Energy has quietly become one of the sharpest limits on what financial institutions can build. As banks and fintechs scale AI, real-time analytics and infrastructure that never switches off, the cost and availability of power now shape decisions that used to be made purely on compute price.

Greg Holmes, EMEA Field CTO at Apptio, spends his time helping financial institutions link technology choices to financial outcomes. His argument is that electricity has stopped being an operational footnote and started influencing cloud architecture, vendor selection and competitive position — and that most organisations are only beginning to measure it properly.

The economics of AI now move with the grid

Training and serving models is among the fastest-growing categories of cloud spend in financial services, which means any jump in electricity prices lands directly on the technology budget.

Holmes notes that recent price spikes in the UK and elsewhere have made identical workloads suddenly far more expensive to run. The response has been a harder look at model efficiency and how well GPUs are actually being used, alongside growing demand for consumption data broken down hour by hour so teams can identify cost hotspots and forecast when a given AI task stops being economically worthwhile.

There is a practical optimisation hiding in that, he adds: for a large share of queries, a cheaper model will deliver an equivalent result. Choosing it is a cost decision rather than a quality compromise. Left unmanaged, however, energy volatility becomes a genuine brake on how fast an institution can scale AI at all.

Energy profile is now a procurement question

Holmes is direct on whether efficiency should influence which cloud and AI providers a firm selects: it should, and it increasingly does.

Providers differ substantially in how they source and manage power, and those differences feed straight into cost predictability, sustainability reporting and resilience. Some run in regions with stable, low-carbon supply; others depend on grids exposed to volatility. Efficient infrastructure, transparent carbon accounting and the ability to schedule workloads intelligently have become meaningful points of differentiation rather than sustainability window dressing.

The implication is uncomfortable for buyers who have not asked the question. Signing with a provider whose energy profile you do not understand is, in his framing, an increasingly risky bet — efficiency and transparency now belong on the evaluation checklist next to performance, security and compliance.

Cloud strategy is being redrawn around power

Rising prices are already changing where and how workloads run.

Some organisations are relocating workloads to regions with cheaper or more stable electricity. Others are spreading them across multiple providers to limit exposure to any single market's volatility. Hybrid models are gaining ground too, with a number of firms revisiting private data centres to regain direct control over how their power is sourced.

Perhaps the clearest signal of the shift: FinOps teams are beginning to model cloud decisions against energy curves rather than compute pricing alone. That is also accelerating interest in workload automation — letting non-urgent AI jobs run in off-peak windows when power is cheaper, or shifting them to markets with more sustainable supply.

Scaling AI now carries a real cost premium — and the size of that premium depends on where your electricity comes from. — On why power has become a strategic variable

What it means for the UK's AI ambitions

National AI ambitions ultimately rest on affordable, reliable power, and current prices make that harder to guarantee.

With British businesses facing some of the steepest electricity costs in the developed world, scaling AI domestically carries a significant premium — with knock-on effects for competitiveness, investment decisions and the country's ability to attract global AI workloads. If volatility persists, Holmes expects some organisations to run compute in cheaper markets instead, slowing adoption at home.

He is careful not to write the position off. The UK already hosts a substantial number of data centres, including high-end cloud and AI capacity that is in strong demand; the difficulty is that this capacity is expensive and constrained by how easily additional electricity can be secured. Prioritising efficient infrastructure, modernising the grid and encouraging transparency around consumption would, in his view, keep the country in a leading position.

Where firms go next on energy security

Several shifts are already visible among institutions trying to protect both resilience and margin.

The first is investment in detailed energy visibility — understanding which workloads draw power, at what times and at what cost. The second is workload scheduling, moving flexible compute into cheaper periods or steadier regions. The third is far more rigorous vendor vetting, with energy sourcing, carbon intensity and exposure to grid shocks all now part of the assessment.

Some are going further and standing up their own private data centres to lock in capacity while addressing data sovereignty and independence. Tying the whole picture together, Holmes argues, is FinOps: the framework that helps leaders anticipate price spikes, avoid operational disruption and make sure AI growth does not outrun the power available to support it.