Skip to content

Northwest Louisiana · A plain-language explainer

Amazon is building a $12 billion data center campus in Northwest Louisiana.

Here is how it uses electricity, what it powers, and what it could mean for your power bill.

Caddo & Bossier Parishes

Announced February 2026Served by SWEPCODeveloped with STACK Infrastructure

Every figure on this page is tagged by source, so you can see where it comes from.

The scale

How a data center uses power

A data center is a building full of computers that runs every hour of every day. The newest campuses draw an enormous, steady amount of electricity, and most of it never reaches a screen you can see.

A modern hyperscale campus can draw

90 to 100 MW

several times the around 30 MW a large site drew only a few years ago, as AI workloads grew.

For a sense of scale, a single large data center can use about 100,000 homes worth of electricity. That is the kind of load Northwest Louisiana is planning around.

One caveat before the widgets below: Amazon has not disclosed how much power the Louisiana campuses will draw. The 100 MWused here is an industry reference size for a modern hyperscale campus, not a figure for this project.

Where the power actually goes

Not all of a data center's electricity does computing. Efficiency is measured by PUE, the total power drawn for each unit that reaches the chips.

1.50
1.1 · efficient (as low as 1.1)1.6 · industry average upper end
Computing
67%
Servers and chips doing the actual work
Cooling
25%
Keeping all that hardware from overheating
Other overhead
8%
Power conversion, lighting and building loads

Put differently: of a 100 MW campus running at this efficiency, about 67 MW does computing while roughly 33 MW goes to cooling and overhead.

A single large data center can use about 100,000 homes worth of electricity.

A typical AI chat query is estimated at about 0.3 Wh, about the same order of magnitude as a traditional web search. Estimates have fallen as hardware and models improved; per-query numbers are small, and scale comes from total volume.

Note: the split between cooling and other overhead is a modeling assumption. Roughly three quarters of overhead is cooling, and the rest is power delivery, lighting and building loads. The exact share varies by site and season.

Source: Uptime Institute Global Data Center Survey (2024). Industry-average PUE has sat near 1.5 to 1.6 for several years. • Lawrence Berkeley National Laboratory, “2024 United States Data Center Energy Usage Report,” and Uptime Institute. The best new facilities operate near 1.1. • International Energy Agency (IEA), data-center electricity analysis (“Electricity 2024” and “Energy and AI,” 2025). The largest facilities now draw on the order of 100 MW. • International Energy Agency (IEA), “Energy and AI” (2025). A large data center can consume roughly as much electricity as 100,000 households. • Epoch AI energy estimate (February 2025); OpenAI has cited about 0.34 Wh per query. A typical modern AI chat query is estimated at roughly 0.3 Wh, the same order of magnitude as a traditional web search. Earlier figures near 2.9 Wh (IEA, “Electricity 2024”) are now considered about 10x too high for typical queries, as hardware and models improved.

What it powers

The cloud behind everyday life, and jobs close to home

Most of what a data center does is invisible. It is the engine behind apps and services people use without thinking about them, and increasingly behind artificial intelligence.

Streaming and apps

The video you watch, the maps you follow, the photos backed up to the cloud.

Online business

Storefronts, payments and the back-office systems that keep them running.

AI assistants

The newest, power-hungry workloads writing, searching and answering questions.

Closer to home

That load also shows up as local economic activity. These are the figures the project and the state have put forward for the Northwest Louisiana campus.

$12 billion

Amazon's planned investment in the campus

540

Direct on-site jobs at the campus

1,710

Indirect community jobs supported

1,500

Construction jobs during the build

A wider, statewide number you will see quoted

$4.7 billion

Amazon's total investment across Louisiana since 2010

Statewide cumulative since 2010, not this campus alone.

7,500

Amazon jobs created across Louisiana since 2010

Statewide cumulative since 2010, not this campus alone.

These two figures are Amazon's cumulative totals across all of Louisiana since 2010, reported by the state. They are not from this campus alone, so it is worth keeping them separate from the project numbers above.

How the grid delivers it

From the power plant to the server room

Electricity reaches a data center the same way it reaches your house, through a chain of generation, transmission and distribution. A very large customer interacts with each link a little differently.

In Northwest Louisiana the local utility is SWEPCO (Southwestern Electric Power Company), an AEP utility. The project is also tied to up to 200 MW of new Louisiana solar, enough to power about 44,000 homes.

From power plant to data center

Pick a stage to see what it does, and how a very large customer changes the picture.

Generation

Energy vs. capacity

Two different things are bought here. Energy is the actual kilowatt-hours produced over time. Capacity is having enough megawatts ready for the moment everyone needs power at once. A steady, around-the-clock data center buys a lot of energy, and it also needs firm capacity.

In Northwest Louisiana the project is tied to up to 200 MW of new Louisiana solar, enough to power about 44,000 homes.

Source: North Louisiana Economic Partnership (NLEP), Amazon, and Louisiana Economic Development (LED), 2026 project announcement.

The shape of the flow above follows the cost structure of the grid itself: costs build up across generation, transmission and distribution, and a big new customer touches each layer in a different way.

Who pays?

The question on everyone's power bill

This is the part people most want answered. A large new customer adds costs to the grid, and it also adds revenue. Whether neighbors come out ahead depends on how those two sides balance.

It adds costs

Serving a huge load can mean new generation, upgraded lines and equipment. If those costs were spread to everyone, other customers could end up paying for them.

It adds revenue

A large customer also pays a great deal into the system. When it pays its full share, the grid's fixed costs spread across more sales, which can ease pressure on rates.

100%

Under its arrangement with SWEPCO, Amazon covers 100% of its own energy-infrastructure costs. The point of that structure is to keep the new load from landing on other ratepayers.

How a big customer's bill is built

The E3 study describes a large customer's bill in three parts. How a utility sets each one decides whether a big load helps or hurts everyone else.

Demand charge

based on peak megawatts

What you pay for the maximum power you pull at once. A steady campus has a high, predictable peak.

Energy charge

based on total kilowatt-hours

What you pay for the electricity actually used. Running around the clock means a lot of energy.

Customer charge

a fixed monthly amount

A flat fee that covers metering, billing and basic service, the same idea on your own bill.

Independent analysis points the same direction. Studying data centers comparable to this one, E3 found that a well-designed large load can create surplus value for other customers, around about $3.4 million a year for a 100 MW facility today, putting gentle downward pressure on rates.

What a big load could do to rates

When a large customer pays its full share, the grid's fixed costs spread across more electricity sales. E3 modeled how much surplus value that can create. Size the facility and see the modeled figure.

This is proxy analysis, not a SWEPCO measurement. The figures come from E3's national study of comparable Amazon data centers, modeled on PG&E (California), Dominion (Virginia), Entergy Mississippi, and Umatilla Electric Cooperative (Oregon). They are not a SWEPCO rate, and not a measurement of your Louisiana bill.

100 MW
Smaller facility100 MW referenceLarger facility

Modeled surplus value, 2025

$3.4 million

at $33,500 per MW

Up to, by 2030

$6.1 million

up to $60,650 per MW

For reference, E3's 100 MW reference facility lands near about $3.4 million in 2025 and up to $6.1 million by 2030 in E3's model.

Source: E3 (Energy + Environmental Economics), “Tailored for Scale: Designing Electric Rates and Tariffs for Large Loads,” December 2025. National analysis of proxy utilities, not SWEPCO.

The bigger picture

Water, clean energy, and what is still uncertain

Electricity is one piece. A project this size also touches water and new power supply, and it carries real open questions. A neutral look means holding all of it at once.

Water

Cooling the hardware takes water. The project's reported figures put that use in context.

114 million gallons

Water used per year for cooling

less than 2%

Share of the regional water system's capacity

about 1,052 homes

A relatable size for that annual water use

up to $400 million

Investment in regional water infrastructure

New power supply

Some of the load is being matched with new generation, locally and, in the wider analysis, across the country.

up to 200 MW

New Louisiana solar tied to the project

about 44,000 homes

What that new solar could power

about 4.2 GW

New solar and wind Amazon is poised to enable in E3's case-study regions

E3 proxy-utility modeling (PG&E (California), Dominion (Virginia), Entergy Mississippi, and Umatilla Electric Cooperative (Oregon)), not a Louisiana-specific result.

more than 1 million homes

What 4.2 GW of new renewables could power

E3 proxy-utility modeling (PG&E (California), Dominion (Virginia), Entergy Mississippi, and Umatilla Electric Cooperative (Oregon)), not a Louisiana-specific result.

The project also includes a $250,000 community investment fund.

What is still uncertain

None of the figures above remove the open questions. Three are worth keeping in view.

Load growth is real and fast

Big new loads can strain a grid. If demand outpaces new supply, costs can rise for everyone, which is exactly why these arrangements get close scrutiny.

Who pays over the long run is still being decided

The surplus today depends on how shared upgrades are paid for across decades. The E3 study calls this the central rate-design question, not a settled answer.

These are models, not local measurements

The surplus figures come from national analysis of comparable utilities. They are a useful guide, not a reading of a SWEPCO bill in Caddo or Bossier Parish.

A data center is a large, steady electricity customer. Whether it helps or hurts your bill comes down to the rules that decide who pays for the grid, and those rules are worth watching closely.

How to read the source tags

  • LANorth Louisiana Economic Partnership (NLEP), Amazon, and Louisiana Economic Development (LED), 2026 project announcement.
  • E3E3 (Energy + Environmental Economics), “Tailored for Scale: Designing Electric Rates and Tariffs for Large Loads,” December 2025. National analysis of proxy utilities, not SWEPCO.
  • IndustryIndependent public sources on data centers and the grid (IEA, Lawrence Berkeley National Laboratory, Uptime Institute).

Every figure on this page is tagged inline by source. See the full, linked list of sources below.

Sources

Where every figure comes from

Each number on this page is tagged inline by source family. Here is the full, linked list, grouped the same way, so you can go check the record yourself.

Louisiana project facts

North Louisiana Economic Partnership (NLEP), Amazon, and Louisiana Economic Development (LED), 2026 project announcement.

E3 national study

E3 (Energy + Environmental Economics), “Tailored for Scale: Designing Electric Rates and Tariffs for Large Loads,” December 2025. National analysis of proxy utilities, not SWEPCO.

General-industry context

Independent public sources on data centers and the grid (IEA, Lawrence Berkeley National Laboratory, Uptime Institute).

Links open publisher pages or official posts. Where a figure is a national or proxy-utility model rather than a Louisiana measurement, the inline tag and its caption say so.