Bain: A $10 Billion Company's AI Run-Rate Could Hit $550 Million by 2035 Without New Cost Discipline

Bain modeled what happens to a typical $10 billion consumer products company’s annual IT run spend if AI cost management stays on its current trajectory: from roughly $250 million today to $550 million by 2035 — and even under strong cost discipline, the number still climbs to $450 million, a 75% increase. The brief, from Bain’s Danielle Burgs Escobar, Simo Zerrifi, Chris Bell, and Mac Dinsmore, argues traditional IT budgeting simply cannot manage a cost category this volatile: many organizations now exhaust their annual AI token budgets by mid-year, and only 13% of tech executives say they have adequate spending transparency today.

Bain’s fix is to extend the FinOps Foundation’s four domains — visibility, benefits, right-sizing, and operating model — specifically to AI workloads, and it backs the framework with real numbers: 10-20% documented savings from cloud infrastructure optimization, 10-30% from application rationalization, and up to 40% productivity gains in customer service and IT operations. One example cited: a global media company found tens of millions of dollars in unmanaged AI and cloud spend scattered across more than 80 separate general ledgers — a discovery that was only possible once someone built the visibility layer to look.

That visibility gap is the same one Forrester’s Greg Zorella flagged in a rate-volume-mix framework for diagnosing “runaway” token spend, arguing that blaming raw token counts is too blunt a diagnosis — cost overruns can come from unit-price changes, consumption growth, or shifts in which model tiers get used, and untangling the three requires clear category ownership and per-token-type tracking most finance teams don’t yet have. Anthropic’s own May 2026 shift to consumption-based pricing — cutting flat per-seat rates in favor of billing tied to token usage and model selection, with Agent SDK credits that expire monthly without rollover — is a live example of the volatility Bain is describing: a small group of heavy users can now swing a company’s bill in ways a seat count never could.

For consulting and enterprise-AI advisors, Bain’s framework is less a warning than a sales tool: the CFOs writing AI budgets need this vocabulary before their next renewal cycle, not after.