FinOps X Foundation: FinOps for AI comes 'before' AI for FinOps
The Computer Weekly Developer Network followed the FinOps Foundation to San Diego this month for just about the last technology conference before the summer recess, this was FinOps X 2025 and it was all about the cost of cloud and the full set of tools, technologies and tribulations that circulate throughout this space… day two was set to extend the ideals and methodologies here one stage further.
FinOps Foundation executive director J.R.Storment kicked off day two this year with a note on attendee feedback so far (not everyone loves a 7.45 am keynote) and the fact that so many use cases (among FinOps practitioners) are now touching AI.
That, of course, means AI for FinOps and FinOps for AI (the former being how AI can help FinOps and the latter being for FinOps is applied to AI workloads), so with the billions of dollars/pounds now spiralling in global cloud spend, it’s important to look at what happens next.
FinOps Foundation CTO Mike Fuller took over main stage duties to explain how he had come to the foundation after being the technical lead at Atlassian for many years.
“On the surface, it may appear that you need to learn so much when it comes to AI for FinOps, but really – under the surface – its all FinOps and you can apply many of these principles and practices right across your whole FinOps topography,” said Fuller. “Reasoning models take a lot of different paths, so working with AI means being able to work with consumption-based pricing and cost forecasting in new ways.”
As we have said then there are new terminologies to take on here, but many of the same FinOps principles still apply… what matters now is how practitioners differentiate how the FinOps practice applies to AI spend.
Define FinOps for AI Scope
Across planning and estimating, forecasting, budgeting and optimisation in FinOps for AI environments is – right now – at a particularly challenging state because FinOps practitioners are trying to lay the groundwork for understanding their AI costs before they start to think about cloud service optimisation. This will encompass complex considerations such as anomaly management and an appreciation for outliers.
Hopefully by now it’s clear, FinOps for AI needs to come before AI for FinOps i.e. we need to start from a FinOps mindset first and understand how AI is costed differently in the cloud universe before we start to think about augmenting FinOps in any way with AI automations and accelerations.
Additional cost dimensions for AI
Going forward, we need to appreciate the fact that there are additional (and different) cost dimensions for AI. We can see scenarios where firms will used enhanced cost data to pre-validate metrics in model tracking dashboards here.
As we look at AI architectures right now, organisations will be looking at the “build vs. buy” question when it comes to layout out spend plans. This means working across on-premises, co-location services, public cloud and so on. If a business is going to build and train its own models, then that’s one cost consideration, but crucially, the foundation’s Murray also says we need to think about GPU usage considerations and the power efficiency of these units as a key entry on the FinOps balance sheet.
Pepsico: medallion (data) architecture
There’s obviously a trade-off between using third-party functions and tools at this level. Guest speaker Kimberly Floss from Pepsico explained how her team had structured her company’s FinOps organisation with satellite arms dedicated to data * analytics and SAP, with FinOps in the centre.
Using a “medallion architecture” (bronze is raw, silver is enriched and gold is validated) to data quality is central to how Pepsico has created its analytics architecture.
The medallion architecture is succinctly described and detailed on Bismart here, with image source credit also.
Fuller spent some extra time talking about the FinOps Foundation’s AI certification options.
FinOps Certified
The FinOps Foundation thinks it has an answer to these challenges with its new launch of FinOps Certified: FinOps for AI. This is a new education series and certification built to help DevOps (and indeed FinOps) practitioners to understand, manage and optimise AI spend.
Extended sections of this second-day keynote featured speakers from AWS. The company is now talking about how Amazon Q developer is now featuring the Q for cost optimisation functions and help teams find savings and come up with execution plans. The company also announced I/O optimisations for Aurora alongside new cost comparison functions for AWS Cost Explorer.
NOTE: Non-hyperscaler speakers generally don’t feature on main stage presentations at FinOps X shows, it’s the big cloud service providers, major customer use case representatives and FinOps board members. This largely appears to be because (like a Kubecon CloudNativeCon, there are just so many members.
Platinum players club
With Google Cloud, Microsoft Azure and Oracle Cloud all ranking as platinum sponsors for this show (along with IBM Cloudability, Flexera and CloudHealth), the gold and silver sponsor section of this community is packed with smaller names… and some not such smaller names.
Microsoft also presented and spoke about its approach to application modernisation guidance, especially in the realm of building agentic AI services in environments where optimisations can be made to align costs to project goals. Soon to come from Redmond in this arena is Azure AI Foundry Provisions Throughput Reservation. Microsoft also says that it is taking non-technical users into account as well with its FinOps approach and providing visualisations that help leaders contribute to data-driven decisions.
Given the “progression of growth” for FinOps at a higher level that exec director Storment started off his day one keynote with, there is surely little chance of this arena failing to further grow.
FinOps Foundation executive director J.R.Storment.
