A 20-Year Bet on AI

Nuclear is Powering Meta Platforms

The most important story in tech infrastructure this week is the landmark 20-year nuclear power agreement between Meta Platforms (META) and Constellation Energy (CEG). This deal is a game-changer—not just for AI, but for the entire US power grid and data center ecosystem. Under the terms, Meta will source up to 1,121 megawatts of carbon-free nuclear power from Constellation’s fleet, securing a stable, long-term energy supply to fuel its rapidly expanding AI data centers.^[1] This is one of the largest clean energy deals we have seen this year.

Why is this so significant? Meta’s CEO, Mark Zuckerberg, has made it clear: the company is determined not to miss the next platform shift, as it did with mobile devices in the early 2000s. By locking in nuclear power, Meta is trying to make sure their AI ambitions won’t be constrained by grid volatility, regulatory uncertainty, or carbon mandates. The deal guarantees price stability and energy security for Meta’s hyperscale data centers, which are projected to be among the most power-hungry in the world as generative AI workloads ramp up. For Constellation, the agreement brings a predictable, high-value revenue stream and cements nuclear’s role as one of the drivers for the digital economy.

This partnership is a blueprint for the next era of US infrastructure, tech giants directly underwriting the clean, reliable power needed to drive AI innovation. It’s a bullish sign for investors – proof that the intersection of digital and energy infrastructure is not just a trend, but a structural shift with decades of runway ahead.

AI Data Center Boom: Demand, Dollars, & Duration

The numbers behind the data center buildout are nothing short of staggering. U.S. data center construction backlogs have stretched to nine years, up from seven just last quarter. Eaton (ETN) reports an 11% year-over-year increase in data center orders, with a $10.1 billion backlog. The sector is tracking an 18% CAGR through 2030, with 42 new mega projects worth $169 billion added this quarter alone, bringing the total mega project pipeline to $1.8 trillion. Yet only 15% of these projects have broken ground, highlighting the immense runway ahead.

What’s driving this? The generative AI revolution. Eaton estimates that data centers are expected to see a 160% increase in global annual energy demand by 2030.^[2] In the U.S., data centers’ share of total power consumption has already jumped from 2% in 2018 to 4.4% in 2023, and projections suggest this could triple to around 12% within five years, with the lion’s share coming from AI workloads. Vertiv Holdings (VRT) is also a beneficiary as they specialize in providing mission-critical power and thermal management infrastructure to data centers. Evercore ISI estimates that Vertiv’s total addressable market is going to reach $65 billion by 2028, fueled by rising power and thermal demand trends. [HH1] 

Eaton is keeping the momentum going, like Meta, as they struck a deal with Siemens Energy (SMNEY) this week to provide power and technology to accelerate the delivery of new data center capacity. Eaton and Siemens are going to combine their grid-independent energy supplies and standardized modular power systems. The collaboration will enable the simultaneous construction of data centers and on-site power generation capabilities.

Chart 1: USA Data Center Annualized Private Construction Value ($B) – 1/14-12/24^[3]

AI Tech Titans Go All-In: Capex & Control

Meta’s nuclear deal is just one piece of a larger capital spending surge. Mark Zuckerberg has repeatedly emphasized that Meta is building out its own data centers to “control its own destiny” in the AI era. The company’s capex guidance has been raised from $64 billion to $72 billion, up from an original $60–$65 billion, with the majority earmarked for data center and AI infrastructure. Meta doesn’t want to pay “tolls” to third parties as it did during the mobile revolution—this time, it’s all about owning the ecosystem.

And Meta isn’t alone. Hyperscale capex is projected to rise over 20% year-over-year in 2025, with GPU power consumption growing by 30% annually and custom silicon shipments expected to increase tenfold from 2023 to 2027. The demand for fiber optic connectivity, advanced semiconductors, and supporting infrastructure is accelerating in lockstep.

AI Grid Power: The New Battleground

All this growth is putting immense pressure and opportunity on the U.S. electric grid. Quanta (PWR), a grid infrastructure company, just reported a record $35.3 billion backlog, with its electrical segment’s growth driven by technology, power generation, and energy storage.^[4] The fastest way to bring new energy online in the next few years? Renewables like solar, wind, natural gas, and battery storage. But the real challenge is logistics: while there’s ample power in aggregate, getting it to the right locations, where new data centers are being built, is the key hurdle.

In 2024, U.S. peak electricity demand stood at 745 GW.^[5] Now, thanks to generative AI, data centers’ annual energy use could reach between 74 and 132 gigawatts, or 6.7% to 12% of total U.S. electricity consumption, by 2028.^[6]

GE Vernova (GEV) is ramping up gas turbine production, aiming to expand to 70–80 heavy-duty units per year, up from 48, as gas power equipment orders grew 80% year-over-year. New agreements with hyperscalers are set to convert to orders in 2025-2027, highlighting the symbiotic relationship between tech and energy.

But it’s not just about gas. Renewables and battery storage are the quickest solutions to bridge the near-term gap, and the industry is investing heavily in these areas. Nuclear, as underscored by the META-CEG deal, is increasingly part of the conversation as the only scalable, dispatchable zero-carbon option for the decades ahead.

The bottom line is that the intersection of AI, data centers, and grid infrastructure is creating a self-reinforcing cycle of investment, innovation, and economic growth. We think automation around the grid and U.S. infrastructure is the next natural play in the lifecycle of U.S. mega project spend. The U.S. is uniquely positioned to lead this transformation, with deep capital markets, tech companies, and a robust (if challenged) energy sector. As the AI revolution accelerates, the winners will be those who build, power, and connect the digital infrastructure of tomorrow. We remain bullish in the sector—and excited for what’s next. The future isn’t just coming; it’s being built, megawatt by megawatt, right before our eyes.

Chart 2: Data Center Energy Consumption by Data Center Type & Equipment, Global – 2005-2024^[7]

Stephanie Link’s TV Schedule:

Sources:

^[1] Source: Bloomberg. As of June 3, 2025.

^[2] Source: Eaton Corp. As of June 3, 2025.

^[3] Source: USA Census Bureau. As of July 16, 2024.

^[4] Source: Quanta Corp. As of May 1, 2025.

^[5] Source: EIA Energy. As of October 10, 2024.

^[6] Source: Berkeley Lab Report. As of December 20, 2024.

^[7] Source: IEA. As of May 30, 2025.