On October 13, Mainspring Energy announced that American Electric Power had agreed to buy up to 1.6 gigawatts of its linear generators through 2034, with a first tranche of 190 megawatts for delivery beginning in 2027. The next day, AEP announced that it plans to invest approximately $4 billion in a 702-megawatt fuel cell project near Cheyenne, Wyoming, using Bloom Energy technology, under a 20-year power purchase agreement with Google. Two announcements, two technologies, two headline figures of roughly $4 billion each — and one of the country’s largest regulated utilities repositioning itself as a buyer, owner, and developer of on-site generation at gigawatt scale.
Read it as a structure story first and a technology story second, as this series read Generac’s Amazon deal and Helix. The interesting part is not that a utility bought generators. It is that AEP is underwriting two different kinds of on-site generation, through two different structures, with two very different risk profiles — one contracted to the world’s strongest corporate credit, the other with no disclosed customer, location, or regulatory treatment — and asking investors to treat both as the same strategy. They are not the same bet.
Start with the arithmetic. The combined ceiling across both deals is about 2.3 gigawatts. Against AEP’s approximately 33,000 megawatts of owned and contracted capacity, that ceiling is roughly 7 percent of the fleet — if every tranche is exercised. The firm portion is much smaller: 190 megawatts of Mainspring units actually purchased, plus 702 megawatts of Bloom fuel cells under a signed PPA, or 892 megawatts combined — about 2.7 percent of the fleet. The gap between 2.3 gigawatts and 892 megawatts is the gap between an option and an obligation.
The two $4 billion figures are not equally solid
The week produced two $4 billion headlines with very different evidentiary status. The Bloom figure comes from AEP itself: its October 14 release says the company “plans to invest approximately $4 billion” in the Cheyenne project. That is the buyer’s own estimate for a defined project.
The Mainspring figure comes from somewhere else entirely. Neither company’s announcement disclosed a price. The “up to $4 billion” characterization surfaced in a Bloomberg Tech interview with Mainspring chief executive Tom Linebarger, as reported by Utility Dive, which described the agreement as one “which could be worth up to $4 billion.” That is a ceiling suggested by the seller’s CEO to a reporter — not a contract value AEP has confirmed. Treating the two $4 billion figures as equivalent would be a category error: one is the buyer’s own estimate, the other is the seller’s upside characterization.
Dividing the Cheyenne $4 billion by 702 megawatts yields roughly $5,700 per kilowatt — arithmetic on disclosed figures, not a company-stated unit cost, and AEP has provided no cost breakdown. The same arithmetic on the reported Mainspring ceiling would yield roughly $2,500 per kilowatt, but that ceiling is not a disclosed price.
What these machines are — and aren’t
A linear generator and a fuel cell are both on-site, modular, gas-fueled machines that produce electricity without a spinning turbine. The resemblance mostly ends there. Mainspring’s linear generator burns fuel in a low-temperature, flameless reaction that drives parts back and forth in a straight line through copper coils, producing electricity directly from the linear motion. The units come in standard 2.5-megawatt configurations, can switch among natural gas, renewable gas, biogas, propane, and hydrogen according to the company, and produce close to zero nitrogen oxide emissions. Mainspring says units can be online within 12 months, and AEP will be able to buy about 200 megawatts a year — a pace that would fill the ceiling by 2034 if sustained.
Bloom’s fuel cells are a different machine. A solid oxide fuel cell converts fuel into electricity through an electrochemical reaction rather than combustion — no flame at all, not even a low-temperature one. Bloom says its fuel cells can run on 100 percent hydrogen or any blend with natural gas. When running on natural gas, Bloom claims 34 percent lower carbon dioxide emissions than the displaced marginal generation in PJM Interconnection, and virtually no sulfur or nitrogen oxide emissions. Neither is zero-emission on natural gas. The “near-zero” language in AEP’s Cheyenne announcement depends entirely on carbon capture that is planned, not built.
Why a utility would own these instead of building wires
The strategic logic is speed. AEP’s own framing gives it away: chairman, president, and chief executive Bill Fehrman said the Mainspring units “can provide additional capacity quickly and efficiently,” and the Cheyenne release emphasizes technology that “can be deployed rapidly.” AEP operates the nation’s largest transmission system — 40,000 line miles — and serves 5.6 million customers across 11 states. It knows better than almost anyone how long transmission takes: years of siting, permitting, and construction, with interconnection queues as the binding constraint.
On-site generation sidesteps the queue: a modular generator at or near the load needs no new transmission line, substation, or place in an interconnection process. That is why the same logic has now appeared on both sides of the meter: Generac selling backup generators to Amazon, Black Hills building gas generation and a private microgrid for Google’s Cheyenne campus, and now AEP buying linear generators and building fuel cells. The buildout is routing around the grid’s slowest step by putting generation next to the load.
There is a second logic, specific to AEP: the Cheyenne project sits outside its regulated footprint. Wyoming is not among the 11 states where AEP’s utilities operate; Fehrman said the project demonstrates opportunities “both within and beyond our traditional service territory.” Mainspring unit locations are undisclosed, but the pattern is clear: AEP is using its balance sheet to serve large-load customers wherever they are, not only where its wires run — growth migrating outside the regulated model into negotiated, contracted structures.
The Google contract changes everything about the risk
The Cheyenne project’s structure is what makes it financeable. Google will buy power from Cheyenne Fuel Cell, LLC, under a 20-year PPA. The project company is approximately 92 percent owned and controlled by AEP, with nearly 8 percent owned by funds advised by Basalt Infrastructure Partners. Construction is planned to start in 2026, with full operation expected by the end of 2028. Counterparty quality is the load-bearing wall: a 20-year PPA with Google converts a $4 billion construction project into a contracted cash-flow stream lenders can underwrite. The Basalt minority stake shares capital and risk at the project level without diluting AEP’s control. AEP has not disclosed the PPA price, the financing structure, the debt terms, or target returns, and this analysis will not guess at them.
It is also worth placing Cheyenne in its full context. The AEP project is not Google’s only Cheyenne power arrangement. It follows agreements between Google and Black Hills, effective September 30, under which Black Hills is supplying up to 590 MW of grid-connected service plus management of roughly 2.1 GW of third-party resources through a private microgrid, with terms through 2048. Google is assembling a portfolio of suppliers for a single campus, and AEP’s 702 megawatts is one piece of it. That confirms Google’s Cheyenne demand is real and multi-sourced — and makes AEP one vendor among several competing for a single customer’s favor.
The Bloom relationship also predates Cheyenne. Bloom announced in late 2024 a supply agreement with AEP for up to 1 gigawatt of fuel cells, with an initial 100-megawatt order. The 702-megawatt Cheyenne project sits under that umbrella, consuming most of the 1-gigawatt supply agreement in a single project. That concentrates the Bloom relationship: an agreement meant to seed fuel cells across many sites is now substantially committed to one campus, one customer, and one 20-year contract.
The Mainspring side has no such anchor
Everything the Cheyenne project discloses, the Mainspring agreement does not. No customer is named, no locations disclosed, no regulatory treatment described — rate base, competitive affiliate, or contracted service, all unknown — and no price disclosed. The only firm commitment is the 190-megawatt first tranche, roughly an eighth of the ceiling, for delivery beginning in 2027. The rest is an option of up to about 200 megawatts a year through 2034.
That optionality is a feature, not a flaw — for AEP. An option on 1.4 gigawatts of fast-deployable capacity, exercisable as load materializes, is exactly the flexibility a utility wants when demand forecasts are uncertain. But optionality for the buyer is uncertainty for everyone else. The Mainspring agreement’s value will be determined tranche by tranche. Until those decisions are disclosed, the 1.6-gigawatt figure is a ceiling on a possibility, not a pipeline.
One governance wrinkle: AEP is already an investor in Mainspring, alongside NextEra, the Gates Frontiers Fund, Khosla Ventures, and others, with its stake size undisclosed. A customer that is also an investor sits on both sides of the negotiation, and the terms deserve the same arm’s-length scrutiny this series applied to Vistra’s dual role inside Helix. There is no suggestion of impropriety, and AEP’s stake may be small. But the disclosure gap means the market cannot test the arm’s-length question.
The execution risks are concentrated in the suppliers
Both deals transfer AEP’s execution risk onto two suppliers that have never delivered at this scale. Mainspring, founded in 2010, began commercial shipments in 2020 and says it has more than a gigawatt in development and field operations. It recently closed a $242 million Series G round led by Valiant Peregrine, and has raised a little more than $1 billion across all rounds, according to Utility Dive. Real numbers for a private company — and small against delivering 200 megawatts a year for eight years. Manufacturing scale-up, field reliability across hundreds of units, and service in states where Mainspring has not operated at utility scale are all unproven at that pace.
Bloom’s execution risk is different in kind but not in degree. Bloom has deployed more than 1.3 gigawatts, according to the company, and says its factory can deliver gigawatts per year. But Cheyenne — 702 megawatts at a single site, on a two-year timeline, with carbon capture integration — would be among the largest fuel cell deployments ever attempted. Bloom’s parallel commitments are also large: an expanded Oracle agreement covering up to 2.8 gigawatts, a Nebius deal signed in May 2026, and Equinix deployments across 19 data centers. Manufacturing capacity, not demand, is the constraint that could bind across all of them at once.
The carbon capture question deserves its own paragraph, because it carries the emissions claim. AEP says Cheyenne “includes plans to integrate carbon capture technology to deliver reliable, near-zero-emissions power.” Plans, not equipment: no capture rate, vendor, cost, or start date disclosed. On natural gas without capture, Bloom’s fuel cells beat combustion — its claim of 34 percent lower CO2 than PJM marginal generation, if accurate, is meaningful — but they are not near-zero. The near-zero outcome depends on a carbon capture system that has been announced as an intention. Treat the emissions claim the way you’d treat the Mainspring ceiling: a possibility conditioned on future execution.
What to watch
Milestones that convert announcements into earnings, in order: the first Mainspring tranche delivery in 2027 and the disclosure of where those units go and on what terms; the second tranche decision, which will show whether the option is being exercised on schedule; Cheyenne’s construction start and financing package; carbon capture vendor selection; and full operation by the end of 2028. Watch the regulatory filings too: if Mainspring units enter rate base, the rate-case treatment will define the deal’s economics more than the equipment price.
The bottom line
AEP’s two deals belong in the same analytical frame as Generac’s Amazon warrant, Black Hills’ Google contract, and Helix: they are structure stories, not just sales stories. Generac fused the generator manufacturer and the data center into one aligned contract; Black Hills became an infrastructure partner outside the regulated model; Helix fused the data center, the chip, and the power plant into one vehicle. AEP’s move fuses the regulated utility and the on-site generator: the company that owns the wires is now buying the machines that route around them.
The two deals are not equal bets, and the market should not price them as if they were. Cheyenne is a contracted, single-customer infrastructure investment with a defined price tag, term, and counterparty. The Mainspring agreement is an option on a technology, with a disclosed first tranche and an undisclosed everything else. The first can be underwritten. The second can only be watched, tranche by tranche. What unites them is the shared judgment: put generation next to the load, and treat the interconnection queue as a bottleneck to bypass. Whether that judgment earns its keep will be decided the way these things always are — delivery by delivery, megawatt by megawatt.
None of this is investment advice, and nothing in this analysis recommends buying or selling any security. What it does suggest is a structure worth watching: a regulated utility buying the generators, a $4 billion contracted project beside a $4 billion reported ceiling. Whether AEP’s bet pays depends on two suppliers’ execution, terms not yet disclosed, and a carbon capture system that exists so far only as a plan.