S&P webinar: Data center power demand to more than double by 2030

➤ AI data center racks consume as much power as 80 to 100 homes, with some new campuses requiring gigawatt-scale power equivalent to that of entire cities.

➤ Hyperscalers have contracted 135 GW of clean energy capacity, with 70% contracted in the past two years.

➤ Power unavailability poses severe business risks, with AI training runs not easily checkpointed when interrupted.

Data center power demand is set to more than double between 2026 and 2030 as AI infrastructure expansion creates significant challenges for power grids and utilities, experts said during the "AI Infrastructure Growing Pains: Ensuring Energy Resilience Amid Grid Constraints and Public Demands" webinar by S&P Global Market Intelligence and S&P Global Energy Horizons.

"We're talking about a data center campus with equivalent energy consumption of Orlando, Florida," said Dan Thompson, principal research analyst at 451 Research from S&P Global Energy Horizons, referring to gigawatt-scale facilities under construction. "So remember all of Disney, all of Universal Studios, airports, restaurants, all the things that are in Orlando, Florida."

The US dominates global data center capacity with 62,242 megawatts as of March 2026, far surpassing China's 34,015 MW. By 2030, US capacity is projected to reach 151,734 MW, with Northern Virginia alone expected to reach 17,578 MW.

Hyperscalers drive clean energy surge

The four major hyperscalers — Amazon.com Inc., Google LLC, Meta Platforms Inc. and Microsoft Corp. — have quintupled their contracted clean energy capacity since 2022, reaching nearly 135 GW by February 2026, according to Adam Wilson, senior research analyst at S&P Global Energy Horizons.

"Seventy percent of the tech industry's carbon-free capacity in the US has been announced in just the last two years," Wilson said. The technology sector now accounts for 87% to 90% of all US corporate clean energy procurement.

Amazon leads with 41 GW of global clean energy capacity, while Microsoft has closed the gap by adding 34 GW since 2024. Recent activity shows no signs of slowing, with Meta adding 3.5 GW of US capacity and Google announcing 2,700 MW in Michigan in the past three months alone.

Solar has emerged as the dominant energy source, comprising 55% to 60% of corporate clean energy contracts, overtaking wind, which now represents less than 25%. Nuclear energy is gaining traction, with 30 GW of nuclear deals announced in the past two years as hyperscalers seek baseload power for their facilities.

Access the webinar here.

Grid constraints force alternative strategies

The unprecedented demand is straining grid infrastructure and forcing data center operators to pursue alternative power delivery strategies, said Ben Levitt, associate director of Americas Gas and Power at S&P Global Energy Horizons.

Utilities now expect approximately 1,000 terawatt-hours of new data center demand over the next decade, with 700 TWh expected in the next five years alone. "This isn't what they're considering in their pipeline. This is what they are building into their resource plans," Levitt said.

The lack of lead time has proven particularly challenging. "We've seen demand expectations on average rise by about 30% per quarter" since early 2024, Levitt noted, forcing utilities to build infrastructure that typically requires years of planning and regulatory approval.

In response, developers are increasingly turning to behind-the-meter solutions. The pipeline of projects planning to colocate power supply with data centers has reached 180 GW of maximum capacity, with about 50 GW expected in the near term over the next five years.

Power interruption risks mount

The business risks associated with power unavailability are enormous, particularly for AI workloads, said Tony Lenoir, associate director at 451 Research.

"The data center business model revolves around generating income on a per-compute basis. And so every potential interruption in the flow of electrons powering the chips behind those computes turns into revenue losses," Lenoir said.

AI data centers face particularly acute risks because they charge higher rates for GPU compute than for CPU processing. Additionally, AI training runs cannot be easily checkpointed, meaning power interruptions often require restarting jobs from the beginning.

"If the job gets cut off, say, due to a power outage, you generally have to start the whole thing all over again," Lenoir said. "And with that come multiple costs," including the initial interrupted run, the cost of rerunning the job, and the opportunity cost of delayed AI model deployment.

Data centers are now being designated as critical infrastructure in growing numbers of markets, including the US, UK, Australia and Germany. The four major US hyperscalers are expected to spend approximately $640 billion on capital expenditures in 2026, almost double their 2025 spending.

The convergence of massive power demands, grid constraints and business continuity requirements is driving a fundamental shift, according to analysts, in how data center operators view energy procurement — from a sustainability initiative to a strategic security imperative.

"Whether you call it sustainability or energy autonomy, AI data centers need solar and wind energy," Lenoir said, noting that renewable energy provides insulation against volatile international commodity markets and geopolitical disruptions. The effective closure of the Strait of Hormuz earlier this year marked the third major shock to global energy systems in six years, following COVID-19 disruptions and Russia's invasion of Ukraine.

However, behind-the-meter solutions come at a steep cost. All-in costs for on-site power generation, including redundancy and microgrid equipment, are now at least twice as high as typical retail tariffs for large loads, at about $160 to $170 per megawatt-hour. These costs have risen 30% to 50% in the past two years due to escalating equipment and labor expenses.

Most analysts expect behind-the-meter configurations to serve as interim solutions until grid capacity catches up. By 2030, behind-the-meter generation is projected to meet about a quarter of new data center demand, with most facilities eventually transitioning to grid power as their preferred long-term solution due to greater reliability, resilience and cost-effectiveness.

– Access the Market Intelligence webinar replay portal. – Access more Energy webinars here. – Dive into Technology, Media and Telecommunications content with S&P Capital IQ Pro. – Access more TMT webinars. – Dive into the S&P Global Energy Horizons page.