Greg Holmes, EMEA Field CTO at Apptio an IBM Company, explains why energy visibility must move from the engine room to the boardroom as AI changes the economics of compute.

Energy has quickly become one of the defining constraints of modern digital infrastructure. Every AI query, cloud-native application and training exercise ultimately depends on a resource that is no longer cheap, stable or predictable. Yet, despite this dependency, many organisations still cannot answer a basic question: how much energy does our technology estate consume, and what does that consumption actually cost us in real time?

The rapid expansion of AI workloads, combined with persistent volatility in global energy markets, has made this a much more pressing question. The economics of running digital infrastructure are evolving, and energy is no longer simply a background utility; it now has a direct influence on the ROI of many technology decisions.

The true cost of an AI query

AI has introduced a level of energy intensity that traditional IT has rarely had to contend with. Modern GPU clusters draw sustained, high-density power, and their consumption patterns are far less predictable than those of conventional workloads. The International Energy Agency recently highlighted the scale of this shift, reporting that the combined capital expenditure of five major technology companies exceeded $400 billion in 2025, with a further 75% increase expected in 2026. As AI adoption accelerates and energy-intensive use cases, including layers of AI agents, become more mainstream, the IEA has warned that global electricity consumption for data centres could double by 2030, with power use from AI-enabled data centres tripling over the same period.

The physical toll of these digital interactions becomes clearer when examined at the unit level. A single generative AI query can consume 0.3 to 0.5 watt-hours of electricity. In practical terms, generating one AI response uses roughly the same amount of energy as running a high-efficiency light bulb for several minutes or briefly powering a kitchen appliance. Multiply that by millions of daily queries and the impact becomes considerable.

Organisations may understand the size of their cloud bill, but they often lack visibility into the cost per query or the specific workloads driving consumption. When energy prices spike, as they recently have across Europe and other markets, the cost of running these workloads can swing dramatically. Without a clear understanding of the energy footprint behind each AI initiative, a product that appears profitable on paper can quickly become a significant financial liability.

Cloud, repatriation and the illusion of control

We’re seeing energy availability become a limiting factor for AI expansion. Denmark recently paused approvals for new hyperscale and AI data centres after receiving requests totalling 60 gigawatts of capacity, nearly nine times the country’s peak electricity demand of around 7 gigawatts. Of those requests, AI-specific facilities accounted for roughly 14 gigawatts. This is by no means an isolated case; it is an early indication of the pressure many European grids will face as AI adoption accelerates.

Some organisations are repatriating AI workloads in response to rising cloud costs, but the control they gain can be an illusion. Moving workloads on-premises can simply swap cloud-pricing volatility for exposure to unpredictable energy markets and the need to produce or source reliable energy. For companies running large GPU clusters, those fluctuations can turn into major, unexpected costs.

Public cloud isn’t immune to volatility either. Providers frequently change pricing, licensing and egress fees, and AI workloads can spike consumption far faster than teams expect. Without guardrails, a single dev team can burn through a year’s budget in weeks. Many organisations know their total cloud bill but not the workloads driving it, and it’s that lack of visibility that leaves them exposed. In addition, cloud providers may run out of reservable capacity, limit the supply of reservable instances and increase on-demand costs during peak periods.

Why energy visibility must become a first-class metric

The question is no longer simply where workloads run, but whether organisations can see what is happening. Without granular visibility into consumption, it becomes difficult to forecast accurately, identify cost-intensive services or understand the true financial impact of AI initiatives. Leaders need to know not just the total cost of their infrastructure, but what is driving that cost up hour by hour.

This matters because energy prices, and therefore the cost of compute, fluctuate throughout the day. There are periods when electricity is significantly cheaper, and others when it becomes extremely expensive. Some suppliers even offer free or negative-priced energy when renewable generation is high. If workloads do not need to run at a specific time, shifting them into these cheaper windows can materially reduce spend while making better use of available energy.

Engineering for flexibility and resilience

This is where engineering and financial strategy converge. Many workloads, particularly batch jobs or non-urgent processing tasks, can be scheduled flexibly. With the right systems in place, organisations can take advantage of cheaper energy periods, whether workloads run in the cloud or in their own data centres.

In cloud environments, this aligns naturally with FinOps practices. Cloud billing and consumption data can be viewed at an hourly level, and organisations can set alerts for sudden changes in usage. More importantly, this data can be exposed directly to developers through APIs, giving engineering teams real-time insight into the financial impact of their decisions and allowing their solutions to schedule workloads automatically according to need.

Building resilience requires reducing the traditional divide between technologists and finance leaders. Engineers need to understand the cost implications of their designs, and finance teams need to understand the operational realities of modern infrastructure. Static annual budgets no longer reflect how cloud and AI consumption behave; a more dynamic, real-time approach is becoming essential.

Engineering for the future

Technology strategy and energy strategy are now closely linked. Organisations preparing for the next decade will increasingly need to treat energy not just as a utility, but as a core measure of software efficiency and business value. With clearer visibility into digital infrastructure costs, CIOs and CFOs can build operating models that are better able to withstand market shocks and support continued innovation.