and occasionally thousands of trillions (quadrillions)—of tokens have been expended, primarily for “agentic use,” wherein autonomous agents perform tasks without human intervention. The resulting expenses have become staggering, prompting many of those firms to impose rationing on model usage.
This development suggests that there may be limits to how much work can be automated; virtual workers could prove more expensive than human employees, depending on the specific task.
A notable trend to monitor is the shift by numerous companies, including Western enterprises, toward cheaper AI alternatives derived from openly available Chinese models.
While uncertainty remains, several broad patterns are beginning to emerge.
Read the full story at BBC News →