That matters because the study also points to an important split inside the broad “AI power demand” narrative. Clean-energy equities and traditional energy do not simply offer interchangeable ways to express rising electricity demand. Their market behaviour can diverge because their cash-flow profiles and financing needs differ. An ETF investor chasing the same structural theme through different vehicles may therefore be taking very different rate exposure, valuation risk and cyclical sensitivity than the headline story suggests.
In practice, this argues for looking beyond theme labels and checking overlap across holdings, sectors and factor exposures. A portfolio that combines AI, clean energy and infrastructure ETFs may look diversified on the surface while still clustering around one shared driver: expectations for data-center buildout and power availability. Rebalancing discipline matters here. If one narrative is pushing several sleeves of the portfolio at once, concentration can build gradually rather than all at once.
The deeper long-term lesson is that second-order beneficiaries are not automatically better diversifiers. They can be valuable exposures, but investors should separate the structural case for higher electricity demand from the shorter-term market forces—especially interest rates—that may dominate returns for long stretches.
Behind every AI prompt sits a heavy, physical infrastructure with a growing appetite for electricity. This could lift demand for natural gas and energy storage, renewables, grid equipment, and transmission, but each option carries different economics, physical limits, and exposure to interest rates. How, if at all, do these energy connections show up in public-market returns? The new working paper, “The AI-Energy Nexus: Data Center Demand and the Dynamics of Clean and Traditional Energy Stock Returns,” co-written by Original Postrofile.aspx?facId=15705" target="_blank" rel="noreferrer noopener" shape="rect">George Serafeim of the Climate and Sustainability Lab at the HBS AI Institute, tests that question by studying how energy investments moved alongside data center and AI-related assets from 2023 to 2025, how clean and traditional energy stocks respond differently, and how interest rates complicate those relationships.
Key Insight: The Demand is Real
“The key point is that data center load growth increased the value of any scalable, financeable, and connectable source of electricity and grid infrastructure, not only the value of a single technology.” [1]
The scale of what’s coming is hard to overstate. Worldwide data center capital expenditure jumped 51% in a single year, hitting roughly $455 billion in 2024. The International Energy Agency estimates data centers already consume around 460 terawatt-hours a year (about 2% of the world’s electricity) and could more than double by 2030. That kind of hunger rewards any power source that can actually meet demands, and clean energy has genuine advantages: it can be built in modular chunks that match phased construction, it lets tech giants lock in long-term prices and meet their emissions promises, and its deployment drives broader investments across supporting electrical hardware, equipment, and grid upgrades. But there are hard limits, too. Renewables are intermittent, and data centers need continuous, dependable power. The best solar and wind sites often sit far from where the computing happens, and connecting new projects to the grid means waiting in interconnection queues clogged with thousands of applications. This is exactly why natural gas, with its ability to ramp up on demand and sit near the load, retains a strategic foothold. Reflecting growing infrastructure demand, the co-movement between data center returns and traditional energy assets accelerated in 2025. The authors suggest this may be an early sign of a new channel as natural-gas producers explore direct supply arrangements with data-center operators.
Key Insight: Bytes vs. Volts
“[I]t is the electricity-consuming data centers, not the AI algorithms running inside them, that drive the energy market connection.” [2]
To trace the AI-energy link, the researchers analyzed 753 trading days using six market benchmarks: a global clean-energy exchange-traded fund, a global traditional-energy index, a data center and digital-infrastructure fund, an AI-enablers index, a global stock-market benchmark, and a 10-year U.S. Treasury bond index. A casual observer might assume that AI software companies drive energy stock movements. However, when the researchers ran a statistical “horse race” placing AI technology stock returns side-by-side with data center infrastructure returns, the AI measure became much weaker. After the researchers accounted for the broader market and interest rates, the AI index added no meaningful explanatory power for clean-energy returns, while the data-center measure remained significant.
Key Insight: Two Different Energy Stories
“[C]lean and traditional energy exhibit opposite interest rate sensitivities.” [3]
The data center fund, which primarily holds data-center real-estate and digital-infrastructure companies, became the key measuring stick. If markets are recognizing AI-related electricity demand, the portfolios most exposed to that demand should move together. That’s exactly what turned up, but not evenly. Clean energy stocks tracked data center stocks about twice as closely as traditional energy stocks did. In regression terms, data center returns alone explained roughly 34% of the daily ups and downs in clean energy, versus less than 9% for traditional energy. Interest rates explain another important divide. Clean energy behaved like a growth stock, thriving when rates fall and struggling when they rise. Traditional energy did the opposite. Why? Many renewable projects require substantial upfront investment and generate cash over decades. When rates rise, financing becomes more expensive and those distant future cash flows are worth less today. Just as AI ignited demand for power, the Federal Reserve staged its most aggressive rate-hiking campaign in four decades, lifting rates by 525 basis points. The two forces worked against each other: expectations of higher electricity demand supported clean energy, while rising financing costs reduced the value of long-lived projects.
Why This Matters
Most organizations will not need to make decisions about clean-energy stocks or data-center power contracts. Even so, this study offers a useful way to think about technological change: major innovations rarely remain contained within the industries that create them. AI is already generating new pressures and opportunities across industries because its growth depends on so many parts of the entire economy. For business leaders and executives, the challenge is therefore to look beyond direct applications of the technology and consider how it could reshape the customers, suppliers, and constraints surrounding their own businesses.
References
[1] Cheema-Fox, Alex, Megan Czasonis, Piyush Kontu, and George Serafeim, “The AI-Energy Nexus: Data Center Demand and the Dynamics of Clean and Traditional Energy Stock Returns,” Harvard Business School Working Paper, No. 26-096 (May 2026): 5-6.
[2] Cheema-Fox et al., “The AI-Energy Nexus,” 14.
[3] Cheema-Fox et al., “The AI-Energy Nexus,” 16.
Meet the Authors
Alex Cheema-Fox is Head of Flow and Investor Behavior Research; Sustainability Research at State Street Associates.
Megan Czasonis is Managing Director at State Street Associates.
Piyush Kontu is a Quantitative Researcher at State Street Associates.
George Serafeim is Charles M. Williams Professor of Business Administration at Harvard Business School. He co-leads the Climate and Sustainability Impact Lab at the HBS AI Institute.
The post Forget the Algorithms: Follow the Electricity appeared first on Harvard Business School AI Institute.
Enjoyed this article? Sign up for our newsletter to receive regular insights and stay connected.