Data centers have become the power sector’s most important new customer — and, increasingly, its newest asset class. The electricity supply built to serve hyperscale computing load is taking on the financial characteristics of infrastructure: long contracts, creditworthy counterparties, and project-financed capital structures. For energy investors, understanding how data-center power is contracted, built, and financed is now core curriculum.
Why this load is different
Power demand from data centers differs from traditional load growth in ways that matter for investors. First, scale: a single hyperscale campus can draw hundreds of megawatts to more than a gigawatt — the equivalent of a mid-sized city, arriving as one customer with one procurement team. Second, speed: data center developers want power on timelines measured in years, not the decade-long cycles of traditional utility planning. Third, credit quality: the buyers are among the largest and most creditworthy corporations in the world, which transforms the risk profile of any contract they sign.
The result is a structural mismatch with the traditional utility model. Utilities plan generation and transmission years ahead through integrated resource plans; data centers want firm capacity commitments now. That gap — between the speed of digital infrastructure and the speed of grid infrastructure — is where the new asset class is being built.
The contract structures
Data-center power deals take several forms, each with different implications for who bears what risk.
Utility tariff service is the traditional path: the data center takes service as a large customer under a special tariff, and the utility builds the supply. The utility captures the rate base growth, and investors in the utility capture the earnings. The risk is timing — the utility must get generation and transmission approved and built before the customer needs it.
Power purchase agreements (PPAs) between the data center operator and a generator or developer are the growth structure. The hyperscaler signs a long-term contract — often 10–20 years — for power from a specific project or portfolio, frequently with clean-energy attributes attached. These PPAs are what make project finance possible: a 15-year contract with an investment-grade technology company is bankable collateral, and lenders treat it as such.
Behind-the-meter generation — power plants built on the data center campus itself, bypassing the grid — is the structure for sites where grid interconnection is too slow or too constrained. Gas turbines, and in some concepts small modular reactors or fuel cells, provide dedicated supply. The economics trade the cost of duplicating infrastructure against the value of speed and certainty.
Capacity and tolling arrangements split the difference: the data center secures the right to capacity from a plant, paying for availability whether or not it runs, while energy costs pass through. These structures are familiar from traditional power markets, now applied to a new customer.
Why it qualifies as an asset class
An asset class needs definable cash flows, a deep buyer base, and repeatable structures. Data-center power is acquiring all three. The cash flows are the long-term contracts — PPAs, tariffs, capacity payments — with counterparties whose credit is among the strongest in the corporate world. The buyer base is the infrastructure capital universe: pension funds, sovereign wealth funds, and infrastructure managers already comfortable with contracted energy assets see data-center power as a natural extension. The structures are adaptations of project finance — SPVs, non-recourse debt, contracted revenue — applied to a new load.
For illustration, consider the economics from the financier’s perspective. A 500-megawatt data center campus needs roughly 4.4 terawatt-hours a year at high utilization. Supply contracted at, say, $70 per megawatt-hour represents over $300 million in annual revenue — a revenue stream comparable to a mid-sized utility’s entire retail book, concentrated in a single contract with a single investment-grade counterparty. That concentration is both the attraction and the risk.
What investors underwrite
Financing data-center power means underwriting a different risk set than traditional generation.
Counterparty concentration. A project serving one customer has binary credit exposure. If the hyperscaler thrives, the contract is gold; the underwriting question is what happens if it does not — can the power be resold, and at what price? Lenders look hard at the remarketability of the supply: grid-connected power with multiple potential buyers is more financeable than a bespoke behind-the-meter plant with one possible user.
Contract tenor versus asset life. Data center operators prefer flexibility and may resist 20-year commitments; financiers want tenor to match debt. The negotiated middle — 10–15 year contracts with extension options — is where most deals land, and the residual value assumption after contract expiry is a key underwriting variable.
Interconnection and transmission. The grid connection is often the critical path. A project with a signed PPA but no firm interconnection is not financeable, and interconnection timelines in constrained regions stretch for years. Investors have learned to treat the interconnection agreement as a gating document, on par with the PPA itself.
Technology and fuel risk. Gas-fired supply carries fuel-price and emissions exposure; renewables carry intermittency that must be firmed with storage or backup; nuclear carries construction and timeline risk. The “right” answer varies by region, and the financing structure has to match the technology’s risk profile.
The AI demand question. Underneath every data-center power investment sits a demand forecast for computing itself. Financiers do not underwrite AI adoption curves directly, but they do ask whether contracted load is backed by built and operating data centers or by speculative development — the difference between contracted revenue and contracted hope.
Who is financing it
The capital is coming from every direction. Hyperscalers themselves fund supply directly, signing PPAs that effectively pre-finance projects and occasionally investing equity in generation. Utilities build rate-based supply for data center load, capturing regulated returns. Independent power producers develop contracted projects specifically for large-load customers. Infrastructure funds buy operating portfolios and development platforms, applying the same playbook they use for renewable assets. And project finance lenders — the same banks that finance wind and solar — are extending the model to data-center supply, because the contracted cash flows look familiar.
The risks that are specific to this asset class
Three risks deserve special attention. First, stranded-asset risk from demand shortfall: if computing demand grows more slowly than the contracted supply, someone holds excess capacity. The contract protects the generator, but the system-level risk lands on whoever committed the capital. Second, regulatory backlash: large loads raise questions about who pays for grid upgrades and whether residential customers subsidize data centers — the cost-allocation fights are already underway in several jurisdictions, and adverse outcomes can reshape project economics. Third, technology displacement: the computing stack changes fast, and power infrastructure built for today’s architecture must remain useful for tomorrow’s.
The utility load-growth playbook
For regulated utilities, data-center load is the best growth story in a generation — and the playbook for capturing it is well defined. The utility files a large-load tariff: a special rate class for customers above a threshold (often 50–100 megawatts) with terms that protect existing customers. The key provisions are minimum bill and take-or-pay commitments, which ensure the data center pays for the capacity built to serve it even if its own plans change, and upfront contribution toward interconnection and system upgrades.
The regulatory politics are delicate. Commissions must be convinced that existing residential and commercial customers are held harmless — that they are not subsidizing the infrastructure built for a hyperscaler. Utilities that navigate this well get approval for tariffs that underwrite billions in new rate base; those that navigate it poorly face proceedings that delay load interconnection and sour the regulatory relationship. The tariff proceeding, not the press release, is where the value is created or destroyed.
For utility investors, the metric to watch is contracted large-load megawatts in the pipeline versus the capex plan. Load that is contracted with creditworthy customers and approved tariffs converts directly into rate base growth; load that is merely “in discussions” does not. The gap between announced interest and signed commitments is where hype lives.
Behind-the-meter economics in detail
Behind-the-meter generation — building power plants on the data center campus — deserves a closer look, because its economics invert the usual logic. A grid-connected project benefits from scale, shared infrastructure, and diversified customers. A behind-the-meter plant duplicates all of that for a single user, which is inherently more expensive per megawatt — so why do it?
Three reasons. Speed: in constrained regions, a grid interconnection can take longer than building the data center itself; on-site generation collapses the timeline. Certainty: the operator controls its own power supply rather than depending on utility planning cycles and commission decisions. Resilience: on-site generation with backup provides uptime independent of grid events, which has direct commercial value for customers selling compute reliability.
The financing follows the logic: behind-the-meter projects are typically funded by the hyperscaler or a dedicated developer under a long-term energy services agreement, with the data center’s credit underwriting the investment. For illustration, the premium over grid power can be substantial — 20–50% or more per megawatt-hour — but against the revenue of a fully utilized AI training cluster, the power premium is a rounding error. Speed has a price, and the buyers can pay it.
The long-term question is whether behind-the-meter remains a niche for constrained sites or becomes a structural parallel grid. If interconnection timelines stay long and load growth stays fast, the parallel grid grows — with implications for utility load forecasts, grid planning, and the allocation of system costs that regulators are only beginning to confront.
What could break the thesis
Every asset class has a bear case, and intellectual honesty requires stating this one’s plainly. Demand disappointment is the first: if computing demand grows more slowly than the contracted supply — because of efficiency gains, architectural shifts, or simply overbuilding — the industry faces excess capacity. Contracts protect individual projects, but the system-level overbuild would strand capital and reprice every development pipeline.
Regulatory rebalancing is the second. The political economy of large loads is unstable: as data centers’ share of system costs becomes visible, pressure builds to reallocate those costs more heavily onto the large users. Adverse cost-allocation rulings, moratoriums on new large-load interconnections, or environmental restrictions on backup generation could each reshape project economics after capital is committed.
Technology displacement is the third and hardest to underwrite. The history of computing is a history of doing more with less — and a step-change in compute efficiency would reduce the power intensity of the load these assets serve. The contracts are long; the technology cycles are short. That mismatch is the fundamental tension in the asset class.
None of these invalidates the thesis — contracted cash flows with investment-grade counterparties remain financeable through cycles. But they define the underwriting: shorter tenors, stronger remarketability analysis, conservative residual value assumptions, and constant attention to the demand signal underneath the contracts.
The bottom line
Data-center power is becoming an asset class because it has the three things infrastructure investors need: contracted cash flows, creditworthy counterparties, and financeable structures. It is not without novel risks — concentration, timing mismatches, and demand uncertainty are real — but the direction is clear. The electrons behind the cloud are being packaged, priced, and sold like infrastructure. Investors who learn the structures now will be the ones holding the assets later.