Remember the early cloud days, when "just spin up another server" felt gloriously free — right up until the invoice arrived and someone in finance started breathing heavily? Congratulations: AI is about to speedrun that exact lesson, and Goldman Sachs kindly brought the receipts.

Tokens Are the New Cloud Spend

According to a Wall Street Journal report, Goldman Sachs expects AI agents to drive a 24x increase in token consumption over the next four years. As autonomous agents get stitched into everyday business workflows, every task they perform quietly meters more tokens in the background — and unlike your interns, the meter never sleeps, takes lunch, or goes on vacation.

That's the hidden catch with agentic AI. A human asks a single question, but an agent might fire off dozens of model calls to plan, act, check its own work, second-guess itself, and try again. Multiply that across an entire organization and the token counter starts spinning like a slot machine that has forgotten how to pay out.

From Pilot Budget to Cost Control

The result is a very familiar corporate arc. AI spending is graduating from experimental "innovation budget" play money into the same rigorous cost-control territory that cloud computing occupies today — complete with dashboards, FinOps teams, and someone senior squinting at a chart and asking why the bill doubled again.

The angle most people miss is that this is actually a maturity signal, not a warning siren. Nobody builds elaborate cost-governance around a passing fad; you build it around infrastructure you can no longer imagine living without. Painful invoices are, counterintuitively, one of the clearest signs that AI has genuinely arrived and plans to stay.

The honeymoon phase of "just add more AI and see what happens" is officially ending. The spreadsheet phase is beginning — and it's going to be considerably less fun at the quarterly review.

Source: Tech Startups