Industry

The great inference price collapse is remaking AI business models

The cost of running frontier-quality models has fallen roughly tenfold in a year, turning products that were uneconomic in January into commodities by December.

By Tom Becker, Industry Reporter — NEW YORK

NEW YORK — The price of intelligence keeps falling through the floor. Running frontier-quality models now costs roughly a tenth of what it did a year ago, the product of better chips, distilled models and brutal competition among inference providers — and the collapse is quietly rewriting every AI business plan.

Products that were margin-negative at launch are suddenly profitable; features once rationed behind premium tiers are becoming free defaults; and startups that differentiated on 'we made it cheaper' are watching that moat evaporate quarterly.

The countervailing force is appetite: agentic workloads consume orders of magnitude more tokens than chat ever did, and total inference spending continues to rise even as unit prices crater.

Enable JavaScript to read the full story on Neural Daily News.