AI has gotten cheaper more quickly than any other transformative technology in history.
The cost of achieving a given level of AI performance has fallen about 47% per quarter since 2023, or 13× per year.
That price drop is 4x faster than DNA sequencing, 6x faster than compute, 18x faster than lithium batteries, and 54x faster than electricity.
The cost of "AI intelligence" is collapsing at an unprecedented pace, with access to large language models becoming as cheap in just three years as PCs did over a 15‑year cycle.
Since 2022, the price per million tokens on leading models has dropped over 90%, and that’s before accounting for big quality improvements, which make effective intelligence even cheaper.
At the same time, hyperscalers are on track to spend around $700–760 billion on AI infrastructure in 2026, with total AI CapEx likely exceeding $5 trillion by decade’s end.
When your product’s price falls 90%, you need usage to explode just to keep the economics stable; if price erosion outruns demand growth, the payback on that CapEx gets pushed far out.
AI semis and infra stocks are effectively a leveraged bet that demand will outpace collapsing prices — if that assumption breaks, the whole AI trade’s valuation needs to reset.
For the U.S., this isn’t just about margins; AI is framed as an existential race with China, which means a politically driven reluctance to slow infrastructure spending.
The real constraint won’t be money, but physics: power grids, land, permits, chips, memory, and the ability to turn trillions in CapEx into working compute at scale.