One AI chatbot query burns as much power as a lightbulb running for half an hour — and that's just the start of the bill

Apptio (an IBM company) Field CTO Greg Holmes puts it bluntly: "Energy has quickly become one of the defining constraints of modern digital infrastructure." A single generative AI query consumes 0.3 to 0.5 watt-hours of electricity — trivial on its own, but multiplied by billions of queries a day, it becomes one of the biggest forces shaping the economics of the entire industry, Data Centre Review reports.

The International Energy Agency projects global data center electricity consumption could double by 2030, with AI-specific power use inside those data centers tripling. Five of the biggest tech companies together spent more than $400 billion on capex in 2025, and that figure is expected to grow another 75% in 2026. Denmark offers a vivid example of grid strain: the country received requests for 60 gigawatts of capacity, against actual national peak demand of just 7 gigawatts — with roughly 14 of the requested gigawatts earmarked specifically for AI facilities.

• Energy providers vary widely in stability and carbon footprint — that already affects cost predictability and business resilience, not just sustainability reporting
• Most companies still have little to no real visibility into how much energy their AI workloads consume or where that energy comes from
• Holmes advises moving from boardroom talk about energy to real-time tracking by individual workload
• Some workloads can be flexibly scheduled into cheaper energy windows — a lever that works in both cloud and on-premises setups

“Global electricity consumption for data centres could double by 2030, with power use from AI-enabled data centres tripling”

— Greg Holmes, Field CTO, Apptio (IBM)

Holmes's practical advice: treat energy as a core efficiency metric on par with response latency or cloud cost, and put engineers and finance in the same decision loop instead of reconciling the electricity bill after the fact. Against the backdrop of New York already pausing AI data center construction and OpenAI's compute shortage persisting at least until 2028, energy is no longer a background line item — it's turning into the thing that actually decides who can afford to scale AI and who can't.

Nothing here should be taken as financial advice — just information to consider.