From FinOps to AI-Ops: Cloud Stops Being a Cost Centre

Introduction

For most of the last decade, the cloud conversation at the leadership level was almost entirely about cost. FinOps teams were built, dashboards tracked spend down to the service level, and every quarterly review included a slide on “cloud optimisation.” That conversation isn’t going away, but I’ve noticed it’s no longer the main event. The question that actually gets airtime now is different: is our cloud platform capable of being the substrate our AI strategy runs on, or is it still just where we happen to keep our servers?

The old lens no longer fully applies.

Cost optimisation made sense as the dominant framing when cloud workloads were mostly predictable web traffic, batch jobs, storage that grew linearly. You could forecast it, and FinOps existed to keep that forecast honest. AI workloads don’t behave the same way. Inference costs scale with usage in ways that are genuinely hard to predict a quarter out. A single successful agentic feature can 10x its own compute footprint in a month if adoption takes off, and no spreadsheet built for steady-state web traffic saw that coming.

That’s forced a real shift in how leadership thinks about cloud spend. It’s not that cost stops mattering it very much still does, arguably more, because the numbers grow faster. It’s that cost has to be evaluated alongside a second axis: is the platform actually built to support the kind of workload AI represents bursty, model-hungry, dependent on fast iteration or is it optimised for the workload of five years ago?

What “AI-Ops” actually looks like in practice

I don’t love the term it’s already getting overused in vendor decks but the underlying shift is real. It’s FinOps discipline extended to cover model selection, inference routing, and capacity planning for workloads that don’t follow a predictable curve. Teams doing this well are asking questions FinOps alone never had to answer: which workloads actually need the most capable (and most expensive) model tier, and which can run on something cheaper without anyone noticing the difference? Where does it make sense to reserve capacity in advance, and where do you need the flexibility of on-demand pricing because usage genuinely can’t be forecast yet?

This also changes who needs a seat at the table. Cost governance used to be mostly a conversation between finance and platform engineering. Now it needs someone who actually understands model behaviour in the room too because the biggest cost lever isn’t infrastructure sizing anymore, it’s routing the right task to the right model tier, and that’s a decision finance can’t make alone.

Where leadership tends to get this wrong

The most common mistake I see is treating this as purely a cost-control exercise bringing in the same FinOps playbook that worked for compute and storage and expecting it to work unchanged for AI spend. It doesn’t translate cleanly. Capping usage the way you’d cap a compute budget can quietly throttle the exact feature that was supposed to differentiate the product. The teams getting this right aren’t minimising AI spend; they’re making sure the spend goes to the workloads that actually justify it, and cutting the ones that don’t which is a fundamentally different exercise from a blanket cost-reduction target.

The other mistake is waiting too long to build this muscle. By the time an AI feature is in production and costs are already climbing unpredictably, you’re doing triage instead of strategy. The orgs ahead of this built cost visibility into the platform from the first pilot, not after the first surprising invoice.

Conclusion

The cloud conversation isn’t leaving cost behind it’s absorbing a new dimension it was never built to handle. Leadership teams that keep running the old FinOps playbook unchanged will find themselves either overspending on capabilities they don’t need or under-provisioning the one workload that actually matters to the business. The ones who adapt the playbook bringing model literacy into the cost conversation and building visibility from day one are the ones who’ll treat cloud spend as an investment lever rather than a recurring surprise.