A few months ago the posture inside American companies was simple. Buy the best model money could rent, point it at everything, and let the tokens fly. The Wall Street Journal now calls the new posture “thrift-maxxing.” Teams route each task to the cheapest model that can actually finish it, mixing cheaper open models out of China with the offerings from OpenAI and Anthropic. The best is no longer the default. The best is the exception, kept for the work that earns it.
That one change rattles the math behind the lab valuations. OpenAI and Anthropic are edging toward public markets on a story of endless, undiscriminating demand for the top tier. The thrift-maxxing buyer breaks that story. A procurement lead who has learned that a two-dollar summarization does not need a twenty-dollar model is a procurement lead who will not pay twenty dollars. The IPO slides still show the curve climbing forever. The procurement lead is quietly rewriting the assumptions underneath it.
This is not the boom failing. It is the boom growing up. Railways got laid in a frenzy, then the accountants arrived to make the lines pay. Cloud went from “move everything, cost be damned” to FinOps teams clawing the bill back. The moment buyers start optimizing is the moment a technology stopped being a bet and started being a tool. Discipline in the buyer is a sign of health, not decay.
What the buyers are really buying
The same discipline shows up in the questions nobody has cleanly answered. The New York Times ran experiments to see whether AI could stand in for an office worker, and the honest result read less like a verdict than a confession: it can do a great deal, and it still needs someone watching. On a separate front, the hard part of building software has moved. The puzzle is no longer “can a model write the code” but “can you ship something you would actually trust to run.” Capability has commoditized at the bottom. Reliability has not.
So the moat was never raw capability. It was, and still is, the unglamorous part: does it behave the same way twice, and who answers when it doesn’t. The lab that earns its premium on the tasks that genuinely need it will keep that premium. The lab that was selling “always buy our best” just lost its easiest customer. Thrift does not kill the leaders. It separates the ones charging for brand from the ones charging for result.
There is a pattern underneath all of this that the room tends to laugh at before it admits. The thing that turns out to be true often gets dismissed first, by the people with the most invested in the old story. Cheap open models would never matter. The office would either vanish by spring or the whole thing was hype. Each claim got its laugh, and each is now what procurement teams act on. The laughter was the tell. When the obvious gets ridiculed, it is usually because it is about to be proven right by the people doing the work, not the people selling the dream.
The labs are not finished. The ones with taste and execution, the ones who cut ruthlessly and ship something that actually holds, will do fine. What ends is the free ride of unexamined spend. A market that makes its buyers choose is a market that has finally started to respect them.
The buyers chose. With the choice comes the consequence, and the consequence no longer stops at the lab’s door. The company that routed the cheap model owns what it produced. The team that shipped the AI-written code owns the bug. Accountability moved downstream to the people who actually use the thing, and that quiet shift is the part nobody put in the slide deck.
A technology stops being a story the moment someone has to make it pay. The AI boom just got its first real invoice, and the bill and the blame now land in the same inbox.
Leave a Reply