Gas Is Letting Data Centers Grow Beyond the Grid (Podcast)

BloombergNEF projects data-center electricity demand of 194GW by 2035, but grid-connection delays are pushing developers toward faster on-site power solutions. The article highlights a growing shift to gas generation alongside turbines/engines, fuel cells, and battery storage to accelerate project timelines. The net implication is a more complex, higher-cost buildout for AI infrastructure where power availability—not compute—drives execution risk.

Analysis

The near-term beneficiary set is not “AI” broadly, but the picks-and-shovels layer that monetizes speed to power: heavy-duty turbines, reciprocating engines, switchgear, and integrated microgrid controls. That favors vendors with installed bases and service revenue, because the first wave of projects will prioritize uptime and commissioning certainty over theoretical efficiency; once a developer commits behind-the-meter, the follow-on revenue stream is usually maintenance, fuel optimization, and replacement parts. By contrast, regulated utilities and transmission owners may see some load growth pushed out in time, which matters because delayed interconnection can defer rate-base expansion for years.

The second-order winner is natural gas infrastructure at load centers, not necessarily upstream gas producers. If large campuses self-generate, basis-sensitive hubs near constrained regions should tighten relative to Henry Hub, while pipeline and local distribution operators with spare capacity gain leverage. The loser is the “all-in on grid buildout” narrative: transmission, transformers, and interconnection services remain structurally needed, but the addressable market is less linear because behind-the-meter generation can bridge multi-year delays and steal share from utility-delivered electrons.

The contrarian risk is that on-site gas is a temporary workaround, not a secular replacement. Emissions permitting, noise, water, air-quality, and local political opposition can become the new bottlenecks, and those constraints will be more binding in dense data-center corridors than the original grid queue. Over 6-18 months, if interconnection reform or utility buildout improves, the premium paid for emergency self-generation could compress quickly; if gas prices spike, the economics also deteriorate fast because these projects are extremely sensitive to delivered fuel and capacity-factor assumptions.

Key Decisions for Investors

• Long GE Vernova (GEV) and Caterpillar (CAT) versus short a broad utilities proxy such as XLU for a 3-6 month relative-value trade; thesis is that behind-the-meter demand shifts capex toward industrial power equipment faster than it expands regulated utility earnings. Falsify if utility load forecasts and interconnection approvals inflect materially higher within two quarters.
• Watch-list long Bloom Energy (BE) only on pullbacks: optionality on fuel-cell adoption is high, but the market will likely overpay for the theme before proof of durable backlog conversion. Use only if order-book growth and gross-margin expansion are independently confirmed.
• Long natural-gas exposure via KMI or WMB against short a power-infrastructure basket if local gas constraints and distributed generation adoption accelerate; this is a 6-12 month basis and transport story, not a pure commodity call. Falsify if gas prices rally enough to make self-generation uneconomic or if permitting slows deployments.
• Pair long industrials benefiting from power equipment demand against short utility developers most exposed to delayed data-center hookups; the cleaner expression is GEV/CAT over XLU or a regulated-utility-heavy basket. Entry should wait for evidence of announced on-site generation orders rather than headline speculation.