FinOps is now where AI budgets live or die | Friday, July 3, 2026

Burn more tokens, run more agents, top the internal leaderboard, call it progress. Enterprises that treated token volume as a proxy for value are now staring at budgets exhausted in a quarter and asking the question they skipped: did any of it move a number?

The FinOps Foundation says the share of finance teams managing AI spend went from 31 percent to 98 percent in two years. Gartner projects that by 2028 the cost of AI coding will surpass the salary of the developer using it. The constraint on enterprise AI is no longer capability, and no longer even deployment. It is the bill, and it just came due.

The receipts are piling up with names attached. Uber exhausted its entire 2026 AI budget in four months and capped spend at $1,500 per tool per month. The FinOps Foundation's J.R. Storment says companies started calling in April already three times over their full-year token budget. Meta took down the internal tokenmaxxing leaderboard its staff had built. Microsoft cancelled Claude Code subscriptions across several product divisions. The vendors are responding in kind: Google Cloud shipped Spend Caps that auto-pause API traffic at a budget ceiling, and Anthropic added spend alerts to Claude Enterprise.

The reframe for buyers is sharp. The problem was never that AI is expensive; it is that consumption-based pricing makes cost invisible until the invoice lands, and no output benchmark has matched the 5-to-20x jump in per-developer consumption that agentic mode produces. Gartner's projection that AI coding will out-cost the developer by 2028 is only shocking if you were not watching the meter.

The reflex this week is to clamp down, and for ungoverned budgets that is correct. But the failure was never spending; it was spending blind. The danger now is over-correction: enterprises so burned by a runaway invoice that they freeze, wait "12 to 18 months to prove ROI," and quietly cede ground to competitors who learned to spend well instead of little.

The uncomfortable truth underneath the cost panic is that value and consumption are not the same thing, in either direction. Just as heavy token use never guaranteed results, a slashed budget does not guarantee efficiency; it can just as easily starve the one deployment that was actually working. The teams that win the next year will not be the cheapest or the biggest spenders. They will be the ones who can name, in dollars and in outcomes, what each dollar of AI spend returned, and then move money toward what works. Discipline is a budget you can defend line by line, not a budget you are afraid to use.

We have watched this movie before, in every technology cycle. A new capability arrives, usage explodes because usage feels like progress, and then someone in finance asks the unglamorous question: what did we get for it? The answer is rarely nothing. But it is almost never what the consumption numbers implied, because inputs were never the point. Outcomes were. Tokenmaxxing was just the latest way to confuse motion with progress.

The good news is that this is a solvable problem, and a familiar one. You meter it, you cap it before you scale it, you tie it to a baseline you can defend, and you move money toward the deployments that actually move a number. None of that requires spending less. It requires spending on purpose.