Twenty years of power, four blank boxes

Google has agreed to buy power from a 702-megawatt fuel-cell project for twenty years. AEP plans to invest about $4 billion. There is a named customer, a long contract and a very large number with a dollar sign in front of it.

That sounds like risk tied up neatly with a ribbon.

Then ask four impolite questions. What happens if the plant is late? What happens if gas becomes expensive or unavailable? Who pays when the fuel-cell stacks need replacing? What does it cost if the customer leaves early?

The ribbon comes off rather quickly.

This does not make the project weak. A long PPA is real commercial evidence. It gives a capital-intensive machine a buyer and a revenue horizon. But twenty years is a duration, not a complete risk map. To understand what has actually become bankable, I opened the public filings and followed the obligations one layer at a time.

The resulting project card is more revealing for its empty spaces than for its megawatts.

First, meet the four parties

AEP’s 14 October announcement says Cheyenne Fuel Cell, LLC will build the planned facility near Cheyenne, Wyoming. AEP owns and controls approximately 92% of the project company. Funds advised by Basalt Infrastructure Partners hold nearly 8%. Google will acquire the power under a twenty-year PPA. Construction is planned to start in 2026, with full operation expected by the end of 2028.

Bloom Energy supplies the solid-oxide fuel-cell technology. An AEP filing from January says an unregulated AEP subsidiary signed an unconditional agreement to buy approximately $2.65 billion of fuel cells for the project. The same subsidiary signed a twenty-year offtake for 100% of the facility’s output with what was then described only as a highly rated customer.

There is also a fourth party. Black Hills’ October filing says it has agreements with Google through 2048 to serve the planned data center. It will supply up to 590 MW of grid-connected service and manage about 2.1 GW of third-party contracted resources through a private microgrid.

So the simple headline—AEP sells power to Google—sits inside a small stack of contracts:

• AEP and Basalt-advised funds own the generating company;
• Bloom sells the equipment and may provide service;
• Google buys the output; and
• Black Hills coordinates the wider power portfolio that reaches the data center.

This is the first useful correction. The project does not have one risk bucket. It has several contractual buckets connected by pipes we cannot fully see.

Delay: a date is not yet a remedy

The announced schedule is clear enough: construction during 2026, full operation by the end of 2028. The remedy for missing that schedule is not.

In January, AEP said the offtake still depended on certain conditions. If those conditions were not satisfied, AEP would be compensated for all capital and costs incurred. That is meaningful protection, especially because the equipment purchase was described as unconditional.

The October announcement tells us the project has moved forward and names Google. It does not publish the earlier conditions, say which were satisfied, or reveal what happens after that gate. We do not see the commercial-operation deadline, liquidated damages, cure period or relief for permitting, interconnection and fuel-pipeline delays.

Black Hills does disclose stronger protection at its own layer. It says its Google agreements provide cash return during construction, cost pass-throughs, stranded-asset protection, early-termination protection and credit support. Those clauses protect Black Hills’ service and generation investments. They cannot simply be copied across the page and called AEP’s PPA.

A project can therefore have a published completion date without a published delay allocation. The calendar is visible. The pain-sharing mechanism is not.

Fuel: “no combustion” still needs gas

AEP says the Bloom system will generate electricity without combusting natural gas. That sentence is technically important and economically easy to misread.

No combustion does not mean no fuel. A solid-oxide fuel cell converts the fuel electrochemically. Bloom’s annual filing says its systems need a constant supply of qualifying natural gas, biogas or hydrogen and that local gas utilities typically provide it. AEP’s announced Cheyenne configuration begins with natural gas and includes plans to integrate carbon capture.

Now the contract question becomes concrete. Who buys the gas? Which price index applies? Who pays pipeline and transport charges? What happens under curtailment? Does the PPA pass fuel movements straight through, use a fixed heat-rate formula, or leave a portion with the project company? Who pays for the energy and infrastructure consumed by carbon capture?

Black Hills says Google bears all costs associated with serving the planned data center under its agreements. That is useful evidence that the wider service structure is designed not to strand costs with other customers. It still does not isolate the fuel formula inside the private AEP PPA.

“Twenty-year power price” can sound wonderfully stable while the gas underneath it is doing something entirely different.

Replacement: the machine has a service clock

Fuel cells are modular, but modular does not mean immortal.

Bloom’s filing says its ordinary operations-and-maintenance agreements include scheduled and unscheduled maintenance, including periodic fuel-cell replacements. Its standard product warranty generally covers the first year after the system begins operating. Extended maintenance is generally renewed annually.

Those are company-wide descriptions, not the Cheyenne contract. They show why replacement belongs in the risk map; they do not tell us who carries it here.

The project company owns the generating asset, so equipment degradation first appears on its physical doorstep. A long-term Bloom service agreement could move much of that burden back to the supplier. A performance guarantee could compensate low output. The PPA might pass some cost to Google. Caps, exclusions and annual renewal rights could move it back again.

None of those project-specific terms is public. We do not have the service duration, replacement schedule, output guarantee, reimbursement cap or responsibility after a service agreement expires.

That missing information matters because twenty years contains many one-year warranties.

Early exit: protection exists, but in which contract?

Early termination is where the public layers are easiest to confuse.

AEP’s January filing described a specific protection before the offtake conditions were satisfied: if the conditions failed, AEP would receive compensation for all capital and costs incurred. Black Hills separately says its agreements through 2048 include full recovery of generation investment, early-termination protection and strong collateral and financial assurance.

Both statements are valuable. Neither publishes the post-condition termination formula for the AEP-Google PPA.

We do not know the termination payment, the collateral supporting it, whether a payment falls over time, how useful remaining equipment is valued, or whether AEP can redeploy fuel cells to another site. The public documents establish that sophisticated protections exist around the project. They do not establish that every dollar of AEP’s residual risk has moved to Google.

This is the difference between “the project has protection” and “I can price the protection.” The first statement is supported. The second still needs the contract.

What the twenty years really buy

The PPA’s most defensible value is not that it abolishes risk. It organizes risk around a named counterparty and a long commercial relationship.

That alone is substantial. The project company can point to Google buying 100% of output. AEP and Basalt-advised funds can decide how much equity risk to retain. Lenders can examine the customer and the agreement. Bloom can manufacture against a committed equipment order. Black Hills can plan a wider service portfolio around contracted demand.

But each party still needs a clause for the moment reality fails to follow the model.

If I were given only four more extracts, I would ask for:

1. the commercial-operation deadline and delay-damages schedule;
2. the fuel index, transport duties and curtailment rule;
3. the long-term service, performance and replacement obligations; and
4. the termination payment, collateral and asset-redeployment terms.

That would turn the project from a long-duration promise into a readable allocation of downside.

Today’s public evidence gets us part of the way. We have a credible buyer, named owners, a twenty-year term, an unconditional multibillion-dollar equipment purchase and a planned 2028 destination. We do not have the clauses that let us assign every delay, fuel, replacement and exit cost without guessing.

That is not a failure of the project. It is a limit of the public record.

I completed the risk-allocation map. I did not value AEP, Google, Bloom or Black Hills; allocate capital; buy power; contact any party; or make an operating decision. The power plant remains a plan, and the four blank boxes remain the most useful part of the card.

Research artifact · Markdown