Power Purchase Agreements for Data Centers: How Developers Structure the Deal That Now Leads the Deal

Power purchase agreements have moved from a late-stage operational task to a primary site selection and capital structure input for data center developers -- and the structuring decisions made at PPA stage now determine whether a project can be financed at all.

Until recently, a data center developer's sequencing looked like this: find land, complete due diligence, secure entitlements, sign the lease, then figure out power. That sequence is no longer viable for most institutional-scale projects. Power access -- and specifically the structure of the contract that secures it -- has moved to the front of the development workflow.

In 2026, the PPA is often the deciding document in whether a project proceeds.

Meta's January 2026 announcement of agreements to procure up to 6.6 GW of firm nuclear power across Vistra, TerraPower, and Oklo illustrated the hyperscaler version of this logic: secure dispatchable, carbon-free baseload power first, then build the data centers around it. For institutional developers building wholesale and colocation capacity, the same principle applies at smaller scale. Projects that cannot demonstrate power delivery certainty are struggling to close construction financing.

Why PPA Structure Matters More Than Price

A common mistake in early PPA conversations is optimizing for power price first. The more important variable is the risk allocation the PPA creates -- and whether that allocation is compatible with development financing.

There are three common PPA structures for data center developers. Each carries different risk profiles.

Physical PPAs deliver power directly from a generating facility to the data center. The developer takes physical delivery under a fixed or formula-based price agreement. The advantages are price certainty and direct supply chain control. The risks are geographic: physical PPAs require the data center and the generating facility to be on the same grid region. For projects in congested markets like Northern Virginia or Phoenix, locating a physical PPA counterparty with available capacity nearby is increasingly difficult.

Virtual PPAs (VPPAs) allow developers to purchase power from a generating facility while procuring operational electricity from the local utility. The VPPA creates a financial hedge -- the developer pays a fixed price to the generator and receives a floating market price for the power sold into the grid. If market prices exceed the fixed price, the developer receives a payment; if they fall below it, the developer pays the difference. VPPAs avoid the geographic constraints of physical delivery but introduce basis risk and require balance sheet capacity to absorb mark-to-market exposure. Construction lenders and equity investors have different tolerances for VPPA basis risk, so the financing implications need to be modeled early.

Utility service agreements -- negotiated directly with the serving utility -- remain the standard for most projects, but the content of those agreements has become materially more complex. Large-load service agreements now typically include demand response obligations, potential curtailability provisions, network upgrade cost assignments, and rate schedule terms that vary significantly by utility and by regulatory jurisdiction. The FERC RM26-4 proceeding will standardize some of these terms for projects above 20 MW connecting to interstate transmission, but utility-specific terms remain negotiated independently.

Nuclear as the Emerging Baseline for Hyperscalers

The most significant structural shift in 2026 PPA markets is the emergence of nuclear as the preferred source for large-scale data center power.

Nuclear delivers what data center operations actually need: dispatchable, carbon-free baseload power with capacity factors above 90%. For hyperscalers making 24/7 clean energy commitments, nuclear closes the gap that solar and wind cannot. Meta's deal with Vistra for existing nuclear output at the Clinton Clean Energy Center in Illinois -- a 20-year agreement for more than 1.1 GW -- followed Microsoft's restart of Three Mile Island Unit 1 under a comparable structure.

For institutional developers building capacity that hyperscalers will ultimately lease, nuclear-adjacent sites are now explicit acquisition criteria. Within a 20-mile radius of operating or restarting nuclear plants, data center land is pricing at a meaningful premium due to interconnection advantages and reduced transmission loss. Sites on the same transmission path as nuclear output can avoid years of interconnection queue delay in congested markets.

This has created a two-tier land market. Grid-served sites compete on cost. Nuclear-adjacent sites with clear interconnection paths compete on certainty, and certainty commands a lease rate premium of 15-25% based on current market transactions.

For developers, this means nuclear proximity is now a site screening criterion, not a secondary consideration. AI can layer irradiance, transmission topology, interconnection queue position, and nuclear plant operating license data to identify sites where nuclear-adjacent power access is actionable -- but the PPA negotiation itself requires direct utility and generator relationships.

Key PPA Terms to Negotiate for Development Financing

Lenders evaluating construction loans for data center projects increasingly require demonstrated power access before committing. The following PPA terms have the most direct impact on whether a deal is financeable.

Tenor and commencement date. PPAs for data center financing typically run 10-20 years. Lenders want to see a commencement date aligned with the project's expected energization and a term that extends beyond the loan maturity. If the PPA commencement is speculative or contingent on utility milestones that are not under the developer's control, lenders will treat the power commitment as unconfirmed.

Delivery conditions and curtailability. If the PPA includes curtailability provisions -- allowing the generator or utility to reduce supply under system stress conditions -- lenders and tenants will require analysis of how often curtailment could occur and what the operational fallback is. For hyperscale anchor tenants with guaranteed uptime SLAs, curtailability may require backup generation coverage that needs to be sized and permitted at the development stage.

Termination triggers and force majeure. Data center construction timelines have extended significantly due to transformer lead times and interconnection delays. A PPA with narrow force majeure language or tight milestones that the developer cannot reliably hit creates termination risk that will show up as a financing haircut. Negotiate for construction delay flexibility at the PPA stage, not after the lender has flagged it.

Cost assignment for network upgrades. Under the evolving RM26-4 framework, network upgrade costs may shift to the load customer. Understand whether the utility service agreement or PPA assigns any upgrade cost responsibility and model those costs explicitly in the pro forma before committing to site control.

AI's Role in PPA Workflow

AI has a practical role in the PPA diligence and monitoring layer.

Before LOI, AI can aggregate utility integrated resource plans, FERC interconnection queue filings, transmission constraint reports, and historical curtailment data to build a power risk profile for each candidate site. That compressed research layer allows development teams to screen sites on power quality and contract risk, not just headline availability.

After PPA execution, AI can monitor utility regulatory dockets, generator operating reports, and interconnection milestone filings to flag events that could affect delivery obligations -- curtailment frequency increases, utility rate case filings that affect the underlying tariff, interconnection queue positions that could affect the generator's own supply.

The negotiation itself -- PPA structure, counterparty credit, risk allocation, term length -- remains human work. But the analytical support that informs negotiating position and monitors ongoing performance is where AI compresses the workflow.

Power purchase agreement structuring has become a development competency, not a utility task. The developers who treat it that way -- starting PPA conversations at site screening rather than after entitlements -- are the ones closing deals in the current market.