10JUN

xAI wins 41 gas turbines for Colossus

4 min read

10:06UTC

xAI received approval in March 2026 for 41 natural gas turbines totalling 1.2 GW to power its Colossus complex; Meta's El Paso project will run on a 366 MW behind-the-meter gas array.

← Back to Oregon bills data centres, not homes Jump to analysis ↓

IndustryDeveloping

Key takeaway

xAI, Meta and Wonder Valley are committing decades of fossil generation behind the meter to bypass the grid queue.

xAI received approval in March 2026 for 41 natural gas turbines totalling 1.2 GW to power its Colossus data centre complex outside Memphis. ¹ Meta's El Paso project will run on a 366 MW behind-the-meter (BTM) natural gas array. Kevin O'Leary's Wonder Valley campus in Alberta, announced at $70 billion, has a 7.5 GW bring-your-own-power (BYOP) gas plan attached.

"behind-the-meter" describes generation built on the customer's side of the grid connection, dispatched to serve the customer's load directly without selling into the wholesale market. "Bring-your-own-power" is the operator-led variant: a campus is sited specifically to host its own generation rather than wait for transmission upgrades. Both routes solve the same problem: years-long waits for new transmission interconnection. Both also commit the operator to a fossil-fuel asset that runs for decades. A combined-cycle gas turbine ordered today comes online late in this decade and is depreciated over a twenty- to thirty-year operating life.

Colossus is xAI's Memphis training cluster, the site that hit roughly 100,000 H100 GPUs in late 2024 and was reported at 200,000 by mid-2025. Its previous power supply attracted complaints from the Southern Environmental Law Center over Clean Air Act permitting on its existing turbines. The March 2026 approval is the formal regulatory consent for the larger 1.2 GW build-out. Meta's El Paso project sits inside ERCOT, the Texas grid that has explicitly relaxed large-load interconnection rules to attract data centre demand the rest of the US grid will not connect.

The BYOP build-out runs alongside hyperscaler net-zero commitments published to investors and ESG indices. The cooling-water and renewables work covered in their sustainability reports does not retire the gas plant the same operator just queued behind the meter. Wonder Valley's 7.5 GW would meet the entire peak demand of a mid-sized European country, attached to a single private campus. Whether this gas capacity ever runs at the contracted load factor depends on how fast renewable interconnection clears, but the contracts are being signed now, and turbine lead times mean the bet is locked in.

Deep Analysis

In plain English

Most buildings get their electricity from the public grid, the network of power stations and cables that delivers electricity to homes and businesses. Large data centres need so much power that getting connected to the grid can take three to five years. Behind-the-meter gas is a different approach: the operator builds its own gas-fired power stations on or next to the data centre site, generating electricity directly without ever connecting to the public grid. The gas comes from a pipeline; the turbines convert it to electricity on site. xAI, the AI company founded by Elon Musk, won approval for 41 of these turbines at its Colossus facility in Memphis, totalling 1.2 GW. Meta and Kevin O'Leary's planned Wonder Valley campus in Alberta are taking the same route. The advantage is speed: turbines can be delivered and installed in 18-24 months. The disadvantage is that they burn fossil fuel and emit carbon and local air pollutants.

Deep Analysis

Root Causes

Grid interconnection timelines in the US average three to five years for large industrial loads. AI data centres require power delivery on 12-18 month deployment cycles to remain competitive. The gap between what the grid can deliver and what operators require has made BTM gas economically rational even at the cost of higher per-MWh fuel prices, permitting complexity, and reputational risk.

The secondary cause is the GE Vernova backlog: gas turbines are deliverable within 18-24 months (subject to queue position), whereas new grid connections in congested areas are not. Equipment manufacturers have pulled forward delivery capacity that transmission planners cannot match.

What could happen next?

Consequence
The IEA's 15-27 GW US onsite gas projection by 2030 is now on track to be achieved early: xAI alone accounts for 1.2 GW, Meta El Paso for 0.37 GW, and Wonder Valley (if built) for 7.5 GW.
Medium term · Medium
Risk
Memphis's minor modification permitting pathway for 41 turbines is now the subject of EPA regional scrutiny; if challenged successfully, it could invalidate the approval and strand capital already committed.
Short term · Medium
Precedent
Three major BTM gas commitments in a single quarter establishes BTM self-generation as the default model for AI campuses above 300 MW, which will shape utility planning and grid investment models for the next decade.
Long term · High

Source Landscape

This story draws on neutral-leaning sources

Primary parallel: The 1980s gas co-generation boom under PURPA (the Public Utilities Regulatory Policies Act of 1978) saw industrial operators install BTM gas at scale to bypass utility rates and avoid grid interconnection costs, producing a wave of self-generation that utilities litigated for a decade.

The structural parallel is direct: the same grid-avoidance motivation, the same BTM permitting pathway, and the same regulatory gap between industrial self-generation and utility-scale generation review. Congress amended PURPA in 2005 to close that gap for large qualifying facilities, the same legislative response now being discussed for data centre BTM generation.

Counter-parallel: Norway's Mongstad refinery onsite gas generation, developed by Equinor (then Statoil) from the 1980s, shows a different outcome in a different regulatory context: sovereign-owned industrial self-generation at scale proceeded without the same community opposition because the industrial site was sited away from residential areas and the economic benefits were directly visible to the state as shareholder.

The US context differs in that the benefits accrue to private operators and the costs to public health budgets.

A 1.2 GW BTM gas installation at current Henry Hub prices costs approximately $70-80 per MWh all-in (fuel plus O&M plus capital recovery), compared with roughly $50-65 per MWh for grid power in the same region. The premium is 15-30%.

Operators accept this premium because the alternative is a three-to-five year grid queue delay, which at a data centre annualising $500-700 million of revenue per GW represents a present-value cost far exceeding the lifetime fuel premium. The BTM gas decision is financially rational at current queue lengths even before factoring in the competitive value of earlier AI capacity deployment.

Consensus view: Bloomberg NEF's power markets team assessed in Q1 2026 that behind-the-meter gas generation at data centres has crossed the threshold from exception to strategy: operators are no longer treating BTM gas as a temporary bridge to grid connection but as the permanent primary supply model for new AI-focused campuses above 300 MW. BTM gas avoids TNUoS-equivalent charges, grid interconnection queues, and capacity market obligations simultaneously.

Counter-view: The Atlantic Council's Global Energy Center argues that the BTM gas wave represents a regulatory failure rather than a rational market outcome: data centre operators are avoiding costs that should price the full externality of concentrated fossil generation near urban areas.

The Memphis xAI site sits within an environmental justice community under existing EPA air quality designations, and the 41-turbine approval bypassed the normal major source permit review via a minor modification pathway that critics argue was improper.

Key tension: Whether regulators treat BTM gas at data centres as an industrial self-generation decision (historically lightly regulated) or as a new utility-scale fossil fuel addition requiring the same scrutiny as a power station, and whether the Memphis approval sets a national precedent for the minor modification pathway.

Sources:Data Center Frontier

Mentions:Meta →Kevin O'Leary →Wonder Valley →Atlantic Council →hyperscale →Alberta →Equinor →Bloomberg →Prima →The Atlantic →Norway →xAI →ERCOT →Colossus →BYOP →behind-the-meter →El Paso →Texas →

First Reported In

Update #1 · Boom hits wall: grid says no, states freeze

Data Center Frontier· 26 Apr 2026

Read original →

Causes and effects

Caused by

GE Vernova gas backlog hits 80 GW into 2029

GE Vernova's 80 GW backlog and the IEA's 15-27 GW onsite gas projection are a direct economic consequence of operators ordering BTM gas at scale to bypass grid queues.

Occurred 1 Dec 2025

Read story →

This Event

xAI wins 41 gas turbines for Colossus

Hyperscalers are bypassing the grid physically, queuing private fossil generation no public operator approved.

Led to

UK queue: 50 GW DC vs 45 GW peak demand

The UK grid queue saturation documented at 50 GW against 45 GW peak is the structural grid constraint that operators like xAI and Meta cited as the reason to bring their own power rather than seek grid connection.

Occurred 27 Feb 2026

Read story →

DOE lets PJM switch off data centres

xAI's 41-turbine approval and Meta's 366 MW BTM gas array at El Paso established the scale of BTM generation the DOE then targeted with the Section 202(c) curtailment order.

Occurred 18 May 2026

Read story →

Different Perspectives

Data-centre developers and hyperscale operators

Hill County's rescission on 4 June, seven days after RCM Hill filed a $100 million taking suit, shows the Fifth Amendment threat is faster than any appellate route: the damages clock runs from the day the moratorium passed. Rural counties with no large-load permitting framework face a litigation bill that can exceed their entire annual budget.

Kenya and President Ruto

Kenya's suspension of the $1 billion Microsoft-G42 Olkaria project in early May applies raw-capacity logic at national scale: President Ruto stated the full 1 GW build would mean switching off half the country against a 3 GW installed base. A single hyperscale campus can consume a third of a Sub-Saharan grid with no equivalent constraint in Europe.

Denmark and Energinet

Energinet's 27 May extension of its large-load connection pause, with a 60 GW queue against 7 GW peak demand, demolishes the assumption that surplus renewable generation is a relief valve for compute demand. Denmark has more wind than it can use and still cannot connect data centres, because transmission pace is the binding constraint.

France and EDF

EDF's repurposing of the Bouchain former power-station site for SoftBank's Phase 1 campus gives France a replicable siting instrument, a brownfield nuclear connection bypass, that no other G7 grid operator can match. France's Choose France summit on 30 May secured the boom's largest European bet without a connection-queue fight or community moratorium.

SoftBank Group

SoftBank's EUR 75 billion France commitment on 30 May anchors at EDF's Bouchain nuclear baseload, bypassing the UK's four-times-US electricity cost premium (cited by OpenAI as reason to pause Cobalt Park) and Germany's grid-queue delays. EDF's supply relationship is bilateral; SoftBank never enters the French connection queue.

US residential ratepayers and state regulators

Portland General Electric's 4 June tariff is the first evidence that PJM's cost-transfer warning to governors on 19 May can run in reverse: Oregon households get a 1.3% bill reduction as data centres absorb their grid costs. The 12 other states carrying active cost-attribution bills now have a filed tariff with actual numbers to cite.