AI data center and power plant co-location Market Outlook 2026–2034: Growth Drivers, Key Companies, Strategic Developments & Adoption Forecast — A Meticulous Next™ Foresight Brief

This Meticulous Next™ brief examines the co-location of AI data centers with power plants and on-site generation, where a compute operator contracts power directly from an adjacent generator, builds its own supply behind the meter, or funds new generation in exchange for faster access to electricity. It tests how far compute-for-power deals become the standard route to power for AI campuses by 2034, and what that means for generators, equipment makers, technology buyers and utilities.

The pivot began with a regulatory refusal. In November 2024 the Federal Energy Regulatory Commission rejected, on a 2-1 vote, an amended interconnection agreement that would have let Talen Energy sell 480 MW of behind-the-meter power from its Susquehanna nuclear plant to an adjacent Amazon data center. In June 2025 the parties replaced it with a front-of-the-meter power purchase agreement for up to 1,920 MW through 2042, worth about USD 18 billion to Talen over its life. The conclusion is that the buyer kept the site and the supply, and only the metering arrangement changed, so the regulator shaped the structure of the deal and not its direction.

Rules then caught up with demand. On 18 December 2025 FERC found the co-located load rules of PJM Interconnection unjust and unreasonable, created three new transmission services and set a three-year transition for existing behind-the-meter arrangements that ends on 18 December 2028. The order followed a capacity auction that cleared 6,623 MW short of the reserve target at the price cap of USD 333.44 per MW-day. The conclusion is that grid scarcity and priced access to transmission are pushing data center operators to contract generation directly.

Planning has moved fast. Behind-the-meter generation was planned at under 2 GW of proposed AI data center capacity in late 2024 and at about 48 GW in early 2026, roughly 30% of planned US data center capacity, with natural gas making up nearly 75% of the planned equipment. The conclusion is that on-site supply has moved from an exception to a routine option within about 15 months, and the first campuses are now entering service.

The brief treats AI data center and power plant co-location as a high-impact signal with an eight year horizon to 2034. It identifies which generators, equipment makers and technology buyers move first and what each means for contracts, campuses and grid service. It is a focused 108-page decision brief that presents an indicative trajectory rather than a segmented market model.

Behind-the-meter gas is the first source. The Socrates project of Williams for Meta in New Albany, Ohio, totals 400 MW across two 200 MW sites, with the south site in service in July 2026 and the north site due in late 2026, and Williams has committed USD 5.1 billion since 2025 to modular behind-the-meter gas plants. The conclusion is that gas can be delivered in months where grid connections take years, which is why it leads the first wave.

Adjacent nuclear supply is the second source. The Talen and Amazon agreement for up to 1,920 MW from Susquehanna ramps from 840–1,200 MW in 2029 to 1,680–1,920 MW in 2032 and moves the existing 300 MW co-located arrangement to a front-of-the-meter structure. The conclusion is that a buyer can keep the benefit of an adjacent plant while meeting grid rules, at the cost of paying transmission charges.

Fuel cells and hybrid supply fill the gap. Brookfield and Bloom Energy announced in 2025 a framework of up to USD 5 billion for fuel cells at AI data centers, and Crusoe and Microsoft announced a 900 MW AI campus in Abilene, Texas, in March 2026. The conclusion is that operators are combining several sources so that a campus can start on bridge power and shift to grid supply later.

Grid services are being rebuilt around large loads. FERC ordered Firm Contract Demand, Non-Firm Contract Demand and Interim Non-Firm transmission services for PJM, and Ohio and Michigan regulators approved tariffs for loads above 25 MW and 100 MW with minimum payments and long minimum terms. The conclusion is that utilities are creating standard products so that large loads pay for the capacity they reserve.

The limits are turbine supply, permitting and emissions, and the cost allocated to other customers. Large advanced-class gas turbines carry lead times of 5 to 7 years, while aeroderivative and trailer-mounted units can be deployed in months. The brief tests each limit against the cost trajectory to 2034.

The AI data center and power plant co-location market, covering power supply agreements, on-site and behind-the-meter generation, front-of-the-meter campus agreements and related interconnection and transmission services contracted between data center operators and generation owners, measured by annual contract revenue for power delivered and by on-site generation capacity commissioned for data centers, and excluding standard grid purchases from utilities, off-site virtual power purchase agreements and data center construction, is valued at USD 8–12 billion in 2026 and is projected to reach USD 30.07–45.11 billion by 2034, at a CAGR of 18%.

The 2026 base rests on power delivered under co-location agreements and on on-site generation commissioned for data centers, mostly in the United States, with early activity in Europe and Asia. Growth then follows the number of campuses entering service, the share of planned supply that is built and the transition of arrangements to the new rules in PJM.

The forecast is sensitive to turbine delivery, regulator rulings on cost allocation and permitting.

Scenarios.

The base case assumes turbine deliveries improve by the end of the decade and regulators settle cost allocation, bringing the inflection to about 2030 and a 2034 value near the middle of the range. An accelerated case assumes fast permitting and strong buyer funding, bringing the inflection to 2029 and lifting the 2034 value to the top of the range. A delayed case assumes slow turbine supply and ratepayer disputes that hold back approvals, moving the inflection to 2032 and holding growth at the bottom of the range.

Growth drivers

• Speed to power for AI campuses.
• Data center electricity demand.
• Capacity price increases in constrained grids.
• Federal and state support for large-load supply.

Enablers

• On-site generation and interconnection advances.
• Regulator rulings on co-located load.
• Contract and project evidence.
• Generator and technology buyer partnerships.

Restraints and barriers

• Cost allocation and ratepayer disputes.
• Turbine lead times and fuel supply.
• Permitting and emissions limits.
• Differences in national rules and standards.

The direction of AI data center and power plant co-location is set by the speed of AI load growth, technology buyer and generator investment, grid congestion and capacity prices, advances in on-site generation and interconnection, and the readiness of buyers, utilities and regulators to accept new supply structures.

The key participants in AI data center and power plant co-location and compute-for-power deals include Talen Energy Corporation, Constellation Energy Corporation, Vistra Corp., NRG Energy, Inc., The Williams Companies, Inc., GE Vernova Inc., Caterpillar Inc., Cummins Inc., Bloom Energy Corporation, Siemens Energy AG, Mitsubishi Heavy Industries, Ltd., Wärtsilä Oyj Abp, American Electric Power Company, Inc., Amazon.com, Inc., Microsoft Corporation, Alphabet Inc., Meta Platforms, Inc., Oracle Corporation, Equinix, Inc., Digital Realty Trust, Inc., Exelon Corporation, Dominion Energy, Inc., PPL Corporation, NextEra Energy, Inc., Duke Energy Corporation, Chevron Corporation, Energy Transfer LP, Baker Hughes Company, Rolls-Royce Holdings plc, Schneider Electric SE, Vertiv Holdings Co, Eaton Corporation plc, Brookfield Corporation, Switch, Inc., PJM Interconnection, L.L.C., the Federal Energy Regulatory Commission, the United States Department of Energy, the Electric Reliability Council of Texas, Inc., the National Association of Regulatory Utility Commissioners, the Public Utilities Commission of Ohio, the Michigan Public Service Commission and the United States Environmental Protection Agency.

The brief profiles representative participants in each archetype and assesses which are positioned to lead as campuses enter service and transition rules end.

Talen, Constellation, Vistra and NRG lead among generators, GE Vernova, Caterpillar, Cummins, Bloom Energy and Siemens Energy lead among equipment makers, and Amazon, Microsoft, Alphabet and Meta lead among technology buyers. Competitive intensity is rising as buyers compete for sites near generation and for scarce turbines.

Where value migrates.

In 2026 value sits with generators that own sites near load and with equipment makers that can deliver turbines, engines and fuel cells quickly. By 2030 it moves toward providers that combine on-site supply with grid service under the new PJM rules, because large loads must pay for the capacity they reserve. By 2034 it settles with co-location as a standard route to power for AI campuses, where site control, supply contracts and financing carry the margin. Margins hold for generators with firm supply and long contracts, and narrow for equipment makers that depend on a few buyers.

About 48 GW of behind-the-meter supply planned in early 2026, a FERC order on PJM and signed supply agreements support growth, while cost allocation disputes, turbine lead times and permitting limits hold back the pace.

Ratepayer disputes, slow turbine delivery and emissions permits could slow progress. The brief therefore tracks four indicators: on-site capacity commissioned, co-located capacity under contract, turbine delivery times and the cost of on-site power per MWh.

Bridge power and first co-located campuses first, then co-located campuses at scale, then standard route status is the expected order of change.