Behind the Meter: Why Data Centers Are Building Their Own Power Plants

Ask a hyperscaler’s power team what keeps them up at night in 2026 and the answer isn’t chip supply anymore — it’s the queue. Across U.S. transmission territories, 2,200 gigawatts of generation and storage capacity are sitting in interconnection queues, according to Lawrence Berkeley National Laboratory, which is nearly double the roughly 1,300 GW of capacity actually installed on the grid today. The average project now waits about five years from application to commercial operation, up from under two years in 2008, and only 19% of projects that entered the queue between 2000 and 2019 ever got built. For a buyer trying to light up a new campus on a 2027 timeline, that math doesn’t work — so a growing number of operators are simply opting out of the queue altogether and building their own power plants next to the data hall.

The scale of the behind-the-meter shift

“Behind-the-meter” generation means exactly what it sounds like: a power plant built on or adjacent to the data center site, wired directly to the facility rather than through the utility’s distribution network. It was a niche backup strategy a few years ago. It isn’t anymore. Cleanview’s February 2026 tracking identified 46 U.S. data center projects, totaling 56 GW of combined behind-the-meter generation capacity, with 90% of those projects announced in 2025 alone. Three-quarters of the identified equipment is natural gas turbines or engines, and the 56 GW figure now represents roughly 30% of all planned U.S. data center capacity. Meta, Microsoft, Google, and Amazon all have announced onsite generation projects of their own, on top of the independent power developers building merchant behind-the-meter plants to sell capacity directly to whichever operator needs it fastest.

The deal flow backs up the trend. In October, Enerflex (NYSE: EFXT) announced a contract to design, engineer, fabricate, and assemble roughly 450 MW of natural gas-fired generation for a North American data center operator, explicitly marketed as “prime power that does not require grid connection,” with deliveries running 2027–2028 and a pipeline of similar opportunities exceeding 2 GW. That single order is a useful proxy for how fast this market segment has gone from theoretical to backlogged.

Why the grid alone can’t keep up

This isn’t just impatience — the queue delays are structural. FERC’s large-load interconnection reforms are aimed squarely at this problem, trying to give utilities and grid operators a faster, more predictable path to approve data center load. But rulemaking timelines and utility planning cycles still move in years, not quarters, and demand is moving faster than either. BloombergNEF’s September 2026 forecast puts U.S. data center natural gas consumption at roughly 18 billion cubic feet per day by 2035 — a figure that’s nearly double what BNEF projected just nine months earlier, and more than the combined gas consumption of Germany and Japan today. Of that total, grid-connected data centers alone are expected to add 15 Bcf/d of new demand, which BNEF describes as five times more demand growth through 2035 than every other grid-connected sector combined. Utilities planning decade-long transmission upgrades are being asked to absorb that in a fraction of the time.

For a buyer evaluating sites, this reframes the site-selection conversation. The traditional checklist — power availability, land, water, fiber — now has to weight “speed to power” as its own variable, separate from “cost of power.” A site with cheap, available land but a five-year interconnection queue is effectively not available on any timeline that matters for an AI-era lease commitment.

The catch: turbines are backlogged too

Here’s where the behind-the-meter story gets more complicated than the marketing suggests. Going around the grid doesn’t mean going around supply constraints — it just moves the bottleneck to a different manufacturer. GE Vernova’s gas turbine backlog has reached roughly 116 GW (53 GW firm, 63 GW in slot reservations), and the company is already taking reservations for 2031 delivery on some frames. Siemens Energy’s firm backlog sits near 69–70 GW, with lead times of three-plus years even on expedited orders. Mitsubishi Heavy Industries’ large-frame backlog is around 35 GW. Industry-wide average lead times have stretched from about 3.5 years in 2023 to roughly 5 years now, with some heavy-duty frames quoted out to 7 years — while global manufacturing capacity runs only 60–70 GW a year against roughly 110 GW of annual orders. Turbine equipment costs have followed the same curve: average combined-cycle project costs hit roughly $2,157 per kW in 2025, up from under $1,500 in 2023, and turbine prices are projected to approach $600/kW by the end of 2027.

The practical upshot: a buyer who assumes “we’ll just build our own gas plant” solves the timeline problem the way the press release implies needs to check the actual delivery slot, not just the announcement. Enerflex’s own deal doesn’t complete deliveries until 2028. Behind-the-meter power is a real lever, but it’s now its own supply chain with its own queue — just a shorter one than the utility’s, for now.

Emissions, permitting, and local pushback

The other complication is that “behind-the-meter” doesn’t mean “outside the regulatory system.” Onsite gas plants still need air permits, gas supply contracts or pipeline laterals, and increasingly face organized local opposition over emissions and noise — the same community friction that’s already slowing hyperscale campuses in parts of Virginia, Georgia, and the Southwest. The climate math is also becoming a visible part of the story: analysis cited by BloombergNEF estimates behind-the-meter gas generation could produce on the order of 209 million metric tons of CO2 annually at typical capacity factors, which would exceed total U.S. commercial aviation emissions (roughly 180 million metric tons). For buyers with sustainability commitments, PUE targets, or ESG-linked financing covenants, an onsite gas plant can solve a power problem while creating a disclosure problem — one that colocation tenants in multi-tenant facilities don’t typically have to own directly, since the landlord’s power mix is the landlord’s liability on paper, if not always in practice with customers and investors.

What this means if you’re buying capacity in 2026

None of this makes behind-the-meter generation the wrong call — for a hyperscaler with a specific 2027–2028 delivery target and balance sheet to self-finance a gas plant, it may be the only call that hits the timeline. But for most buyers evaluating colocation, build-to-suit, or self-build options right now, the behind-the-meter boom changes the questions worth asking a developer or landlord: What is the actual interconnection timeline for this site, with documentation from the utility, not just a verbal estimate? If onsite generation is part of the power plan, what turbine OEM and delivery slot is it actually booked against, and is that contract firm or a reservation? What’s the emissions and permitting exposure, and who holds it contractually? And is the premium for guaranteed behind-the-meter power — which is running well above historical grid power costs given current turbine pricing — actually cheaper than the holding cost of waiting in the interconnection queue for your specific load profile? Power availability has quietly become the single biggest determinant of data center delivery speed, ahead of construction labor or even chip supply, and it deserves the same diligence buyers have historically reserved for lease rate and SLA terms.

Get the full site-selection and power-diligence checklist, including the questions to put directly to developers and utilities before signing: visit the TechInfraHub Data Center Buyer’s Toolkit.

Written by

Raajeev Ratra

Data Center Infrastructure Expert | 15+ Years in DC Design, Operations & Project Management

Raajeev is a seasoned data center professional with hands-on experience in hyperscale facilities, colocation design, power & cooling infrastructure, and global DC operations. He shares practical insights to help engineers and IT leaders build better infrastructure.

Connect on LinkedIn →