How Hyperscalers Choose Data Center Sites: Power First

“Hyperscaler” usually means the handful of companies that run cloud and AI infrastructure at the largest scale: Microsoft, Google (Alphabet), Amazon (AWS) and Meta, with Oracle and a few AI developers increasingly in the same class. Their hyperscale campuses set the pattern for much of the market, because colocation developers and landowners compete to offer what these buyers want.

Their spending explains the influence. As of the first-quarter 2026 earnings reports in April, the four largest companies’ combined 2026 capital spending plans totaled roughly $700 billion, much of it for data centers and the equipment inside them.⁹ Each company runs its own site selection teams and its own process, and none publishes a full checklist. What follows is drawn from their public statements, announcements and regulatory filings, and from patterns visible across their recent projects.

This guide focuses on what the largest buyers have said and done. For the complete set of technical criteria any site must pass, see our site selection checklist.

The clearest public statement of the power-first shift came from Microsoft. In a late-2025 interview, CEO Satya Nadella said the company’s problem was not a shortage of chips but a lack of “warm shells” to plug them into, and that it might have chips sitting in inventory it could not plug in.¹

The other companies have answered the same problem with deals rather than quotes. In December 2025 Alphabet agreed to buy Intersect, a developer of energy and data center projects, for about $4.75 billion in cash plus assumed debt.² Coverage of the deal described a model of “energy parks” that co-locate hyperscale data centers with generation and battery storage, and quoted Sundar Pichai saying Intersect would help Google build new generation in lockstep with new data center load.¹⁰ The acquisition closed in March 2026, and some of Intersect’s existing operating assets in Texas and California were excluded from it.¹¹

Google has also worked on the demand side. In August 2025 it announced demand response agreements with Indiana Michigan Power and the Tennessee Valley Authority, the first time it targeted machine learning workloads for demand response, while noting that this flexibility would be available only at certain locations.¹² Our guide to flexible data center loads explains how such agreements can speed a connection.

The recent pattern is that hyperscalers do not wait for the grid to show up. They pair a site with a specific source of supply, either by contracting with an existing plant, financing new utility generation, or owning generation outright.

• Contracting with an existing plant. Talen Energy agreed to supply Amazon up to 1,920 MW from the Susquehanna nuclear plant through 2042, ramping up through 2032.³¹³ The arrangement was restructured from a co-located, behind-the-meter design to a front-of-the-meter supply through the PJM grid after FERC rejected an expanded interconnection agreement, and Talen said the new structure did not need FERC approval.¹³
• Funding new utility generation. In August 2025 the Louisiana Public Service Commission approved Entergy Louisiana’s plan to build three natural gas plants, about 2.26 GW combined, to serve Meta’s Hyperion campus in Richland Parish.⁴ The vote was 4–1; two units in Richland Parish are expected online in late 2028 and a third at Entergy’s Waterford site by the end of 2029.¹⁴
• Owning the developer. Google’s purchase of Intersect brings generation development in-house rather than relying only on power purchase agreements.²¹⁰

The lesson for sites is that a hyperscaler is often choosing a power arrangement first and a parcel second. Land near a plant with spare output, or in a utility territory willing to build generation for a large customer, can move ahead of land that is better in every other respect. Our guides to nuclear siting and behind-the-meter vs. front-of-the-meter power cover those structures.

Once a power path exists, the companies’ own announcements describe a consistent set of criteria. Meta’s are the most explicit because its economic development team states them in each site announcement.

Read together, these point to five practical tests: large and reliable power, ready infrastructure (transmission, fiber, water, roads), enough contiguous buildable land, a labor pool for construction and operations, and a local government and utility willing to move at the company’s pace. Clean energy matters too. Meta has said its data centers’ electricity use is matched with 100% clean and renewable energy, and its Louisiana package included a utility plan for up to 1,500 MW of solar.⁶¹⁴

What the announcements leave out is also telling. Proximity to users and low land cost rarely appear as headline reasons for large AI campuses. That is consistent with the shift toward remote, power-rich locations discussed in our guide to choosing between metro and power-rich sites.

Local opposition and concern about electricity bills have pushed hyperscalers to make public commitments that shape where and how they build. In January 2026 Microsoft announced a community-first approach to AI infrastructure built around power costs, water, jobs, taxes and community investment, including asking utilities to charge it for the power it uses rather than raising rates across the community.⁷ Meta said its Canadian project would fund all energy costs associated with the data center and finance new generation and grid upgrades.⁸

Not every community is persuaded. Near Meta’s El Paso site, residents raised concerns about water use, electricity and effects on their neighborhoods.¹⁵ In Louisiana, the lone dissenting commissioner said he could not verify many of the costs and benefits, and consumer groups argued that new plants could eventually raise bills for existing customers.¹⁴

For sites, this means local politics now sit beside power and land in the screen. A county with a clear large-load tariff, workable zoning and a history of approving industrial projects is easier to sell to a hyperscaler than one that has never hosted a large load. Our guides to community engagement and large-load tariffs go further.

The internal process differs by company, but the outward sequence a landowner or utility sees tends to look similar.

Most searches start under a code name, and landowners often do not know who the end user is until late in the process. Our guide to confidentiality and project code names covers what that means for negotiations.

A few features recur. Searches usually run several regions and sites in parallel, so a parcel is competing with options in other states. Land is typically tied up through an option or purchase agreement with long diligence periods before anything is public. Utility conversations often run alongside, and sometimes ahead of, land negotiations, because the power answer decides whether the land matters. State and local incentives, rezoning and the utility agreement tend to come together near the end, which is why announcements often arrive with a power deal, an incentive package and community commitments at once. A site that cannot clear the first step rarely gets a second look.

If the largest buyers are choosing power first, the most useful work a landowner or developer can do is to establish a site’s power story before marketing it. That means knowing the nearest transmission lines and substations, whether the utility has capacity or a plan for new generation, what the utility’s large-load rules require, and how long equipment and construction would take. Equipment scarcity matters here too, as our guide to equipment supply chain constraints explains.

• Lead with power: available MW, voltage, utility, and a realistic energization window.
• Show readiness: zoning status, contiguous buildable acres, water and fiber routes.
• Know the community: local attitudes, tax treatment and any prior large-load approvals.
• Be realistic: a large, cheap tract without a credible power path is rarely a hyperscale site.

None of this assures interest from a particular buyer, and each company weighs factors differently. If you want an independent read on how a parcel compares, you can get a site reviewed.