TL;DR
AI data center electricity costs now shape software pricing, vendor contracts, and automation budgets. Global data centers used about 415 TWh of power in 2024, and demand could more than double by 2030. We break down the numbers, the grid pressure, and how to plan your automation budget around it.
The Numbers Behind the Surge
Data centers turned into a front-page energy story in under three years. The IEA's Energy and AI report puts global data center use at about 415 terawatt-hours for 2024, roughly 1.5% of world electricity. Its base case sees that figure passing 945 TWh by 2030. Indeed, AI workloads drive most of that growth.
Efficiency gains hid the trend for years. Between 2015 and 2019, faster servers and cloud consolidation kept consumption nearly flat while internet traffic soared. That cushion ran out once GPU fleets arrived; since then, each new AI deployment adds real load to the grid.
The pace matters as much as the size. Global data center demand grew about 12% a year over the past five years. Power plants and transmission lines, however, take years to plan and build. In contrast, a GPU cluster can go from purchase order to full operation in months.
Consequently, demand keeps arriving faster than supply. That gap shows up as higher wholesale prices and longer connection queues. Also, utilities across North America and Europe now rank data centers among their biggest planning headaches.
The American Picture
The United States sits at the center of the boom. Lawrence Berkeley National Laboratory found that US facilities used 176 TWh in 2023, about 4.4% of national electricity. The Department of Energy projects that share could reach 6.7% to 12% by 2028. Notably, annual growth rose from 7% before 2018 to 18% through 2023, and AI servers drove the jump.
The Power Mix Problem
The fuel mix behind that electricity shapes both cost and carbon. Pew Research reports that natural gas supplies over 40% of US data center electricity, with renewables near 24% and nuclear near 20%. Meanwhile, US consumption could grow another 133% by 2030, reaching about 426 TWh. Power bills for AI infrastructure therefore track fuel prices closely.
Why Are AI Data Centers Driving Electricity Costs Higher?
AI chips draw far more power than the servers they replace, and companies keep adding them faster than grids can expand. That mismatch between fast-moving demand and slow-moving supply pushes rates upward across the grid.
Start with the hardware. Deloitte notes that a five-acre site can jump from 5 MW to 50 MW when it swaps standard servers for GPU racks. Moreover, the largest planned campuses approach 2 gigawatts each, the scale of a major nuclear plant. Deloitte also expects US AI facility demand to grow from 4 GW in 2024 to 123 GW by 2035, a thirtyfold rise.
Grids move slowly because each piece takes years to approve and build. A new transmission line needs land agreements and environmental review before construction even starts. Meanwhile, each quarter of delay leaves more demand chasing the same supply, which keeps upward pressure on prices.
Inside the Facility: Where the Power Goes
Servers take roughly half of a typical facility's electricity. Cooling and power conditioning consume most of the rest. Efficiency varies widely across operators: a facility running older cooling gear buys up to 60% more electricity for the same computing work. In particular, that overhead spread separates cost leaders from the rest.
Usage patterns matter too. Recent studies suggest inference, the everyday serving of AI answers, accounts for about 60% of AI-related energy. Training grabs headlines, yet daily queries add up to the bigger bill. Each chatbot session your team runs adds to that load.
What a Megawatt Actually Costs
A 100 MW facility running around the clock consumes about 876,000 MWh a year. At a typical industrial rate near $50 per MWh, that means roughly $44 million in annual power spending. At double that rate, the same site pays close to $88 million. Furthermore, the largest planned campuses would multiply those bills several times over. Operators chase each point of efficiency because the savings run into millions.
Grid Hotspots Feel It First
The national averages look tame next to the local numbers. Irish data centers used 22% of the country's metered electricity in 2024, up from 5% in 2015, according to official statistics. Dublin sits near 80%. Similarly, data centers consumed between 33% and 42% of electricity in Amsterdam, London, and Frankfurt in 2023.
Data center developers now shop for cheaper grids. Reuters reported in August 2026 that European AI builders favor smaller cities with cheaper grid-ready land. Powered land in core hubs such as Amsterdam runs about 2.4 million euros per megawatt, while some smaller regions charge under a tenth of that. Accordingly, builders now pick sites for power first, ahead of talent pools or tax breaks.
Canada Enters the Conversation
Canada keeps coming up in siting discussions for good reason. Provinces with large hydro surpluses, such as Quebec, British Columbia, and Manitoba, offer cheap, low-carbon power, exactly what AI builders want. In addition, cooler weather trims cooling loads for much of the year. From a Calgary vantage point, interest in Alberta and Western Canada keeps building as well, and provincial utilities will likely court these projects over the next few years. For Alberta firms, that could mean new demand on the provincial grid and, over time, fresh pressure on industrial rates.
What the 2030 Projections Show
Forecasts differ on size but agree on direction. The table below compares the major published outlooks.
| Source | Scope | Recent baseline | Projection |
|---|---|---|---|
| IEA | Global data centers | 415 TWh in 2024 | About 945 TWh by 2030 |
| LBNL and DOE | United States | 176 TWh in 2023 | 325 to 580 TWh by 2028 |
| McKinsey | United States | 147 TWh in 2023 | 606 TWh by 2030 |
| Ember | Europe | 96 TWh in 2024 | 168 TWh by 2030 |
| Deloitte | US AI facilities | 4 GW in 2024 | 123 GW by 2035 |
Treat these ranges as scenarios rather than guarantees. Efficiency gains, chip supply, and grid buildout speed all shift the curve. Still, even the low end implies the fastest electricity demand growth in decades. Plan, therefore, for rising power costs; the open variable is how fast they rise.
What Rising Power Costs Mean for Your Business
Most companies never see a data center power bill. Instead, AI data center electricity costs reach you through the price of every AI tool you subscribe to. Electricity ranks among the largest operating costs in modern facilities, so pricing pressure builds as power bills grow. Vendors pass those costs along through seat prices, usage fees, and API rates.
Regional electricity rates feel the pressure too. Utilities fund new plants and transmission lines by raising rates across their customer base. Alberta businesses know this pattern well from past industrial booms. Accordingly, treat AI spending and energy exposure as one connected budget line, not two separate ones.
Budgeting for this takes an afternoon. Start by listing every tool with AI features and its renewal date. Next, note which vendors disclose energy or infrastructure surcharges. Then set a quarterly review so price changes stop surprising you.
Practical Steps to Take This Quarter
You cannot control the grid, but you can control your exposure. We start clients with the same short list:
- Track AI subscription pricing at renewal; energy pass-through shows up there first.
- Right-size models to tasks. A small model handling invoices costs a fraction of a frontier model doing the same job.
- Batch non-urgent AI work overnight rather than running each job in real time.
- Ask vendors where their compute runs and how they manage energy efficiency.
- Ask about off-peak pricing for batch workloads; utilities now reward flexible demand.
- Finally, review contracts for price-adjustment clauses tied to infrastructure or energy costs.
Our Take: Right-Sized AI Beats Brute-Force Compute
The industry's default answer to rising demand has been more compute. We take the opposite view for business automation: the smallest system that does the job wins. Most business workflows need reliable, focused agents rather than frontier-scale models burning megawatts. Specifically, that means matching each task to the lightest model that clears the quality bar.
Efficiency also compounds over time. A workflow running on a right-sized model costs less today and shields you from tomorrow's rate increases. Then the savings stack with each run. Automation audits turn up workflows running on heavier models than the task needs. Energy-aware design has become a competitive edge, and buyers now ask about it in procurement.
The Road Ahead
Expect more regulation and more location shifts through 2027. The EU already requires facilities above 500 kW to report energy performance every year. Ireland now makes new data centers bring their own generation or storage. Similar rules will likely spread as grids tighten, though these rules take years to bite.
The same AI driving demand can also help manage it, from sharper load forecasting to smarter cooling controls. The World Economic Forum calls this the energy paradox: AI strains the grid while offering tools to stabilize it. Likewise, the IEA expects renewables to meet nearly half of the extra data center demand through 2030, though buildout speed remains the open question.
The AI buildout will keep growing, and the money will chase cheap power. Businesses that read this shift early can lock in better pricing, choose efficient vendors, and build automation that stays affordable. Ultimately, the winners will treat energy as a design constraint from day one.
