📈  Chart of the Week 7/20/2025

By Michael Allison, CFA

Electrifying Opportunity: AI Infrastructure and a New Energy Supercycle

For most investors, the AI narrative begins and ends with NVIDIA, OpenAI, and the so-called Magnificent Seven. But beneath the surface of model training and inference lies a growing demand curve that few are pricing in fully: electricity.

According to the U.S. Energy Information Administration (EIA), the source of this week’s Chart, U.S. electricity consumption flatlined for a decade and a half (2005–2020), growing at a mere +0.1% annually. That’s now changed. Since 2020, electricity demand is accelerating and is projected to grow +1.7% per year through 2026—and potentially accelerating even further over the next several years.

Why? You guessed it—AI-focused data centers.

AI’s Hidden Cost Driver: Power

AI infrastructure is fundamentally different from traditional cloud computing. Large language models and generative AI workloads require tens of thousands of GPUs, all running at full capacity and drawing enormous amounts of electricity. A single AI training run can consume several hundred megawatt-hours (MWh)—an order of magnitude more than traditional enterprise compute.

This demand doesn’t stop at training. Inference—the process of running AI in real-time applications—is a 24/7, always-on operation. Multiply this across sectors (finance, healthcare, manufacturing, logistics), and the implication is clear: we’re likely only at the beginning of a structural demand shift for electricity.

McKinsey notes that AI-focused data centers can consume 2x to 3x more electricity per square foot than legacy data centers.

Investment Implications: The Infrastructure Opportunity

This surge in electricity demand isn’t just an interesting anecdote—it could very well be an investable theme.

• Utilities and Grid Modernization: Regulated utilities with large service territories, especially in AI-intensive regions (think Northern Virginia, Texas, the Pacific Northwest), are quietly becoming critical infrastructure partners to the AI arms race.

• Power Providers and Developers: Independent power producers (IPPs), especially those with renewable portfolios and strong interconnection queues, stand to benefit from long-term power purchase agreements (PPAs) signed with hyperscalers like Amazon, Google, and Microsoft.

• Data Center REITs: Select real estate investment trusts (REITs) specializing in high-power-density sites are experiencing secular demand tailwinds.

Microsoft alone is expected to spend $50 billion this year on AI data center infrastructure, with a sizable share earmarked for power procurement and grid partnerships.

From Utility Bills to Strategic Alliances

Perhaps most interestingly, we’re witnessing a transformation in the relationship between data center operators and electricity providers. No longer just ratepayers, AI companies are now forging long-term, vertically integrated partnerships with utilities.

We likely can expect to see:

• Joint development of energy infrastructure

• Long-dated renewable PPAs

• On-site generation and storage

Electricity can account for 20% to 40% of total operating expenses for an AI data center, depending on region and cooling method. Energy price volatility and carbon intensity are now board-level concerns for every company deploying advanced AI infrastructure.

A Final Thought

The AI theme is far deeper than many investors realize. The true enablers—energy, infrastructure, and power delivery—may be where the next leg of long-term investment opportunity resides.

I believe that we could be witnessing the birth of a new energy supercycle. And unlike the last one, this time it’s being driven not by oil or steel—but by intelligence.

Sources: Energy Information Administration, McKinsey, The Information, Data Center Frontier

Interested in reading more of Mike's weekly newsletters? Click below to view The Sunday Drive.

The Sunday Drive