Share with your CFO
Generative AI is crashing into enterprise finance operations with a cost structure that FinOps teams were never designed to handle. At FinOps X 2026, Jennifer Hays from Fidelity Investments and Natalie Daley from HSBC argued that AI spend governance now extends well beyond token pricing into database throughput, developer hardware, and workforce redesign. The State of FinOps 2026 report shows 98% of practitioners now track AI spend, yet coverage without comprehension is the recurring failure mode, and both executives warn the lift-and-shift mistake of early cloud adoption is already repeating itself with AI.
What this means for your business
Whether this story is about your organization depends on one question: are you treating AI spend as a line item inside existing cloud budgets, or as a fundamentally different cost structure that touches labor, data infrastructure, and software delivery cycles simultaneously? Companies that got cloud wrong in 2015 did so not from ignorance of compute pricing but from underestimating how the new cost model rewired procurement, architecture, and team design all at once. AI is doing that again, faster, with model refresh cycles now measured in weeks rather than years.
The Hays argument is the sharper one here, and it holds up. Augmenting existing workflows with AI produces incremental value; reimagining those workflows produces structural cost reduction. That distinction matters for the CFO because it determines which budget bucket AI belongs in. Augmentation looks like an operating expense with modest ROI, defensible but modest. Workflow reinvention looks like capital allocation with a longer horizon and higher variance, which means a different approval process, a different success metric, and a different conversation with the board. FinOps teams are being handed responsibility for that distinction without being explicitly chartered to make it.
The practical problem FinOps teams now face is that token costs, the per-unit pricing charged by model providers for processing text and data, are just the visible tip. Daley’s framing from HSBC is the more operationally useful one: the real governance challenge is helping engineers, HR teams, and legal functions choose the right model for the right job at the right cost-speed tradeoff. That is not a FinOps skill set today, it is a cross-functional capability that finance operations is being asked to grow into. I’d revise this read if FinOps foundations start issuing model-selection frameworks with teeth, but right now the charter is expanding faster than the competency.
Concept deep-dive: Token economics
Token economics refers to the pricing model used by AI providers where cost is calculated per token, roughly one token per three to four words processed, covering both what you send to the model and what it sends back. Think of it like a metered phone call where both speaking and listening run the clock. The business implication is that high-volume enterprise use cases, customer service bots, document processing, code generation, generate costs that compound with usage in ways that fixed software licenses never did, making spend visibility a prerequisite for any ROI claim.
Based on reporting from managing AI spend effectively in the generative AI era, originally published 2026-06-10 03:00:00.
