How much electricity will data centers use?
A simple summary of US datacenter energy use for reporters, policymakers, and content marketers
My pet peeve is when people quote statistics about energy use and carbon footprint that are out of date, misleading, or in many cases totally made up.
Here’s the best and latest research I’ve seen so, hopefully, we can avoid that for AI and data centers. If you have other stats, research, or ways to contextualize, please let me know in the comments or via DM and I’ll happily update!
TL;DR: Just give me a quote that I can use
According to Lawrence Berkeley National Laboratory, in 2023, data centers consumed 4.4% of US electricity. Driven by demand for generative AI, data centers may consume as much as 12% by 2028, more than the electricity used by California, Florida, and New Jersey combined.
This quote references an excellent Berkeley Labs paper that summarizes various projections of datacenter energy use. It also references this EIA data about electricity usage by state and this McKinsey report on AI datacenter demand. If you want to attribute it to me, feel free (Brian O’Kelley, CEO of Scope3)
Putting it in context - other options
"By 2028, data centers could consume nearly 12 times as much electricity as New York City" - references this NY ISO report on page 23 for Zone J (NYC)
"The projected 2028 data center energy use would require 72 nuclear power plants” - references US EIA data on nuclear power generation
"The electricity used to power data centers in 2028 would be able to power 55.2 million US households" - references US EIA data on household energy use
"The projected data center electricity use in 2028 would be enough to power Germany” - references IEA data on European energy
Explaining the numbers - other options
From the Berkeley Labs report (page 6): “Together, the scenario variations provide a range of total data center energy estimates, with the low and high end of roughly 325 and 580 TWh in 2028, as shown in Figure ES-1. Assuming an average capacity utilization rate of 50%, this annual energy use range would translate to a total power demand for data centers between 74 and 132 GW. This annual energy use also represents 6.7% to 12.0% of total U.S. electricity consumption forecasted for 2028.”
This McKinsey report says that “around 70 percent of total demand for data center capacity will be for data centers equipped to host advanced-AI workloads by 2030. Gen AI, currently the fastest-growing advanced-AI use case, will account for around 40 percent of the total.”
What about carbon footprint?
Here’s a follow-up post I wrote to try to answer this question…
Here’s a breakdown of the typical energy consumption:
1. Servers and IT Equipment (40-50%)
• The majority of a data center’s energy is consumed by the actual computing infrastructure—servers, storage, and networking devices.
• This demand is driven by continuous processing, high utilization, and increasing workloads from AI, cloud services, and large-scale data processing.
2. Cooling and HVAC (30-40%)
• Cooling is the second largest component. Data centers generate significant heat, and maintaining optimal operating temperatures is essential to prevent equipment failure.
• Advanced cooling techniques, such as liquid cooling or immersion cooling, are being adopted to reduce this expenditure.
3. Power Distribution and Losses (10-15%)
• Power is lost through conversion (AC to DC) and during distribution within the facility. This includes uninterruptible power supplies (UPS) and power distribution units (PDU).
4. Lighting and Ancillary Systems (5-10%)
• Lighting and other facility-related operations make up a smaller portion of the overall energy consumption.
I’m curious is research thinking about disruption in energy reduction solutions?
Leopold Aschenbrenner extrapolates to 20% of US power grid by 2030, you can check his math here: https://situational-awareness.ai/racing-to-the-trillion-dollar-cluster/