AI Is Creating the Biggest Power Boom in Decades — And Energy Infrastructure Is the Real Winner
Introduction – Powering the AI Revolution
The emergence of generative artificial intelligence (AI) and large‑language models has unleashed a new kind of compute demand that dwarfs the incremental additions of the past two decades. Modern AI training clusters use thousands of specialized chips and draw tens of megawatts of continuous electricity. As adoption accelerates and hyperscale operators deploy purpose‑built data centers, the sector is beginning to reshape electricity markets. After decades of flat or declining power demand in mature markets, AI is poised to create the strongest and most durable demand cycle that the energy sector has seen in decades. This report quantifies the scale of that demand, explains the technical and regulatory constraints limiting supply growth, and identifies public‑market winners across the energy value chain. The thesis is simple but powerful: energy infrastructure becomes the true “picks and shovels” of the AI revolution over the next decade, and investors who position early in generation, transmission and equipment providers can capture a multi‑year structural re‑rating.
AI‑Driven Demand Growth: Magnitude, Speed and Sources
Global and U.S. Consumption Trajectories
Data centers already consume roughly 2–3 % of the world’s electricity, but the rise of AI workloads will increase that share significantly. The International Energy Agency (IEA) estimates global data‑centre electricity consumption will grow from about 460 TWh in 2024 to more than 1,000 TWh by 2030 and around 1,300 TWh by 2035. This implies an annual growth rate of roughly 15 %, far above the 2 % average for overall power demand. Accelerated servers—GPUs designed for AI—are the dominant driver: the IEA forecasts these AI servers will account for nearly half of the net increase in data‑centre electricity use between 2024 and 2030. U.S. consumption is expected to rise even faster. The Belfer Center projects U.S. data‑centre demand to grow from 176 TWh in 2023 to between 325 TWh and 580 TWh by 2028, and the IEA sees U.S. per‑capita data‑centre consumption doubling by 2030. Goldman Sachs estimates that global installed data‑centre capacity will rise from roughly 55 GW today to 92 GW by 2027, with AI workloads increasing their share from 14 % to 27 %.
The capital commitments behind these forecasts are staggering. Brown Advisory estimates that nearly $7 trillion will be invested globally in compute power by 2030. Hyperscale operators are the largest customers: Amazon, Microsoft, Google and Meta accounted for more than 60 % of global hyperscale capacity in 2025 and are expected to spend over $350 billion on data centers this year, rising to $400 billion in 2026. These companies have signed or announced long‑term contracts for more than 50 GW of renewable power and have begun securing nuclear capacity through multi‑decade power purchase agreements (PPAs). The data‑centre growth cycle is expected to continue through 2030, with JLL forecasting that nearly 100 GW of new capacity will be added globally between 2026 and 2030, driving a 14 % compound annual growth rate for the sector.
Sources of Demand: Hyperscalers and High‑Density AI Clusters
The surge in power demand stems not simply from more data centres but from the intensity of AI workloads. Training a state‑of‑the‑art large‑language model can require tens of thousands of high‑power GPUs operating continuously for months; inference workloads remain energy‑intensive. Hyperscaler data centres can require more than a gigawatt of capacity for a single campus—enough to serve three‑quarters of a million homes. AI clusters often feature liquid‑cooled racks drawing up to 50 kW per server or more, double that of conventional cloud servers. These systems are being deployed at scale by companies such as Microsoft, Google, Amazon, Meta, Oracle and newer AI‑first firms like OpenAI and Anthropic.
Regional clustering is accelerating demand concentrations. Northern Virginia’s “Data Center Alley” hosts more than 250 facilities and represented about 26 % of Dominion Energy’s electricity sales in 2024. Texas’s ERCOT grid expects 40 GW of planned natural‑gas generation dedicated to data centres, while PJM expects 55 GW of new large loads by 2030 and 100 GW by 2037. Utilities across the U.S. report a 700 % surge in load‑addition requests; CenterPoint Energy in Texas saw requests rise from 1 GW to 8 GW between late 2023 and late 2024.
AI data‑centre growth is a global phenomenon. The U.K.’s National Grid expects to connect up to 19 GW of new capacity by 2031, half from data centres. In Asia, cloud providers are building clusters in Singapore, Malaysia and India, while in Europe the Netherlands and Ireland face moratoria due to grid constraints. Despite regional diversity, the U.S. remains the largest market, with load expected to double by 2030.
Long‑Term PPAs and Co‑Location Deals: Financial Importance
Hyperscalers are securing electricity through long‑term PPAs, often 10–20 years, to lock in price and supply. Microsoft signed a 20‑year agreement with Constellation Energy to restart the 835 MW Three Mile Island Unit 1 nuclear reactor, expected online in 2028. Amazon signed a 10‑year PPA with Talen Energy for up to 1,920 MW from the Susquehanna nuclear plant, generating about $18 billion in revenue for Talen over the contract life. Meta signed a PPA to keep Constellation’s Clinton nuclear plant operating beyond 2027. These contracts highlight the hyperscalers’ willingness to pay premium prices for reliable, carbon‑free baseload power and to commit capital up front. Co‑location deals—where data‑centre campuses are built adjacent to power plants—are growing as utilities sell excess capacity or partner with hyperscalers on SMRs and advanced reactors. Amazon, for example, is co‑locating a data centre at the Susquehanna nuclear site and exploring small modular reactors with Dominion Energy.
The financial benefits of such contracts are twofold. For utilities and independent power producers (IPPs), hyperscaler PPAs provide long‑term revenue visibility and justify investment in new generation. For hyperscalers, locking in power prices hedges against future electricity inflation and secures ESG‑compliant supply. Corporate PPAs reached a record 68 GW globally in 2024, with data centres accounting for more than 17 GW and almost 60 % of deals in the U.S.. JLL expects data‑centre operators to contract around 300 TWh of clean energy per year by 2030, compared with about 200 TWh in 2024.
