In one fortnight, the people who sell compute all agreed it is running out.

Six platform executives printed the same scarcity signal in two weeks, and the bill for it is landing on the workforce.

Start with the OpenAI number, because it set the tone. On May 1, OpenAI finance chief Sarah Friar told Bloomberg the company faces a vertical wall of demand, and she was clear that the constraint is not sales, it is compute. A single gigawatt-scale data center, she said, costs roughly 50 billion dollars fully loaded and takes about three years to bring online, and she expects the shortage to run into 2027. Within days, a cluster of platform-layer executives printed the same three-part message: demand outruns capacity, capacity is now rationed, and the complexity of managing it is rising.

The signal is almost certainly correct. That is exactly why it is useful to the firms that sell advice. When compute was cheap and elastic, the pitch we will manage it for you read as a visible markup on something the client could handle. When compute is rationed, the same sentence reads as risk transfer, and risk transfer is a far easier thing to sell to a nervous board.

The cost of the scarcity is already being pushed down onto people. During the week of May 20, Meta began cutting roughly 8,000 jobs, about a tenth of its staff, and Mark Zuckerberg tied the reductions directly to the bill for AI infrastructure. Meta lifted its 2026 capital plan to as much as 145 billion dollars, the vast majority of it for compute, and the message to employees was blunt: money flowing into hardware is money not flowing into headcount. Expect the advisory version of that logic next. Cut your operations bench to fund your compute commitment, the pitch will go, then rent an AI-enabled managed service to do the same work.

For an executive, the trap is the trade itself. Fixed labor you controlled becomes variable spend you do not. The scarcity is real and worth planning around, but the cure being packaged around it quietly converts your cost into someone else's recurring revenue.