Behind-the-Meter Power: Oilfield Firms Fill AI's Turbine Gap

The grid can't keep pace, so developers are building their own

The AI data center boom has outgrown the U.S. power system's ability to deliver new capacity on schedule. SemiAnalysis projects U.S. data center additions rising from about 21 GW in 2026 to roughly 84 GW by 2030.1 Its review of some 40,000 generators finds that the industry will add less than 10 GW of gas capacity a year in 2026 and 2027. Additions only pick up from 2028, because too few plants are far enough along in construction to finish sooner.1 The firm attributes the shortfall to several overlapping constraints, not a single chokepoint. It expects behind-the-meter generation to reach 40 GW or more by 2028, covering more than half of each year's new data center capacity.1

The result is a model the industry calls Bring Your Own Generation, or BYOG. A provider installs and runs gas engines or turbines at or near a data center under a multi-year contract. The power often flows straight to the facility without touching the public grid.3 The notable shift is who is doing this work. Increasingly it is companies whose expertise comes from oilfields and LNG terminals, rather than traditional utilities or turbine makers.

Turbines became the scarce asset

The main bottleneck is hardware. TechCrunch reports that global gas-turbine orders reached about 100 GW by the end of 2025, while manufacturers can build only 60–70 GW a year. Lead times have stretched to five to seven years, and turbine prices have risen 10–20%.4 BloombergNEF estimates that the cost of building a gas power plant rose 66% in two years.4

Smaller equipment is under the same pressure. Kodiak's estimated lead time for Solar Turbines SMT130 units grew from about 24 months in September 2025 to more than 41 months by August 2026. ERock said that month that gas turbine lead times now exceed four years.3 These figures look shorter than TechCrunch's five-to-seven-year range. A reasonable explanation is that they describe different classes of machines, with large frame turbines facing the longest queues. Across every source, though, the trend points the same way.

This scarcity explains why oilfield service firms matter. Many already own fleets of engines and turbines or have priority purchasing relationships, built up for powering drilling and gas processing sites. Some are now buying equipment before they have customers. According to Measured AI's survey of eight BYOG players, Liberty has purchased engines for customers it has not yet signed. In July 2026 it formed a joint venture with PowerBridge tied to the Alpha Digital campus.3 In practice, firms like this are acting as an informal turbine supply chain, holding inventory that OEMs cannot produce quickly enough.

Venture Global's island-mode playbook

The most striking example comes from LNG. Venture Global's export terminals already run on gas generation that sits inside the facility and is disconnected from the grid. Calcasieu Pass uses a 720 MW combined-cycle plant, Plaquemines has two, and the CP2 terminal under construction is planned for 1,440 MW.2 The company lists six power islands that are ordered, under construction or operating, totaling about 4.5 GW. Roughly 2.2 GW is running today at its two exporting terminals.2

For comparison, data provider Cleanview counts only about 2 GW of behind-the-meter generation online across four data center projects nationwide as of mid-2026, against roughly 90 GW announced.2 That means one LNG developer currently operates about as much off-grid gas power as the entire U.S. data center sector. Its frame agreement for 11.52 GW of new power islands, alongside a target of up to 10 GW of compute, would leave just 1.52 GW for its other needs.2 That is the sense in which its turbines "have two jobs."

The gas question behind the turbine question

Fuel is currently cheap. Henry Hub prices sit near $3.60/MMBtu. LNG exports, not data centers, are driving most new gas demand, at 3.7 Bcf/d versus 0.5 Bcf/d for data centers in 2026.4 That balance may not last. RBC expects data center gas demand to reach 6.1 Bcf/d by 2030, while LNG exports keep setting records.4

What it adds up to

The evidence supports a clear conclusion. Through about 2028, the binding constraint on AI power is access to generating equipment, not demand or fuel. Companies that already hold turbines, engines and gas expertise, whether oilfield contractors like Liberty or vertically integrated LNG players like Venture Global, gain bargaining power that OEMs alone cannot match.

The risks are real. Equipment bought speculatively needs customers who actually sign contracts. The gap between 90 GW announced and 2 GW operating shows how much of the pipeline remains unproven. And if both AI and LNG demand keep growing, the competition for gas supply will intensify. For now, the turbines are going to whoever already has them, and much of that equipment sits with oil and gas companies.