Hyperscale campus

How it works, part by part

1. 1
   Grid connection. Brings electricity from the grid to the site.^[2] In the guide: Grid connection and substation
2. 2
   Substation. Lowers the grid’s high voltage so the site can use it.^[2] In the guide: Grid connection and substation
3. 3
   Switchgear and transformers. Transformers change the voltage. Switchgear, a mix of disconnect switches and circuit breakers, controls and protects the electrical equipment and can isolate parts of it.^[3] In the guide: Transformers and switchgear
4. 4
   Uninterruptible power supply (UPS). Provides backup power, and can run on battery banks or other technologies.^[4] In the guide: Transformers and switchgear
5. 5
   Backup generators. Give the site another source of power if the grid’s supply stops.^[4] In the guide: Backup generators
6. 6
   Data hall. The floor where the computing equipment runs. The lab counts floor area by a module like this, not by the whole campus.^[1] In the guide: Data halls and servers
7. 7
   Server racks. Hold the computing equipment, which runs continuously to answer network requests.^[1] In the guide: Data halls and servers
8. 8
   Cold and hot air paths. Keep the cool air supplied to the equipment from mixing with the hot air it gives off.^[4] In the guide: Cooling inside the building See it in one row of servers
9. 9
   Outside-air cooling unit. Cools the space with outdoor air when the weather allows.^[1] In the guide: Cooling inside the building
10. 10
    Liquid cooling loop. Carries heat away from dense equipment such as AI servers, through plates on the chips. Not every site has one.^[4][1] In the guide: Cooling inside the building
11. 11
    Heat released outdoors. Moves the heat into the outdoor air, as dry coolers do.^[1] In the guide: Heat out
12. 12
    Water connection. Brings in water where the cooling design uses it on site.^[1] In the guide: Water in and out
13. 13
    Generation on the site. A power plant connected on the site’s side of the grid connection. Not every campus has one, so it is drawn in outline.^[5]

How power moves

1. True of data centers generally Power usually arrives from the utility at a higher voltage. A transformer steps it down, and it passes through switchgear and backup equipment before it is shared out to the racks.^[3] Parts 1, 2, 3, 4, 5
2. This type A national lab’s model assumes hyperscale and AI data centers tend to run with higher power-supply efficiency than other categories.^[1] That is an assumption about a category, not a measurement of any site, and not a ranking. Part 4
3. One example The state said the utility would add power plants to its system to meet growing demand, including from this campus.^[6] Part 1

How heat moves

1. True of data centers generally In air-cooled halls, the aim is to keep cooling air from mixing with hot exhaust air.^[4] Part 8
2. This type The lab’s model assumes these sites can use outside conditions for much of their cooling, because they allow wider temperature and humidity ranges.^[1] This doesn’t show how any one site cools. Part 9
3. True of data centers generally An airside economizer uses outdoor air to cool the space in favorable weather.^[1] Part 9
4. This type Water use is uncertain. Some hyperscale sites report 0.1 to 0.3 liters of water per kilowatt-hour of electricity used by computing equipment, much higher than the lab’s simulation of similar systems. The details of those systems are not known.^[1] Parts 11, 12

AI workloads

AI does not get a page of its own here. The national lab report used on this page counts AI servers inside its hyperscale and large-scale colocation categories, and models the cooling of accelerated ones separately.

• A national lab’s model splits accelerated AI servers evenly between hyperscale and large-scale colocation, and other AI servers 80% and 20%, rather than treating AI as a separate kind of building.^[1] That is a modeling choice, not a count of where AI servers sit.
• The same model gives accelerated AI servers their own cooling category, because they use a mix of liquid and air cooling.^[1]
• Liquid cooling is an emerging technology for dense equipment such as AI servers.^[1]
• With cold plates, liquid takes the heat away at the chip instead of room air.^[4]
• The Louisiana example on this page is described as a hyperscaler data center that carries AI workloads.^[6]

On-site power

On-site power is a way of connecting a load to a generator, not a kind of building, so it is covered here.

• Federal regulators describe co-located load by how it is wired: it is physically connected to a generator’s facilities on the generator’s side of the grid connection.^[5] That definition is for the PJM grid region’s tariff, and later orders were not captured, so it may not be the current national rule.
• One example In one Ohio case, state siting staff described planned fuel cells on an Amazon campus running behind the meter, meaning the output goes to that one customer’s electric load.^[7] Amazon — Scioto Darby campus
• One example The same plan would run alongside the grid and would send no power back to it.^[7]
• One example The planned system is rated at a nominal 72.9 megawatts (MW) net. That is generation, not computing load.^[7]

The Ohio report does not call this campus hyperscale, so it shows on-site power only and is not this page’s example.

How this type differs

These differences come from how the sources define each type. None of them makes one type preferable.

• Colocation is built to lease to others. A hyperscale site is built by the company that runs the services on it.^[1] Colocation and leased campuses
• An enterprise data center serves one organization’s own internal use. A hyperscale site runs services at massive scale.^[1] Enterprise and government data center
• A national lab’s model treats edge sites as small, with lower power-supply efficiency and narrower temperature ranges.^[1] That is an assumption about categories, not a ranking. Edge sites

One real example

Meta Richland Parish Data Center is one example of this type, chosen by hand. It is one of 5 such examples; most records on this site are not typed, including 3,110 of the 3,115 source listings on the opening map, which can overlap.

The state’s announcement calls it a hyperscaler data center.^[6]

• The state announced a campus of 4 million square feet on a 2,250-acre former farm site.^[6]
• The state said that at its longest the data center reaches more than a mile from front to back.^[6]
• The state said the utility, Entergy, would add power plants to its system to serve growing demand, including this campus.^[6]

These figures come from an announcement dated 4 December 2024, so they describe what was announced, not what has been finished.^[6]

Questions to ask about a project of this type

1. Which company will use this campus, and does a public filing name it or only a project company?
2. How many buildings are planned, in what phases, and which phases are approved, under construction or operating?
3. Where will the electricity come from: new plants on the utility’s system, a special contract, or generation on the site?
4. Is any figure a request, a contract, a design or a measurement, and what does it cover?
5. How does the design release heat in the hottest weather, and how much water does that use?
6. If there is generation on the site, can the campus also draw from the grid or send power back?

Go further on this site

All types of data center · How data centers work · The 3D map · A megawatt of what?

Sources

1. 2024 United States Data Center Energy Usage Report (LBNL-2001637). Lawrence Berkeley National Laboratory. National laboratory. Cited at Table 4.1, p. 36; p. 36; p. 35; p. 41; p. 42; p. 39; p. 47; p. 38. Captured 3 November 2026.
2. Electricity explained: How electricity is delivered to consumers. U.S. Energy Information Administration. Government. Cited at How electricity is delivered. Captured 3 November 2026.
3. Data Center Power Equipment Thermal Guidelines and Best Practices, TC 9.9 white paper. ASHRAE Technical Committee 9.9. Standards group. Cited at section 4; power distribution overview. Captured 3 November 2026.
4. Best Practices Guide for Energy-Efficient Data Center Design (revised July 2024). U.S. Department of Energy, Federal Energy Management Program. Government. Cited at section 6.1.1; section 6.1; section 4; section 5.6. Captured 2 November 2026.
5. 193 FERC 61,217, Order on Show Cause Proceeding, PJM co-location, Docket EL25-49-000. Federal Energy Regulatory Commission. Government. Cited at Part IV.B.1. Captured 3 November 2026.
6. Meta Selects Northeast Louisiana as Site of $10 Billion Artificial Intelligence Optimized Data Center. Louisiana Economic Development. Government. Cited at announcement body; news release date; summary bullets. Captured 3 November 2026.
7. Staff Report of Investigation, Case 25-0649-EL-BLN, Scioto Darby Creek Road Fuel Cell System Project. Ohio Power Siting Board staff, copy hosted by the City of Hilliard. Government. Cited at Staff Report, Project Description. Captured 3 November 2026.

Sources were captured on the date shown and can change. An operator’s explanation is used only for how a part works, never for a definition or a comparison.