Models are becoming increasingly cheaper, but consumption is skyrocketing. Price tracking at the bottom clearly shows a unilateral plunge in the cost per million tokens. High-end models have seen their cost halved from over $2.50, even plummeting below $1, while lightweight models have been reduced to a pittance.
This precipitous price drop directly stimulated enterprise usage, as shown in the middle chart. Starting in early 2026, it surged almost vertically, exhibiting an extremely exaggerated exponential growth, indicating that large models have achieved substantial, large-scale penetration in enterprise workflows and automation scenarios.
However, the total expenditure curve did not increase exponentially in tandem with usage. Instead, in mid-2026, due to the sharp price drops, it experienced a significant period of stagnation and slight pullback, only managing to push total expenditure upwards in the second half of the year thanks to extremely large shipment volumes.
Meanwhile, the color distribution on the two bar charts also reveals a highly concentrated duopoly, with the orange and green bars capturing the vast majority of the enterprise-level market share, while other players are essentially relegated to a faint, long-tail noise at the top. Essentially, enterprises are enjoying the benefits of extremely low-cost high-performance computing power, while model suppliers are forced into a brutal battle of "price wars and relying on high volume and low profit margins."
