Global capital expenditure has entered uncharted territory. Nearly $5 trillion in spending looms by the end of the decade. The bulk traces to the energy transition. Yet AI infrastructure adds its own ferocious momentum.
Eli Horton, senior portfolio manager for TCW’s equity products, doesn’t mince words. “I do believe this is the largest capital cycle the global economy has ever experienced—that being the energy transition,” he told CNBC. Three forces drive it: energy security, surging electricity demand, and decarbonization. Together they unlock that staggering sum. Horton expects the surge to stretch across multiple decades.
The Overlapping Forces Reshaping Investment
After almost 20 years of flat U.S. electricity use, demand has reversed course. Manufacturing revival, broader electrification, and the AI boom provide the spark. Data centers alone will add 374 terawatt-hours of energy consumption and more than 45 gigawatts of peak load through 2035, according to S&P Global Energy CERA analysts cited in S&P Global Market Intelligence research.
Utilities have responded in force. U.S. investor-owned utilities now project $1.4 trillion in capital spending through 2030. That marks a 27% jump from last year’s forecast. The Wall Street Journal reported the figure in April, highlighting upgrades to an aging grid and new generation to feed AI facilities. Regulatory Research Associates, part of S&P Global, tracked 46 companies and forecast $1.295 trillion for 2026-2030, up from a December 2025 projection of $1.169 trillion.
Entergy raised its 2026-2028 outlook to $35.2 billion. Southern Co. lifted its plan 34% to more than $51 billion for the same stretch. American Electric Power eyes a $72 billion base case. These numbers reflect real contracts. Roughly 60% of Entergy’s large-load growth through 2029 ties to data centers.
But the picture extends far beyond the United States. Fortune detailed how the energy transition dominates the $5 trillion estimate. GE Vernova’s gas turbines have sold out through 2030. Only three companies worldwide produce them. “Now they’re sold out to 2030,” Horton said. “There’s three companies in the world that make them. They have a lot of power at the table.” Caterpillar’s equipment businesses—construction, mining, power generation—have ridden the same wave.
AI spending races alongside. Hyperscalers including Alphabet, Amazon, Meta, and Microsoft have committed hundreds of billions annually. Bank of America analysts project hyperscaler capital expenditure will exceed $800 billion this year, a 67% increase from 2025. Next year could hit $1 trillion, fueled by stronger revenue and cash flow. A sizable portion reflects higher chip prices. Still, the dollars flow to semiconductor makers, networking gear, and the physical plants that power large language models.
Recent updates reinforce the scale. TrendForce revised its forecast for the top nine cloud service providers upward to $830 billion in 2026, citing aggressive raises from Microsoft, Google, Meta, and AWS. Microsoft alone now guides toward $190 billion. Such intensity has pushed some hyperscalers’ capital expenditure above 30% of revenue. A decade ago that ratio belonged to railroads or heavy industry. Today it signals a permanent shift toward infrastructure ownership.
Power remains the binding constraint. Data centers demand constant, high-volume electricity. Utilities cannot build transmission lines or plants overnight. Permitting, supply chains, and skilled labor all slow progress. Yet the contracts keep arriving. S&P Global analysts note that large industrial loads, including new factories, compound the pressure.
Investors have taken notice. Caterpillar shares reflect the equipment orders. GE Vernova’s backlog offers multi-year visibility. Semiconductor suppliers enjoy pricing power, as Bank of America noted: “For AI semis vendors, we expect pricing power and margins to generally hold, as major compute/networking vendors could pass on cost increases to customers.”
Overlap exists. AI’s electricity appetite accelerates the very decarbonization and security investments that define the energy transition. But the two streams differ in character. Energy spending emphasizes long-lived assets—turbines, grids, renewables, storage. AI capital expenditure targets compute, accelerators, and specialized facilities. Their convergence creates a feedback loop. More compute demands more power. More power investment enables more compute.
Challenges lurk. Affordability worries mount for utility customers. Rate cases will proliferate as companies seek to recover costs. S&P Global’s research warns that “affordability remains a key issue shaping the US power and utility sector, moving from a background concern to a binding constraint.” Regulators must balance infrastructure needs against bill increases.
Geopolitical risks add another layer. Recent tensions around energy chokepoints have sharpened focus on security. Horton cited them explicitly. Supply chains for chips, rare earths, and critical minerals remain concentrated. Any disruption could alter spending timing even if the overall direction holds.
Still, the momentum appears durable. Hyperscaler earnings demonstrate accelerating revenue despite the heavy outlays. Equipment makers report sold-out books. Utility forecasts keep rising. The $5 trillion figure captures only spending through decade’s end. Horton believes the cycle spans decades.
Markets have priced in much of this optimism. Yet the physical build-out has barely begun. Transmission projects take years. New generation capacity requires even longer lead times. The coming decade will test whether capital, labor, and policy can match the ambition. If they do, the combined energy and AI investment wave could reshape not just corporate balance sheets but the composition of global GDP itself.
One thing looks clear. The era of stagnant power demand has ended. In its place stands a contest between insatiable compute growth and the infrastructure required to sustain it. Billions have already deployed. Trillions more will follow. The only question is how smoothly the pieces fit together.
WebProNews is an iEntry Publication