$725 Billion in AI Spending — and Nobody Is Talking About the Power Bill
The AI arms race has a number everyone knows: $725 billion. That is what Google, Meta, Amazon and Microsoft plan to spend on AI infrastructure in 2026. It is the headline, the benchmark, the proof that Big Tech believes in artificial intelligence absolutely and without reservation.
But there is another number. One that appears in utility filings, grid operator reports and environmental disclosures — rarely in tech headlines. It is the amount of electricity all of that infrastructure will consume. And it is a number that should concern every European citizen, every pension fund manager and every policymaker sitting between Brussels and Berlin.
The Scale of What Is Being Built
The $725 billion AI infrastructure spend is so large it exceeds Switzerland's entire annual gross domestic product. It is buying servers, chips, cooling systems and warehouse-sized data centres that run continuously, around the clock, consuming power at a scale the world's electricity grids were never designed to handle.
Global data centre electricity use reached approximately 415 terawatt-hours in 2024 — and could rise to more than 1,000 terawatt-hours by 2026, largely driven by AI applications including machine learning, cloud computing and generative models.
A single advanced AI model can require between 50 million and over 100 million kilowatt-hours just for training — equal to the annual electricity use of hundreds of households. When combined with ongoing inference usage, total energy consumption rises even further.
A single AI-related task can consume up to 1,000 times more electricity than a traditional web search. That is not a rounding error. That is a structural shift in how much energy the internet requires — driven entirely by five companies building AI at a scale no regulator anticipated.
Meta's Hyperion — The Number That Stops You Cold
Meta's "Hyperion" data centre campus in Louisiana could use up to 5 gigawatts of electricity at any given time — roughly the equivalent amount of power used by 4.2 million homes.
One campus. One company. The power of 4.2 million homes.
Meta's total emissions from electricity use increased nearly 150% between 2019 and 2024. Meta is spending up to $135 billion on AI in 2026 alone — and shareholders have formally requested the company explain how it will meet its climate commitments given the massively growing energy demand from AI and data centres.
In response to growing energy demands from Meta and its peers, power utilities are building new methane gas plants and delaying the retirement of coal plants. In 2025, the US nearly tripled its gas-fired capacity in development.
The Fossil Fuel Contradiction
Every GAFAM company has made public climate commitments. Google has pledged 24/7 carbon-free energy by 2030. Amazon committed to net-zero carbon by 2040. Microsoft aims to be carbon negative by 2030.
Yet natural gas in 2024 accounted for more than 40% of electricity powering US data centres, while coal supplied 30% globally, according to the International Energy Agency.
A backlog of proposed renewable projects awaiting permission to connect to power grids — combined with efforts by the Trump administration to sideline renewable energy — may affect tech companies' climate goals and prolong reliance on fossil fuels.
The gap between what GAFAM says about climate and what GAFAM does to the grid is not a minor inconsistency. It is a structural contradiction — and it is widening with every new data centre that breaks ground.
Europe Is Not Immune
Data centres in Ireland consumed 21% of national electricity consumption in 2022 — with forecasts indicating a share of up to 32% by 2026. In Virginia, US, data centres consumed 26% of all electricity in the state in 2023.
Grid operators and local authorities warn that without investment, overloads may occur — and the costs of expansion are passed on to end consumers.
European households are already paying some of the world's highest electricity prices. AI data centres are competing for the same grid capacity — and winning, because they can pay more. The cost of the AI boom is being socialised onto electricity bills across the continent, quietly and without democratic debate.
Microsoft's European Energy Deal — And What It Reveals
Microsoft's Power Purchase Agreement with Iberdrola for 150 megawatts of dedicated wind power in Spain represents the new model: GAFAM companies are no longer waiting for grid operators to supply clean energy — they are contracting it directly, locking up renewable capacity before it reaches the market.
For smaller European companies and households, that means less renewable energy available at competitive prices. The green transition is being privatised — not by design, but by the market logic of companies with unlimited capital competing for finite clean energy supply.
What This Means for GAFAM
The energy story is the AI story that nobody in Silicon Valley wants to tell — because it has no good ending yet. $725 billion in spending requires electricity. That electricity is currently supplied, in significant part, by natural gas and coal. The climate commitments made by every GAFAM company are being stress-tested not by regulators or activists, but by the companies' own infrastructure ambitions.
In 2026, power becomes the defining intersection of AI growth and data centre operations. Electricity demand is rising faster than the US power grid — much of it built decades ago — was designed to handle.
The companies building the AI future are also, right now, burning more fossil fuels than at any point in their history. That is not a paradox. That is a choice.
The European Perspective
The EU Green Deal commits Europe to climate neutrality by 2050. The EU AI Act commits Europe to responsible AI. These two commitments are now in direct tension — because the AI infrastructure being built by GAFAM to serve European users is straining European grids, consuming European renewable energy capacity and delaying European decarbonisation targets.
Analyses indicate that although data centres currently account for only about 0.5% of global CO₂ emissions, they are one of the few sectors in which emissions are still rising, while many other sectors are expected to decarbonise. The IEA forecasts that by 2030, approximately 40% of additional energy consumption by data centres will still be supplied by gas and coal.
Brussels has the tools to act — the European Green Deal, the Energy Efficiency Directive, the AI Act. What it lacks is the political will to tell five American companies that their infrastructure plans are incompatible with European climate law. That conversation is coming. gafam.ai will be watching.