Topic

AI inference unit economics

Gross margin, cost structure and loss profile of serving model inference at scale.

Current clusters

build1 distinct publisher

Cutting 2,720 tokens from one prompt saves 27,200 across a ten-turn ticket

Because provider APIs are stateless, every token you remove from a static prompt is removed once per turn. That arithmetic is why the fix for a synchronous support agent looks nothing like the fix for an autonomous CI debugger.

Publishers:thenewstack.io

Reality

Evidence42
Adoption22
Hype gap+12
Incentives: Insufficient
Confidence

invest2 distinct publishers

MiniMax's revenue nearly quadrupled. Its gross profit covers 7% of its adjusted loss.

Enterprise inference sales grew about eightfold to $73.9 million, and the reported loss came in $114 million inside consensus while the adjusted loss more than doubled. Both are true.

Publishers:morningstar.comscmp.com

Reality

Evidence82
Adoption62