AI Data Center Power Consumption in the U.S.: 2026 Data & 2030 Forecast

How much electricity do AI data centers use in the United States? A June 2026 Lawrence Berkeley National Laboratory report estimates 649 TWh of total U.S. data-center electricity use in 2030 in its reference scenario—about 11.8% of projected U.S. electricity use. That covers data centers overall, not AI workloads alone.

It is becoming one of the biggest electricity stories in the United States.

The rapid construction of AI data centers is creating enormous new electrical loads, and recent projections suggest that data centers could consume roughly one-tenth or more of all U.S. electricity by the end of this decade.

A 2026 update from Lawrence Berkeley National Laboratory estimates that data centers could account for about 11.8% of total U.S. electricity consumption by 2030, with scenarios ranging from 9.5% to 15.3%.

That is a major change from only a few years ago.

In 2023, data centers consumed about 4.4% of U.S. electricity.

AI is now accelerating that growth.

U.S. Data Center Electricity Consumption: 2026 Report Numbers

The LBNL report, published in June 2026, gives a 2030 reference estimate of 649 TWh per year and an uncertainty range of 521–843 TWh. The corresponding share of U.S. electricity demand is an estimated 9.5%–15.3%, with 11.8% as the reference case. These are projections, not metered 2030 usage. They also include non-AI data-center activity.

For scale, 1 TWh = 1 billion kWh. A data center with a constant 100 MW facility load would use about 876 GWh (0.876 TWh) annually before allowing for downtime or load variation; if 100 MW refers only to IT load, facility consumption depends on PUE. The calculations below show why those definitions matter.

AI Is Changing U.S. Electricity Demand

For decades, electricity demand in many advanced economies grew relatively slowly.

That trend is changing.

The International Energy Agency expects U.S. electricity demand to rise by nearly 2% per year through 2030, more than twice the growth rate of the previous decade.

More importantly, data centers are expected to account for roughly half of the increase in U.S. electricity demand through 2030.

This is already visible.

According to the IEA, data centers alone accounted for around half of the increase in U.S. electricity consumption during 2025.

The reason AI matters so much is that modern AI computing requires extremely powerful processors.

Training and operating large AI models requires thousands of high-performance GPUs and other accelerated computing systems.

Those servers require electricity.

They also produce enormous amounts of heat, which means even more electricity is needed for cooling equipment, pumps, fans and other supporting infrastructure.

AI Data Centers Are Different From Normal Electricity Loads

There is another issue that I think is even more important than the total electricity consumption.

Data centers are concentrated loads.

National electricity consumption statistics can make the problem look smaller than it actually is.

A data center consuming hundreds of megawatts is not spread evenly across the United States.

That demand may appear in one particular utility territory, connected through a limited number of substations and transmission lines.

The IEA specifically notes that this geographic concentration can make data-center demand more difficult to integrate into the grid.

And unlike many residential loads, large data centers operate continuously.

PJM has reported that more than 90% of data-center load in its region can be continuous. Data centers already represent more than 7% of energy use across the PJM system.

That creates a very different electrical problem.

A home air conditioner may run heavily during a hot afternoon and then reduce its load later.

An AI data center may need hundreds of megawatts 24 hours a day.

What Does a 100 MW Data Center Use in a Day?

A 100 MW data-center load sounds abstract until it is converted into energy.

If the facility draws 100 MW continuously at the grid meter:

100 MW × 24 hours = 2,400 MWh per day

100 MW × 8,760 hours = 876,000 MWh per year = 876 GWh per year

That is the maximum annual energy at a constant 100 MW draw. Real facilities can operate below the full nameplate level, so a load-factor assumption changes the result.

Average load	Daily energy	Annual energy
100 MW	2,400 MWh	876 GWh
80 MW average	1,920 MWh	700.8 GWh
60 MW average	1,440 MWh	525.6 GWh

If 100 MW Refers to IT Load, PUE Changes the Facility Total

Another important distinction is whether the stated MW value refers to total facility demand or only the IT equipment.

A simple planning relationship is:

Total facility power ≈ IT power × PUE

For example, if 100 MW is the IT load and the assumed PUE is 1.20, the total facility demand would be approximately 120 MW. At that level continuously, annual energy would be about 1.051 TWh.

A Simple Electricity-Cost Formula

Annual electricity cost = Average MW × 8,760 × 1,000 × Electricity rate ($/kWh)

At an illustrative $0.07/kWh, a constant 100 MW load would correspond to roughly $61.3 million per year in electricity cost before demand charges, taxes, credits or special utility tariffs.

This is why MW, operating hours, PUE and electricity price must be separated when comparing data-center projects.

→ Related Guide: Can Battery Storage Reduce Data Center Grid Costs?

→ Related Guide: Why Can Data Centers Disconnect From the Grid During a Voltage Dip?

→ Related Tool: U.S. Electricity Cost Calculator

Want to turn MW into a real billing example? See our data center electricity cost per MW guide for monthly and annual kWh scenarios, energy charges and the limitations of simple rate assumptions. Actual utility tariffs may also include demand, capacity and infrastructure charges.

My View From the Electrical Construction Side

From my perspective working in electrical construction, I think discussions about AI electricity demand sometimes focus too much on generation.

People ask:

“Can America build enough power plants?”

Of course generation is important.

But electricity has to travel from the generator to the actual load.

That means the problem is also about transmission lines, substations, transformers, switchgear, protection systems and grid interconnection capacity.

👉 One publicly traded company directly exposed to this infrastructure buildout is Vertiv. Read “Vertiv AI Data Center Infrastructure: Power, Cooling and the Business Behind VRT” to see how the company connects AI electricity demand to actual power and cooling equipment.

You cannot simply build a 500 MW data center and assume that 500 MW of usable electrical capacity will immediately be available at that location.

The surrounding electrical infrastructure must be capable of delivering it.

For me, this is where the AI boom becomes particularly interesting.

AI development can move extremely quickly.

Large-scale electrical infrastructure usually cannot.

A new software model can be developed in months.

A major transmission project, power plant or substation expansion can take years.

That difference in construction speed may become one of the biggest bottlenecks for AI growth.

PJM Is Already Seeing the Pressure

PJM, which operates the power system across all or parts of 13 states and Washington, D.C., is already dealing with rapidly rising load forecasts.

Its 2026 planning analysis projects summer peak electricity demand could rise by approximately 82 GW over the next 15 years, reaching around 239 GW.

PJM identifies rapid data-center expansion as a major contributor to this structural demand growth.

The issue has become important enough that the Federal Energy Regulatory Commission took action in June 2026.

FERC ordered the six regional grid operators under its jurisdiction to examine or reform rules governing how data centers and other very large electricity consumers connect to the grid, while also considering protections for existing electricity customers.

That tells us something important.

AI power demand is no longer just a future forecast.

It is already affecting U.S. electricity planning and regulation.

Will AI Cause Electricity Prices to Rise?

This may eventually become the question that matters most to ordinary Americans.

Someone has to pay for new generation, transmission lines, substations and other infrastructure.

If massive grid investments are required primarily because of new data-center loads, regulators will have to decide how those costs should be divided between technology companies, utilities and ordinary electricity customers.

FERC’s recent actions show that this cost-allocation issue is already receiving federal attention.

The challenge is finding a balance.

AI data centers can bring investment, construction activity and economic growth.

But the electrical infrastructure needed to support them is enormous.

AI May Become an Energy Industry Story

The AI race is usually discussed in terms of GPUs, software and computing performance.

I think that view is becoming incomplete.

The next stage of AI development may depend just as much on megawatts, substations and transmission capacity as it does on computing chips.

The United States clearly has the technology and resources to build more electrical infrastructure.

The harder question is whether that infrastructure can be built quickly enough.

That is why I believe one of the most important questions in the AI industry over the next several years will not simply be:

“Who has the best AI model?”

It may also be:

“Who can secure enough reliable electricity to operate it?”

And that makes AI data-center growth an electricity story worth watching closely.

→ Related Guide: Data Center UPS Sizing: Why kVA and kW Both Matter

Sources & Official References

• Lawrence Berkeley National Laboratory — United States Data Center Energy Usage Report: 2025 Update (published June 2026)
• U.S. Department of Energy OSTI — Technical Report Record

About the author: JHOTAS-LAB is maintained by an electrical-construction and project-coordination practitioner based in South Korea. U.S. energy data and forecasts are checked against official publications.