AI Data Center Power
EXECUTIVE SUMMARY
AI data centers are driving unprecedented power demand, projected to consume 6.7-12.0% of U.S. electricity by 2028, up from 4.4% in 2023, necessitating $7 trillion in global capital investment through 2030. Grid constraints have emerged as the primary barrier to growth in 2026, with power shortages potentially restricting 40% of AI facilities by 2027 and creating a 49 GW U.S. shortfall by 2028. Investors should prioritize energy infrastructure and on-site generation solutions amid this structural shift.
MARKET CONTEXT
The AI boom has transformed data center power dynamics, with global electricity consumption from data centers, AI, and cryptocurrency forecasted to reach 945 TWh by 2030—equivalent to Japan's total usage—and potentially 1,050 TWh by 2026 alone. In the U.S., data center energy use surged from 76 TWh (1.9% of national consumption) in 2018 to 176 TWh (4.4%) in 2023, with projections of 325-580 TWh (6.7-12.0%) by 2028. AI-driven demand contributes nearly one-fifth of global power growth, adding 126 GW annually through 2028, nearly matching Canada's total demand, while U.S. data center capacity could hit 74 GW by 2028 amid a 49 GW power gap.
AI workloads, currently 15% of data center activity, are expected to rise to 40% by 2030, requiring rack densities of 50-70 kW—supported by only 20% of operators today—and pushing U.S. demand to 50 GW by 2030. A single AI task consumes up to 1,000 times more power than a traditional search, straining regional grids and halting new builds in areas like Arizona. Hyperscalers now demand gigawatt-scale "AI factories," accelerating direct energy procurement partnerships.
KEY DEVELOPMENTS
- Grid strain materialized in 2025-2026, shifting from theoretical risk to a commercial barrier; "speed to power" now dictates project viability, with Gartner forecasting power shortages curbing 40% of AI data centers by 2027.
- Hyperscalers are forging direct alliances with energy producers, bypassing utilities for on-site generation like nuclear (adoption up from 11% to 33% since 2023) and natural gas, with 62% exploring off-grid solutions.
- Shift to advanced architectures, including direct current systems for AI efficiency and Bloom Energy's 2026 report highlighting gigawatt-scale campuses outpacing grid delivery.
INVESTMENT IMPLICATIONS
Portfolio positioning favors energy infrastructure equities, including utilities, independent power producers, and on-site generation providers like fuel cells and small modular reactors (SMRs), as AI power needs eclipse traditional usage and sustain elevated power spreads through tiered pricing and spot markets. Allocate 10-15% to power-adjacent sectors, targeting firms enabling behind-the-meter sales and fossil-non-fossil hybrids, given Morgan Stanley's view that long-term compute demand justifies current capex without overbuild risk. Data center REITs with secured power contracts offer defensive yield, while AI hyperscalers with energy partnerships provide growth exposure; U.S. demand for 50 GW new capacity by 2028 underscores opportunities in transmission upgrades and renewables.
RISKS TO MONITOR
- Acute power shortages over the next two years, as demand surpasses supply, risking project delays and outages like AWS's October 2025 incident amid rising OpEx and regulations.
- Tighter 2026 regulations on power, space, and emissions, potentially inflating costs and slowing renewable adoption despite 27% current usage.
OUR VIEW
While consensus fixates on grid bottlenecks, the contrarian edge lies in underappreciated on-site power proliferation—direct current architectures and SMRs could resolve 30-50% of the shortfall by 2028, unlocking hyperscaler margins as partnerships commoditize energy access ahead of utility upgrades.
This research note is for informational purposes only and does not constitute investment advice. Solomon Grey Capital and its analysts may hold positions in the securities and asset classes discussed.
