Corporate buyers of artificial intelligence services are rewriting their procurement logic, moving away from high-volume token consumption — what the industry has taken to calling "tokenmaxxing" — toward disciplined, return-focused deployments. The shift lands directly on OpenAI and Anthropic, the two providers most exposed to enterprise API demand, and it carries real consequences for their growth rates. Budget restraint, not a collapse in interest, is the mechanism: customers are still buying, but they are buying differently.
The Inventory Problem: Too Many Tokens, Not Enough Returns
The tokenmaxxing era had a straightforward supply-side logic — feed the model more, get more output, repeat. Enterprises ran with it, piling tokens into workflows without a clear accounting of what each call was returning. Now the ledger is being audited. Companies are tightening AI budgets with an explicit mandate to demonstrate return on investment before committing further spend. That is a demand contraction in all but name. For providers billing per token, fewer calls per customer means less revenue per seat, even if the customer count holds steady.
OpenAI and Anthropic Carry the Most Exposure
Both OpenAI and Anthropic built their commercial cases on the premise that enterprise consumption would keep climbing. The efficiency turn does not invalidate that premise permanently, but it defers the volume ramp. Growth rates at both companies now face a moderating force that is structural rather than cyclical: enterprises are not waiting for prices to drop or models to improve before pulling back. They are recalibrating how much AI spend is justified right now, on current outputs, at current prices.
Single-Cause Skepticism
It would be too clean to call this a pure budget story. The shift toward efficiency reflects a maturation in how buyers evaluate AI tooling — moving from experimentation budgets, where spend is discretionary, to operational budgets, where every line item needs a business case. That transition was always coming. What is notable is the timing: it is arriving before either OpenAI or Anthropic has locked in the long-term enterprise contracts that would smooth out consumption volatility. The companies most likely to weather the shift are those with customers already running AI in production workflows where cutting back carries its own cost.