What is a power purchase agreement?

A power purchase agreement is a contract under which a generator sells electricity to a buyer over an agreed term at an agreed price. It is the standard instrument for contracting power outside a utility's standard tariff, and it comes in three broad forms.

Physical. The buyer takes delivery of electricity at a defined point. Where the generator and the load are not directly connected, the power is typically sleeved through a utility or retailer that handles balancing and delivery for a fee.

Virtual, or financial. No electricity changes hands. The parties settle the difference between the contract price and the market price at a reference point — a contract for differences in substance. The buyer continues to purchase its actual power from its supplier, and the PPA operates as a hedge on the price of that purchase.

On-site. The generator is at the facility, on the customer's side of the meter, and sells its output directly to the load — the contractual layer underneath most third-party-owned [behind-the-meter](behind-the-meter) generation, and where a PPA and a lease start to resemble each other.

In a data-center context the instrument runs in both directions, and it is worth being explicit about which is under discussion. The facility as buyer uses a PPA to fix the cost of power over a long horizon. The facility as counterparty is often what makes a new generation project financeable at all: a long-dated contract with a creditworthy, continuously consuming buyer is precisely the cash flow a generation lender wants.

"They have a PPA" is close to uninformative. Six parameters decide what the contract actually transfers, and each is negotiated separately.

Two consequences follow from the table, and between them they account for most of the disappointment around these contracts.

A data center runs continuously, and most contracted generation does not. An as-generated contract delivers when the resource is available. A facility drawing a flat load around the clock must buy the difference at whatever the market charges at that moment, which is systematically the wrong moment. The cost of covering that gap — firming — is real, it is borne by the buyer, and it is frequently omitted from the comparison that made the contract look attractive. A PPA that matches the load's shape is a fundamentally different instrument from one that matches the resource's, and the two are quoted in the same units.

In the other direction, the buyer's credit is the product. For a generation project, a long-dated contract with a creditworthy buyer is what converts a capital-intensive asset into something lenders will advance against. This is why large loads are courted: the data center is not merely purchasing electricity, it is supplying the credit that lets the plant be built. Anyone signing one should understand that the tenor being asked for is set by the generator's financing, not by the facility's needs — and should price the difference rather than absorb it.

The secondary terms carry more weight here than they appear to. Change-in-law and regulatory pass-throughs decide who absorbs a cost imposed on the industry mid-term. Curtailment provisions decide who pays when power is available and cannot be delivered. Collateral thresholds decide what happens to the contract when either party is downgraded — and a contract that demands collateral precisely when a counterparty is under strain is a contingent liability, not only a hedge.

The most consequential misreading in this area is simple and it recurs constantly: a power purchase agreement is not an interconnection.

A PPA is a commercial contract about volume and price. It does not create the physical ability to deliver electricity to a site, it does not confer standing in an [interconnection queue](interconnection-queue), and it does not shorten a study process. A project can hold contracted power for a decade and be unable to take delivery of a megawatt, because contracting for electricity and being connected to the network that carries it are separate problems solved by separate parties on separate schedules. A facility that treats a signed PPA as evidence that its power is solved will discover the distinction at the point it is least fixable.

What a PPA does, precisely: it converts an open exposure to market prices into a known cost, for a defined volume, in a defined shape, for a defined term, subject to the counterparty performing. That is genuinely valuable — power is a large and volatile share of a data center's operating cost, and fixing it is what allows the layers above to be underwritten. But it is a price instrument, and it should be tested as one.

The last question is tenor, and it is the one that ties this page to the rest of the site. A PPA's term has to be read against the [energization](energization) date at one end and against the obligations it supports at the other. Power contracted from a date the facility cannot yet use is a cost with no revenue behind it. Power contracted for less time than the leases and compute contracts it supports is an exposure sitting under a structure that assumes it is not there. Both are ordinary, both are avoidable, and both are matters of matching contract terms across the seams described in [the data-center capital stack](the-data-center-capital-stack).