Will AI be cheaper in the future?

The economics of artificial intelligence are currently characterized by a sharp divergence between the plummeting unit cost of intelligence and the soaring capital expenditures required to produce it.

Driven by "LLMflation" and intense competition, the price for GPT-4 equivalent performance dropped roughly 240-fold over 18 months, with costs for state-of-the-art models falling by approximately 10x annually.

Conversely, the infrastructure needed to sustain this growth faces severe inflation, with projected data center capital requirements reaching $5.2 trillion by 2030 due to shortages in specialized hardware and energy constraints.

As a result, enterprise spending is shifting from one-time training costs to continuous inference operational expenses, prompting a move toward smaller, more efficient models to manage the total cost of ownership.

This dynamic creates a market where AI becomes a commodity for users while imposing immense financial risks on the hyperscalers building the underlying capacity.