American Electric Power announced on October 14 that it would invest roughly $4 billion in a 702-megawatt fuel cell facility near Cheyenne, Wyoming, and sell the output to Google under a 20-year power purchase agreement. The project reflects a growing problem for the U.S. power sector: data centers need electricity sooner than utilities can expand generation and transmission. Large electricity buyers are turning to on-site generation to reduce their dependence on grid construction schedules. Fuel cells, previously associated with smaller installations, backup power, and corporate sustainability programs, are now being considered for much larger loads. AEP’s investment brings utility ownership to this approach, placing generation near the customer while grid development catches up.
For the Cheyenne project, proximity to the campus does not by itself establish that the fuel cells will be located on the campus, operate behind the customer’s meter, or be exempt from interconnection requirements. The broader case for on-site generation is distinct from the specific connection arrangements described here.
The Wyoming Agreements
Google will buy power from Cheyenne Fuel Cell, LLC, an entity that AEP owns and controls with a roughly 92 percent stake. Funds advised by Basalt Infrastructure Partners hold the remaining share of just under 8 percent. The plant will use Bloom Energy solid oxide fuel cells under a previously announced arrangement covering up to 1 gigawatt of AEP purchases. Construction is scheduled to begin in 2026, with full commercial operation expected by the end of 2028. The expected completion date provides a benchmark for comparison with projects dependent on new transmission, though the article does not establish whether this facility would avoid new transmission requirements. AEP plans to invest $78 billion across its system from 2026 through 2030 and serves 5.6 million customers in 11 states. The company presented the project as a way to serve customers both within and beyond its traditional service territory. Chairman and Chief Executive Bill Fehrman emphasized the role of new technologies and partnerships in delivering reliable power at scale.
The AEP agreement is Google’s second Wyoming power deal in roughly two weeks. The first, effective September 30, is with Black Hills Corporation. Black Hills will invest $1.8 billion to build 564 megawatts of natural gas generation at its Cheyenne Prairie site between 2027 and 2029. It will also manage roughly 2.1 gigawatts of third-party resources, described as turbines and fuel cells, through a privately operated microgrid under Wyoming’s Large Power Contract Service tariff. The Black Hills arrangement alone covers approximately 2.7 gigawatts of generation. The AEP project will also serve Google’s planned 2.7-gigawatt data-center campus on a 716-acre parcel in the Switchgrass Industrial Park south of Cheyenne, with full buildout extending toward 2031. The agreements show how a single campus can require several sources of firm power secured years before it reaches full capacity.
The capacity figures are not necessarily additive. The 564 megawatts of Black Hills generation and roughly 2.1 gigawatts of third-party resources total approximately 2.7 gigawatts. Whether AEP’s 702-megawatt facility is included in those third-party resources is not established here; treating it as additional capacity could double-count part of the campus’s supply.
Why Developers Are Turning to Fuel Cells
For data-center developers, the main attraction is speed. Lawrence Berkeley National Laboratory’s Queued Up 2026 analysis, published in July, found that roughly 2,061 gigawatts of generation and storage were actively seeking interconnection at the end of 2025. These figures concern generation and storage projects, not data-center load connections directly. The total fell 10 percent from the prior year, with more than 750 gigawatts withdrawn, but delays remained substantial. In the regions with available data, the median project waited well over three years to sign an interconnection agreement and more than five years to reach commercial operation. Those are separate elapsed-time measures, not consecutive stages to be added together. Of the capacity that entered queues between 2000 and 2020, only about 13 percent had come online by the end of 2025; roughly three-quarters had been withdrawn. Those historical completion rates do not establish the odds for any individual project, but they illustrate the uncertainty facing developers that need power on a fixed schedule.
On-site generation can reduce dependence on those schedules. Bloom Energy says its units can be deployed in under two months and reports installing fuel cells at Oracle Cloud Infrastructure sites in 55 days against a 90-day target. That installation speed should not be confused with the development schedule for a 702-megawatt facility, but it helps explain buyer interest. Gartner projects data-center electricity consumption of 565 terawatt-hours in 2026, up 26 percent year over year. Goldman Sachs analysts expect U.S. data-center demand to outpace available capacity through 2028, while a July 2026 analysis by Currence estimated that 30 to 50 percent of large-scale data-center capacity expected online in 2026 faced significant delays. For developers facing those constraints, fuel cells offer another route to power, particularly where a timely grid connection is unavailable.
Bloom Energy’s Growth and Exposure
Bloom Energy supplies the fuel cells for AEP’s project and several other large data-center commitments. The company reported second-quarter revenue of $1.065 billion, up 166 percent year over year and its first billion-dollar quarter. It also reported a blended gross margin of 34 percent and diluted earnings of $0.62 per share, and raised full-year revenue guidance to $3.9 billion–$4.2 billion. Oracle expanded its initial 1.2-gigawatt arrangement from last year to a master agreement covering up to 2.8 gigawatts. Bloom’s partnership with Brookfield Asset Management increased to $25 billion from $5 billion a year earlier. Equinix has deployed more than 100 megawatts of Bloom fuel cells across 19 data centers, and a May agreement with Nebius added another hyperscale customer. Including the AEP relationship, Bloom has described potential deployments in the tens of gigawatts. These agreements indicate growing interest in fuel cells for large loads, though potential deployments are not the same as completed installations.
Bloom’s growth depends in part on continued data-center spending. Its shares have risen severalfold since the start of 2025 but have also fallen sharply from their highs. If large cloud providers cut capital spending because of economic pressure or disappointing returns on AI software, orders could slow. A slowdown in demand from these customers remains a risk to the company’s growth outlook.
The Utility’s Changing Role
The Wyoming agreements also show utilities taking on new roles outside their traditional rate bases. AEP holds its fuel cell project in a separately capitalized entity with an infrastructure-fund minority partner. Black Hills’ new gas units will be owned by a nonregulated affiliate rather than its Wyoming utility, which will collect a microgrid management fee for coordinating dispatch and reliability services. Google has advanced $399 million to Black Hills for long-lead equipment, with repayment expected before July 2027. The contract is designed to protect existing utility customers from stranded-asset, early-termination, and cost-overrun exposure. Black Hills projects roughly $150 million in net income from the arrangement in 2030 and billions in unlevered free cash flow over the contract’s life.
These structures address a central question in the expansion of AI infrastructure: who pays for the power facilities it requires? Nonregulated affiliates and customer-specific tariffs can allow utilities to serve large loads while limiting existing customers’ financial exposure. Those protections will be tested if costs rise, demand falls, or a customer exits. The allocation of risk warrants scrutiny from regulators and consumer advocates. Utilities are also moving beyond selling electricity to coordinating supply and reliability across their own generation and third-party resources.
Natural Gas and Unresolved Emissions
Faster deployment does not settle the emissions question. Solid oxide fuel cells convert natural gas to electricity through an electrochemical process rather than combustion. This reduces criteria pollutants such as nitrogen oxides and can offer efficiency advantages over gas turbines, depending on the comparison. The units still consume natural gas and emit carbon dioxide. AEP plans to add carbon capture to the Cheyenne facility to deliver what it describes as near-zero-emissions power, but the cost and performance of that approach remain unproven at this scale. Without installed and demonstrated capture equipment, the planned facility would supply gas-fueled generation under a 20-year power contract.
Jesse Jenkins, the Princeton researcher and Firma Power co-founder, has criticized the Black Hills gas commitment. Contrasting it with Google’s separate 890-megawatt nuclear uprate agreement with Constellation, he argued that data centers should use wind, solar, and storage projects already waiting in interconnection queues rather than commission new gas infrastructure. His criticism highlights a limitation of dedicated generation for data centers: it can address a customer’s power needs without resolving the transmission constraints and queue delays holding back lower-carbon projects. Faster access to gas-fueled generation does not eliminate carbon emissions. For AEP’s planned Cheyenne fuel cell facility specifically, the near-zero-emissions claim depends on carbon capture that has yet to be installed.
Conclusion
The Cheyenne agreements show how fuel cells can supply large data-center loads alongside other generation. Their appeal is practical: they offer firm supply where grid development cannot meet a customer’s schedule. The projects also test whether utilities can serve that demand through nonregulated affiliates and microgrids without shifting costs to existing customers. Both will be judged on timely delivery and effective customer protections. For AEP’s fuel cell facility, the performance of planned carbon capture will also determine whether it delivers the promised emissions reductions. Google and its suppliers will need to weigh the value of faster access to power against those costs and risks.