Whether labs pull compute back from outside customers shows in prices and quotas — resolves 2026-10-31

The debate: one view holds that selling compute to outside customers is a capital-allocation decision frontier labs can reverse every quarter, pulling capacity back for internal use whenever that pays more.

Why it matters: it decides whether supply for companies that depend on these labs' inference can shrink overnight, and the view's leading indicator is new compute rising while incremental revenue flattens.

Who is on each side: our 2026-08-29 deep dive found the labs do hold that decision in part, but the indicator cannot distinguish a pull-back from softening demand, so prices and quotas have to be watched, and between February and August 2026 no observed move fell on the pull-back side.

Our call: the claim that selling compute to outside customers is a capital-allocation decision a lab can reverse every quarter is right about who decides, but its leading indicator, new compute rising while incremental revenue flattens, has no diagnostic power. Telling a lab pulling compute back from softening demand requires watching whether flagship list prices and consumer quotas turn at the same time. Between February and August 2026, no observed move fell on the pull-back side.

Sources still being linked up — the node(s) behind this call don't have a clickable original link resolved yet. Logged for follow-up.

Only official pages count: the API pricing pages of OpenAI, Anthropic and Google, and their pages describing consumer plans and quotas (help centers or release notes); media reports do not count. If, between 2026-08-29 and 2026-10-31, none of the three both raised a flagship model's API list price and tightened consumer subscription quotas, the call is right; if any did both, it missed. Logged but not scored: a lab disclosing its inference allocation or the inference capacity it sells, monthly increments recomputed once two labs report annualized revenue on the same basis, and inference gross margins disclosed in a listing or audit.

What would overturn it

(1) A lab disclosing that its inference allocation fell from 40% to below 13.3% within four quarters, or that the absolute inference capacity it sells declined: the arithmetic argument is void and the strong reading, that labs do reverse it quarterly, holds. (2) Flagship list prices rising while consumer quotas tighten in the same period: the softening-demand explanation is contradicted by behavior, and our record of observed moves has to be read the other way. (3) A lab publicly cutting outside inference supply because internal use pays more, not because of an incident, a shortage or an expiring contract: the leading indicator regains diagnostic power. (4) Monthly increments flattening together once two labs report annualized revenue on the same basis: the observation that the two move in opposite directions is void.