AI and Data Center Power Demand: The New Driver of Energy Investment

Introduction

The explosive growth of AI data center power demand is reshaping energy markets more rapidly than any development since the US shale revolution. After two decades of essentially flat US electricity demand, data center construction driven by AI training and inference workloads is creating the first sustained, structural increase in power consumption since the early 2000s. US power demand is projected to reach 4,239 billion kWh in 2026, surpassing the 2024 record of 4,082 billion kWh, with data centers representing approximately 4% of total US electricity consumption (and growing rapidly). This article examines the scale of this demand, how hyperscalers are sourcing power, and what it means for energy investment banking across every sub-sector.

The numbers are staggering and continue to be revised upward. Global data center power consumption is expected to reach 96 GW by 2026, with AI operations consuming over 40% of that total. Worldwide, AI data centers' annual power consumption is projected to reach 90 TWh by 2026, roughly a tenfold increase from 2022 levels. Morgan Stanley forecasts US data center demand could reach 74 GW by 2028, with a projected shortfall of approximately 49 GW in available power access, meaning demand is growing far faster than the grid's ability to supply it.

The capital investment is equally massive. Hyperscalers (Microsoft, Amazon, Google, Meta) plan to spend nearly $700 billion on data center projects in 2026 alone, a figure that has stunned even energy industry participants accustomed to large capital programs. Goldman Sachs estimates approximately $720 billion will need to be spent on US grid upgrades through 2030 to accommodate the new load. The combined investment in data centers and supporting power infrastructure could exceed $3 trillion over the next decade.

The hyperscaler approach to power procurement has evolved rapidly from simply buying renewable energy certificates (RECs) to signing complex, long-term power agreements that are driving real asset investment.

Nuclear deals. The most headline-grabbing development has been hyperscaler investment in nuclear power. Microsoft signed a 20-year PPA to restart the 835 MW Three Mile Island Unit 1, with Constellation investing approximately $1.6 billion in refurbishments (plus a $1 billion DOE loan), targeting 2028 operations. Google and Kairos Power executed the first US corporate small modular reactor (SMR) fleet deal for 500 MW, with initial reactors targeted for 2030. Meta signed a 20-year agreement with Constellation Energy for nuclear power from the Clinton Power Station in Illinois and issued an RFP for 1-4 GW of new nuclear capacity. Google also pursued the restart of the Duane Arnold Nuclear Power Plant in Iowa, targeting 2028-2029.

Natural gas generation. While nuclear gets the headlines, natural gas is doing the heavy lifting in the near term. Gas plants can be permitted and built within 2-3 years, compared to 5-10+ years for new nuclear. ExxonMobil outlined plans in late 2024 to build a 1.5 GW gas-fired facility dedicated to data center power supply. Smaller gas-fired data center projects (like the Prometheus site in Ohio) are coming online in 2026. The NRG/LS Power and Constellation/Calpine deals were both driven by the strategic value of existing natural gas generation fleets in serving data center demand.

Renewables and novel technologies. Microsoft has surpassed Amazon as the largest corporate buyer of clean power, with 40 GW contracted as of September 2025. Google invested $4 billion in Arkansas solar projects in partnership with Entergy. Microsoft also signed a pioneering agreement to purchase fusion power from Helion Energy by 2028. Google backed Fervo Energy's next-generation geothermal plants in Nevada and Utah. While these contracts diversify the power mix, they cannot fully replace the need for firm, dispatchable generation (gas and nuclear) that provides 24/7 reliability.

Firm Power vs. Intermittent Power

Firm power (also called dispatchable or baseload) is electricity generation that can be relied upon 24 hours a day, 7 days a week, regardless of weather conditions. Natural gas combined-cycle plants and nuclear reactors provide firm power. Intermittent power (solar, wind) depends on weather conditions and produces electricity only when the sun shines or wind blows. Data centers require firm power because they operate continuously; even brief outages can disrupt AI training runs that cost millions of dollars. This is why hyperscalers are willing to pay premium prices for nuclear and gas-fired power, and it explains the power sector M&A supercycle: existing firm generation assets are being repriced as scarce, strategic infrastructure.

The AI power demand thesis generates advisory and capital markets mandates across every energy banking sub-sector.

Power M&A and advisory. The repricing of dispatchable generation has driven the largest power M&A wave since deregulation in the late 1990s. Beyond the headline Constellation/Calpine and NRG/LS Power deals, mid-market power transactions (gas plant portfolios, solar farms with battery storage, grid-connected energy projects) generate consistent deal flow. Banks with strong power and utilities practices (Goldman Sachs, Morgan Stanley, Lazard, Guggenheim) are seeing increased mandate activity.

Infrastructure and project finance. New gas-fired generation, battery storage, transmission lines, and grid upgrades require massive project financing. Each GW of new gas capacity requires approximately $800 million to $1.2 billion in construction capital. The cumulative investment required to meet projected data center demand creates a pipeline of project finance mandates that will persist for a decade.

Midstream and upstream. More gas-fired generation means more gas demand, which supports Henry Hub pricing, incentivizes gas-weighted drilling in the Haynesville and Marcellus/Utica, and drives midstream infrastructure expansion (pipeline capacity, processing plants, LNG feed gas systems). The EIA projects data center demand will add 2-4 Bcf/d of incremental gas consumption by 2030, a significant structural demand increase that was absent from forecasts as recently as 2023. This sub-sector linkage is the kind of analytical connection that energy interviewers expect candidates to articulate.

Grid and transmission. Perhaps the most underappreciated investment implication is the transmission infrastructure buildout required to connect new data center loads to the grid. Many proposed data center campuses are located in areas where existing transmission capacity is insufficient, creating bottlenecks that delay projects by years. FERC Order 1920 (requiring regional transmission planning for anticipated load growth) and state-level interconnection queue reforms are attempting to address the constraint, but the timeline to build new transmission lines (often 5-10 years) means the grid bottleneck will persist as a defining feature of the AI energy landscape. Banks advising on transmission infrastructure projects and regulated utility capital programs are seeing growing mandate activity in this space.

Valuation impacts. The AI demand thesis has fundamentally changed how the market values power assets. Existing gas and nuclear plants that were previously valued at 5-6x EBITDA based on commodity-exposed cash flow assumptions are now being valued at 8-12x based on scarcity premiums and long-term demand contracts. This revaluation has created significant wealth for current power asset owners and generated a wave of M&A as acquirers race to secure dispatchable capacity before prices rise further.