We know exactly what AI costs, but we are unable to measure what it produces. And that is a serious problem

The article argues that AI is creating real economic value that current statistics may miss, a gap Semianalysis calls “dark output.”

It distinguishes between AI replacing human tasks and lowering the cost of work that previously went undone, neither of which is well captured by spending or token-based measures.

This echoes past cases where GDP undercounted the impact of computing and unpaid care work, both of which produced value without showing up clearly in the data.

If AI output is being mismeasured, investors, central banks, and companies may be judging productivity and bubble risk from a distorted picture.