Surprise AI Bills Force Bank CFOs to Tighten Cost Control
CFOs at major banks received unexpected bills during the first AI adoption rush, and cost control has now become a necessity. Agentic AI — precisely the technology banks are betting on — is, according to Bloomberg, inherently difficult to… budget for, and only 12 of 50 global large banks can point to documented returns.
When banks rushed to adopt AI in the first wave of adoption, CFOs were asked to do something they rarely have to: approve spending without knowing what it would actually end up costing. Now, after too many surprise bills, finding a way to control costs has become a necessity, Bloomberg reports in an article carried by Moneycontrol on 22 October 2026.
Why Budget Discipline Is Failing
The problem is not just that AI is expensive — it is that the costs are structurally unpredictable. Much of the hope for productivity gains at banks is tied to agentic AI, which can solve complex tasks autonomously. But this type of AI is inherently difficult to budget for, because the amount of computing power each individual task consumes can vary widely, according to Bloomberg.
The difference from traditional software is mechanical: a classic IT process has an almost fixed cost per run. An AI agent carrying out a task on its own — searching, reasoning, calling tools, retrying — consumes a number of tokens that depends on how difficult the individual task turns out to be. Two seemingly identical tasks can therefore end up with very different bills, and as banks move toward agent-based workflows, token requirements vary considerably, according to the reporting.
Bain Partner: Nobody Has Control of the Costs
How this feels inside organizations was summed up by Bhavi Mehta, a partner at Bain & Co. and the firm's global leader for advanced analytics in financial services, in a quote carried by Bloomberg via Moneycontrol:
"I actually don't think anyone in the organization has good control over it — if I'm going to do XYZ, what's my cost going to be," Mehta said.
It is a striking admission from one of the industry's leading consultants: across roles — from developers to finance chiefs — banks lack a reliable answer to what a given AI task costs. Without that foundation, budgets are by definition guesses, and variances against the accounting side become surprises rather than exceptions.
The Returns Are Missing From Public View
The cost problem is measured against another concern: what the AI investments are actually delivering. Only 12 of the 50 largest global banks tracked by the AI benchmarking firm Evident have publicly attributed expected or realized returns to their AI adoption, according to Evident's latest report, as carried by Bloomberg.
That figure should be read with care: it concerns public documentation, not necessarily banks' internal circumstances. A bank can have good visibility into its returns internally without publishing the numbers. But the contrast is clear — nearly all of the largest banks are investing in AI, while only a small minority publicly ties concrete gains to those investments.
Already-Stretched IT Budgets
The AI bills are not arriving in a vacuum. This is a simmering issue at banks, where technology budgets are already under pressure from competing demands — from modernizing legacy IT to cybersecurity, according to Bloomberg. For CFOs, that means every unexpected AI expense competes directly with system modernization and security upgrades that, in the worst case, are required by regulators.
At the same time, the story lands in the middle of third-quarter reporting season, where rising AI investment was repeatedly raised as a theme — amid a broader Wall Street concern about the scale of AI spending on the buy side.
The Broader Spending Wave
That this is about more than banks is illustrated by another forecast: Bank of America estimates that Alphabet will spend roughly $199 billion in capex in 2026 and a further $301 billion in 2027 — about half a trillion dollars over two years, much of it tied to data centers, computing chips, and infrastructure for growing AI demand. The analysts also project negative free cash flow of $12.4 billion in the third quarter, versus roughly $3.9 billion expected by other analysts.
These are analyst forecasts, not confirmed results, and Alphabet's quarterly figures had not been reported when this note was prepared. The forecasts also do not directly belong to the banks' budget problems — but they show the scale of the AI investment spree that the whole ecosystem must now finance.
Open Questions
The underlying Bloomberg article, the Evident report, and the Bain interviews are available here only secondhand, via Bloomberg/Moneycontrol and further retellings. That means certain details — which banks are specifically affected, additional executive quotes, and quarter-specific figures — cannot be reliably reported, and the original Bloomberg publication time is unconfirmed beyond Moneycontrol's date.
The biggest unanswered question, however, is practical: if nobody in the organization — as Mehta says — has good control over what an AI task costs, how will banks develop the control mechanisms needed? Token monitoring, per-agent cost allocation, and tighter coupling between AI projects and financial management seem like plausible answers, but which models banks actually choose remains unknown in the public documentation.