AI Data Center Boom Requires A Lot Of Natural Gas 

UBS expects the AI data center construction boom, which was ignited during the Trump era, to generate structural tailwinds for the U.S. economy starting in 2026. In a recent conversation, one asset manager backing a mega data center project in Texas explained it’s going to be “sprint” mode for the industry through the end of the decade. Adding this all up is not rocket science, and that’s why a new UBS note maintains a bullish outlook on data center-driven power demand, particularly for natural gas-linked utilities and midstream names.

UBS analysts, led by Manav Gupta, noted that U.S. hyperscalers are pouring hundreds of billions into AI data centers, driving an unprecedented surge in power demand.

He said NatGas infrastructure is quickly emerging as the most reliable backbone infrastructure for this explosive growth, creating a structural tailwind for nat gas-levered midstream names— The Williams Companies (WMB), Energy Transfer (ET), Kinder Morgan (KMI), DT Midstream (DTM), TC Energy (TRP), and Enbridge (ENB).

UBS also reaffirmed Bloom Energy (BE) as the top pick for on-site generation using NatGas.

“Despite some knee jerk reactions from DeepSeek and then MSFT trimming some data center capacity, we still see need for a lot of nat gas powered demand to support the new data centers that have already been announced and new projects that are currently in the works,” Gupta wrote in a recent note. 

The analyst outlined the largest data center construction projects: the numbers are staggering…

• Microsoft is investing $80B globally in AI data centers, with 50% allocated to the U.S. Projects include a $3.3B AI hub in Wisconsin and new capacity in 10 countries this past quarter

• Google has earmarked $75B in 2025 capex, most of it for servers and data centers supporting Google Cloud and DeepMind.

• Meta raised its capex outlook to up to $72B for 2025, citing increased spend for AI-optimized data centers in Louisiana and Ohio.

• Oracle is spending $40B on Nvidia chips to power the $500B OpenAI-led Stargate supercenter in Texas.

• CoreWeave will spend $20–23B this year to expand its footprint to 1.6GW in contracted capacity.

• Amazon, Apple, and Equinix all reaffirmed aggressive buildout plans, citing no slowdown in AI demand.

With expanding nuclear capacity still years away (read the theme: here & here) and renewables like solar and wind deemed unreliable, NatGas is set to serve as the near-to-medium-term backbone of AI infrastructure across Texas and the Heartland.

The analyst cited recent comments from Energy Transfer that pointed to 150 data center opportunities in just Texas, with dozens more across Oklahoma, Mississippi, and the panhandle, which bodes well for NatGas companies. 

He referenced comments from Anthropic co-founder Jack Clark, who estimated that AI data centers will require 50 GW of new baseload power by 2027—the equivalent of roughly 50 nuclear plants. With no major nuclear additions expected until the 2030s, this underscores NatGas as the most practical, reliable, and cheapest power source to fuel America’s AI data center expansion.