Data centers are among the largest electricity buyers in the world, and how they procure power shapes energy markets far beyond their own fences. Over the past decade, large technology companies have moved from simply paying the utility bill to structuring sophisticated power deals — power purchase agreements, hourly carbon matching, and behind-the-meter generation. Understanding these deal structures is essential to understanding how data center demand interacts with the grid.
The starting point: buying from the utility
The default way to buy electricity is the utility tariff: the data center takes service under a commercial or industrial rate schedule, paying demand charges for its peak draw and energy charges for its consumption. For many smaller data centers, this is still how it works.
But large data centers quickly outgrow standard tariffs. A 200-megawatt customer is not a typical commercial account — it may be among the utility’s largest customers, with bargaining power and needs that standard rate schedules weren’t designed for. Utilities have responded by creating large-load tariffs and special contracts: negotiated rates with terms matched to the customer’s size, load profile, and creditworthiness. These deals are where much of the action is now, and they are covered in detail in our piece on utility planning for large loads.
The other limitation of the standard tariff is that it offers no control over the power’s source. A data center buying tariff power gets the grid mix — whatever the utility’s portfolio happens to be. For companies with clean-energy commitments, that is not enough. Hence the PPA.
The power purchase agreement (PPA)
A power purchase agreement is a long-term contract — typically 10 to 20 years — to buy electricity from a specific generator, most often a new wind or solar farm. The data center (or more often, its corporate parent) agrees to buy the project’s output at a fixed price; the project uses that contracted revenue to secure financing and get built. The buyer gets price certainty and a credible claim to clean energy; the developer gets a bankable revenue stream.
PPAs come in two main flavors. A physical PPA involves actual delivery of power — feasible when buyer and generator are in the same grid region. A virtual or financial PPA is a contract for differences: the generator sells its power into its local market, the buyer continues buying tariff power in its own market, and the two settle the price difference financially. Virtual PPAs let a company support a wind farm in Texas while its data center runs in Virginia; the electrons don’t match, but the financial support for new clean generation is real.
The corporate PPA market has grown enormously, with technology companies as the dominant buyers. Data center operators have been the anchor customers for a significant share of new U.S. wind and solar capacity over the past decade. This is genuinely additional clean energy — projects that, in many cases, would not have been financed without the contracted demand.
But the standard PPA has a well-understood limitation: intermittency. A solar PPA delivers power when the sun shines; the data center draws power around the clock. On an annual basis, the megawatt-hours can match — the company buys as much clean energy as it consumes — but hour by hour, the data center still relies on the grid’s dispatchable resources, often fossil-fueled, to fill the gaps. Critics call this “annual matching” and argue it overstates the climate benefit. The industry’s response is 24/7 carbon-free energy.
24/7 carbon-free energy
The 24/7 carbon-free concept — championed by several large technology companies — raises the bar: match every hour of consumption with carbon-free generation, in the same grid region, every hour of the year. Not net-zero over the year, but zero-carbon in each of the 8,760 hours.
This is much harder than annual matching, and much more expensive with today’s technology. Solar and wind can’t cover nighttime or calm periods alone. Hourly matching requires a portfolio: wind and solar for the bulk of hours, plus storage to shift energy across hours, plus firm carbon-free resources — geothermal, nuclear, or long-duration storage — for the hardest stretches. Early 24/7 pilots have shown it is technically feasible but carries a meaningful cost premium over annual matching, a premium that falls as storage gets cheaper and firm clean resources scale.
The significance of 24/7 goes beyond corporate sustainability claims. It creates demand for exactly the resources the grid most needs: firm, dispatchable carbon-free power and long-duration storage. A buyer willing to pay a premium for clean power at 2 a.m. on a windless night is sending a price signal for the technologies that solve the grid’s hardest decarbonization problem. Whether that signal is strong enough, at sufficient scale, to move markets remains an open question — but the direction is notable.
Behind-the-meter generation
A more radical option is to generate power on the data center’s own side of the utility meter — “behind the meter.” Instead of drawing all power from the grid, the facility installs its own generation: historically, diesel or natural gas backup generators, but increasingly solar arrays, battery storage, fuel cells, and even small turbines.
Behind-the-meter generation serves several purposes. The traditional one is backup: data centers have long maintained on-site generators to ride through grid outages, since even brief interruptions are unacceptable. But the new wave is about primary power — generating some or all of the facility’s electricity on-site, continuously, to supplement or bypass the grid connection.
The motivations are practical. In regions where grid interconnection takes years, on-site generation can get a facility powered faster. Where utility rates are high or grid power is carbon-intensive, self-generation can be cheaper or cleaner. And for the largest campuses, on-site generation provides a hedge against grid constraints and price volatility.
The regulatory treatment of behind-the-meter generation is complex and varies by jurisdiction. Utilities may impose standby charges — fees for the grid capacity held in reserve for when the on-site generation isn’t running. Interconnection rules for exporting excess power differ widely. And there are environmental questions: a data center running on-site natural gas turbines around the clock is, functionally, a power plant, raising air-quality and emissions issues that backup-only generators don’t.
The emerging deal landscape
In practice, large data centers now assemble power portfolios rather than signing single deals. A hypothetical 300-megawatt campus might combine: a utility tariff or special contract for the bulk of its grid power; one or more PPAs for wind and solar to cover clean-energy commitments; a battery system for backup and grid services; and on-site gas turbines or fuel cells for fast-start capacity while grid interconnection is completed.
New structures continue to emerge. “Bring your own power” arrangements — where the customer funds dedicated generation, from turbines to small modular reactors — blur the line between customer and power plant developer. Tariff innovations like clean-energy riders let customers buy renewable power through the utility rather than contracting directly. And in restructured markets, large customers can participate directly in wholesale markets, buying energy and capacity like a utility would.
The additionality debate
Not all clean-energy claims are equal, and the data center industry’s procurement boom has intensified scrutiny of what counts. At the center is additionality: did the buyer’s action cause new clean generation to be built that wouldn’t have existed otherwise?
A long-term PPA with a new-build wind farm is the gold standard — the contracted revenue directly enables financing, and the project is demonstrably additional. At the other end of the spectrum are unbundled renewable energy certificates (RECs): cheap credits, often from existing projects, that let a buyer claim “100% renewable” without funding anything new. Between them lies a spectrum of instruments with varying credibility.
The debate matters because corporate clean-energy claims now move billions of dollars and shape public perception of the industry’s climate impact. Critics argue that annual matching with distant virtual PPAs — while financially supporting new projects — does little to decarbonize the actual grid where the data center operates, since the facility still draws fossil power hour by hour. Defenders counter that the projects wouldn’t exist without the buyers, and that transforming the generation fleet anywhere reduces emissions everywhere.
24/7 carbon-free procurement is partly a response to this critique: by requiring hourly, local matching, it forces the buyer to confront the actual grid where it operates. But it raises the cost substantially, which is why adoption remains concentrated among the largest, most climate-committed buyers. The broader market still runs on annual matching — credible when tied to new projects, less so when built on cheap unbundled credits.
Retail choice and wholesale participation
In restructured electricity markets — Texas, PJM, and others — large customers aren’t limited to utility tariffs. Retail choice lets them buy from competitive suppliers, negotiating custom products. The most sophisticated buyers go further, participating directly in wholesale markets: scheduling their own energy, buying capacity in capacity auctions, and hiring scheduling coordinators to manage the complexity.
Direct wholesale participation gives the data center maximum control over its power costs and supply — and maximum exposure to market risk. It also raises a policy question: when a large load bypasses the utility entirely, the utility’s fixed costs spread across fewer remaining customers. Restructured states handle this through non-bypassable charges — fees that even direct-market participants pay for shared grid infrastructure — but the boundary is constantly negotiated.
Colocation arrangements add another layer. When a data center developer builds a facility and leases space to tenants, who is the electricity customer? Structures vary: the developer may be the utility customer and pass power costs through to tenants, or tenants may contract directly. As AI colocation grows — companies renting space in someone else’s AI-ready facility — these arrangements determine who has the incentive and the ability to procure clean power, and the answer isn’t always the party with the sustainability commitment.
Why it matters
These deal structures matter for three reasons. First, they determine how much new clean generation actually gets built — the additionality question. A PPA that finances a new wind farm has a different grid impact than a tariff rider that reshuffles existing renewable credits. Second, they affect cost allocation: when large customers procure their own power, the fixed costs of the grid must still be paid by someone, and rate design has to keep up. Third, they shape the generation mix: the resources data centers choose to buy — or build — today become the grid’s infrastructure for decades.
The data center industry’s procurement evolution, from passive tariff customers to sophisticated power buyers, is one of the most consequential demand-side developments in the history of the power sector. The deals are getting more complex because the stakes are getting higher — and the grid is being reshaped by the fine print.