26% of all AI spend is being wasted.
The number comes from the Harness State of AI in FinOps, from May 2026, and it arrives with the context that makes it urgent: 1 in 5 surveyed companies already spends over $1 million a month on AI. At that tier, it means roughly $260,000 a month going down the drain with no measurable return.
The size of the invoice nobody watches
Two in three organizations spend over $250,000 a month on AI. One in five passed $1 million. And in those same companies, 52% have no clear owner for AI cost.
Add up the roles: engineering decides in 35% of cases; FinOps pays the bill in 27%. Different people, with different dashboards, making decisions about the same invoice neither sees whole. With no owner, no cadence, and no shared visibility, every forgotten instance becomes a yearly salary headed for the trash.
The paradox of 26 points
The number that explains the waste: 73% of companies have an AI cost policy, but only 47% manage to enforce it.
That is a 26-point distance between writing the rule and executing the rule. And 42% review AI costs only once a quarter, by which time the invoice has run and the forgotten resource has had a birthday.
The parallel with traditional cloud is direct and documented here before: 29% of IaaS and PaaS spend is wasted, with the same symptoms (resources with no owner, slow reviews, policy without enforcement). AI reproduced the problem with two aggravations: spending grows faster, and consumption is continuous by design. Agents iterating overnight with no ceiling are the active version of the idle instance.
Owner before tool
The temptation is to buy an AI FinOps panel. The survey points out the hole comes first: an observability tool with no owner is a panel nobody watches.
The working sequence is the inverse: name a responsible party per workload, bring whoever decides and whoever pays onto the same invoice, raise the review cadence to weekly, and only then instrument. With owner and cadence, even a simple spreadsheet finds the obvious waste. Without them, the best tool on the market is decoration.
How Tech86 closes this gap in practice
For anyone running AI in production, the house recipe has three items: a clear owner per AI workload, weekly review instead of quarterly, and governance that takes the policy off the paper with numeric limits and automatic consequences.
Whoever decides and whoever pays see the same invoice, before waste becomes culture. The practical result shows in the first month: the AI bill stops growing out of step with usage, and every real spent answers the question AI FinOps exists to ask: what did this invoice deliver?
Conclusion
26% is the average waste, not yours. Yours may be higher (if nobody owns the cost, it probably is) or near zero (if reviews are weekly and the policy has enforcement).
The good news from the report is that the root causes are administrative, not technological: owner, cadence, and enforcement. None of them needs a new tool; all of them need a decision. Next quarter’s AI bill will exist either way. The question is who will look at it, how often, and with what authority.