American Energy - the AI Data Center Demand
Calisade is an investment research firm that offers bespoke investment research to institutional clients. It also delivers weekly insights to the retail market via its Calisade Digests. Calisade focuses on creating alpha through robust macro, fundamental, technical, and sentiment research.
Introduction
In this week’s Digest, we turn to one of the most important developments in global markets today: the intersection of artificial intelligence with the energy system. AI’s demand for continuous, around-the-clock power is already stretching utilities, grids, and water resources, creating both risks and opportunities across the energy value chain.
History shows that energy bottlenecks can reshape entire market cycles — from the buildout of U.S. natural gas infrastructure in the early 2000s, to the shale boom in the 2010s, and the renewable surge of the past decade. What we are now seeing with AI and data centers may prove even more consequential, given that the load is dense, nonstop, and capital-intensive. In this Q and A, we explore why demand is accelerating, what role nuclear may play, and how investors should think about the evolving power mix.
Q1. Why does artificial intelligence create new energy demand?
Training frontier AI models can consume millions of GPU hours, with each GPU drawing 300–700 watts continuously. These workloads are hosted in hyperscale data centers that run 24/7, unlike factories or offices that can reduce usage during off-peak hours. Cooling adds further load — Microsoft disclosed that a typical hyperscale facility uses 125 million liters of water per year, while Google reported 6 billion gallons in 2023, a 15% increase from the prior year. This combination makes AI-driven data centers among the most power-hungry and least flexible consumers on the grid.
Q2. How much electricity are data centers expected to use in the coming years?
U.S. electricity demand is projected by the EIA to rise from 4,097 TWh in 2024 to 4,305 TWh in 2026 — a 5.1% increase in just two years, driven largely by data centers. Over the past decade, U.S. data-center consumption tripled, expanding at an 11.6% CAGR from 2014 to 2024. Forecasts suggest another doubling or tripling by 2028, pushing their share of national demand into high single digits or low double digits. Globally, the IEA projects data center electricity use will reach 945 TWh by 2030, more than double today and equal to the entire annual demand of Japan. McKinsey estimates global data center capacity demand will grow 19–22% annually through 2030, requiring 171–219 GW of capacity compared with about 60 GW today.
Q3. Why is nuclear power suddenly back in focus for investors?
Nuclear supplied about 19% of U.S. electricity in 2023, or roughly 770–800 TWh, despite having only 95 GW of capacity. The reliability is notable: nuclear plants operated at a 92.7% capacity factor, compared with natural gas at 54%, coal at 49%, wind at 36%, and solar at 25%. Tight markets underscore this advantage — PJM’s 2025–2026 capacity auction saw clearing prices jump tenfold to $270 per MW-day, raising total costs from $2.2 billion to $14.7 billion, directly benefiting nuclear incumbents. Big Tech has also begun backing nuclear directly, with Microsoft signing a 20-year deal in 2024 with Constellation Energy to restart Three Mile Island Unit 1 for clean, always-on power to supply its data centers.
Q4. What makes nuclear different from renewables like wind and solar?
Renewables are intermittent — wind averages a 36% capacity factor, and solar about 25% — compared with nuclear at ~93%. While batteries extend renewable supply, most commercial systems can store only 2–6 hours of electricity, insufficient for long gaps or seasonal shortages. Nuclear, by contrast, delivers firm baseload power, providing steady output regardless of weather or time of day. For AI-driven data centers that cannot afford downtime, this distinction is critical.
Q5. Why can’t renewables alone meet AI’s non-stop electricity needs yet?
The U.S. interconnection queue includes more than 2,500 GW of proposed projects, including over 1,000 GW of solar and 1,000 GW of storage. Yet historically only about 20% of queued projects are built. Wait times have stretched from under 2 years in the 2000s to nearly 5 years for projects entering in 2023. Even when built, renewables cannot deliver firm, around-the-clock power without long-duration storage, which is still in early development. As a result, natural gas continues to serve as the flexible backup that stabilizes the grid, while nuclear provides the firm, carbon-free backbone.
Closing Thoughts
AI is accelerating the demand for firm, reliable energy in ways that are reshaping the U.S. and global power landscape. While renewables remain a vital part of the growth story, utilities, nuclear operators, and gas generators are also essential in meeting nonstop electricity needs. For investors, understanding these dynamics will be critical as the energy system adapts to the AI-driven era.
Interested in going deeper on the above topics? Our FREE Basic Digest provides 3-5 pages of summarized insights to keep you informed on key themes impacting the market.
Or consider leveling up to our paid Premium Digest. Typically running 15-20 pages long, the Premium Digest goes much deeper on a given theme, expanding on historical precedents and providing a bit of visibility into Calisade's institutional insider trading research.
Calisade's Elite Digest is our institutional grade offering, typically 25-35 pages chock full of charts, graphs, historical insights, and, perhaps most importantly, stock tickers that are directly impacted by these themes.
📩 Click to explore Basic, Premium, and Elite Access and see what top investors are reading.
