Global data-center electricity in 2024
About 1.5% of global electricity use. This covers all data centers—not AI alone. IEA 2025 report.
AI, the Environment & Sustainability resource page
Playbook · Evidence guide + source activity
AI, the Environment & Sustainability
Understand the electricity, water, materials, infrastructure, and uncertainty behind AI—and ask better questions about local impact.
EveryoneStudentsEducatorsBusinessStartupsNonprofits
Adapt this playbook with the people who will use it.
Mississippi communities do not have identical needs, infrastructure, histories, or access. Use the evidence and tools as a starting point, then let local people shape the purpose, examples, pace, safeguards, and measures of success.
EditionEdition 1.0
Last reviewedOctober 27, 2026
Review bodyMississippi AI Collaborative
Board leadKrystal Chatman
Next reviewApril 2027, or sooner after a material source change
Report a correctionWhat you will leave with
A balanced way to evaluate AI’s direct footprint, possible wider benefits, local tradeoffs, and the quality of environmental claims.
Start here
Ask these five questions first.
- 01
Does a number describe AI alone or all data-center activity?
- 02
Is it a measurement, estimate, modeled projection, or company claim?
- 03
Does water use mean withdrawal or consumption—and is it direct or indirect?
- 04
What electricity sources, equipment, and local infrastructure are involved?
- 05
Are claimed environmental benefits actually measured against a realistic alternative?
Evidence snapshot
Read every number in context.
These figures describe different years, geographies, and scopes. Projection is not observation, and data-center totals are not AI-only totals.
IEA central projection for 2030
The IEA’s 2026 update projects global data-center electricity use roughly doubling from 485 TWh in 2025. This is modeled, not measured.
U.S. central estimate for 2030
Lawrence Berkeley National Laboratory’s 2026 update estimates a 9.5%–15.3% range for all U.S. data-center electricity use.
Variation in workload water use
A 2025 LBNL-led study found enormous variation across scenarios, driven by servers, grids, cooling, climate, utilization, and refresh cycles.
Words that change the meaning
Know the distinctions.
- Water withdrawal
- Water taken from a river, aquifer, utility, or other source. Some may be returned.
- Water consumption
- Water not returned to the same immediate system, often because it evaporates.
- Direct water use
- Water used at the data center, including by certain cooling systems.
- Indirect water use
- Water associated with generating electricity and manufacturing equipment.
- Operational emissions
- Emissions from running a facility and supplying its electricity.
- Embodied emissions
- Emissions from materials, manufacturing, transport, construction, and replacement.
Evidence behind this guide
Why every environmental number needs scope and context.
The research shows fast growth and real resource use, but also wide uncertainty and enormous variation by equipment, grid, cooling, climate, and workload. This playbook teaches readers to preserve those distinctions.
01 · International energy analysisPublished April 2025
Energy and AI
International Energy Agency
What we learned
IEA estimates data centers used about 415 TWh in 2024—around 1.5% of global electricity—and models multiple future scenarios because AI uptake, hardware efficiency, and energy constraints remain uncertain.
How it shaped this resource
Read the original source ↗We use it for the 2024 baseline and to teach that projections are scenarios, not observations or promises.
02 · Updated international analysisPublished April 2026
Key Questions on Energy and AI
International Energy Agency
What we learned
IEA’s update projects global data-center electricity use rising from about 485 TWh in 2025 to about 950 TWh in 2030 while emphasizing bottlenecks, efficiency gains, and continuing uncertainty.
How it shaped this resource
Read the original source ↗We use it to show why dates matter and why a newer projection may differ from an earlier report without making either one dishonest.
03 · U.S. national laboratory reportPublished June 2026
United States Data Center Energy Usage Report: 2025 Update
Lawrence Berkeley National Laboratory
What we learned
LBNL’s reference case estimates data centers could use 11.8% of U.S. electricity in 2030, with a 9.5%–15.3% scenario range driven by equipment shipments, utilization, chip lifetimes, and cooling.
How it shaped this resource
Read the original source ↗We use it to demonstrate how assumptions create a range and why total data-center demand must not be mislabeled as AI-only demand.
04 · Peer-reviewed primary researchPeer-reviewed · June 2025
The water use of data center workloads
Lawrence Berkeley National Laboratory researchers
What we learned
The study finds workload-level water use can vary by more than 10,000 times because server efficiency, grid water intensity, utilization, cooling, climate, and refresh cycles differ.
How it shaped this resource
Read the original source ↗We use it to reject a universal per-prompt water number and to teach readers which local and technical details change the result.
05 · Peer-reviewed modeling studyPeer-reviewed · November 2025
Environmental impact and net-zero pathways for sustainable AI servers
Nature Sustainability
What we learned
The study models energy, water, and carbon together and finds that location, grid change, efficiency, server utilization, and expansion scale create deep uncertainty and different tradeoffs.
How it shaped this resource
Read the original source ↗We use it to connect operational and indirect impacts and to show why local infrastructure and assumptions belong beside any headline result.
06 · Federal lifecycle guidanceUpdated October 2026
Sustainable Management of Electronics and Batteries
U.S. Environmental Protection Agency
What we learned
EPA places waste prevention, longer product life, reuse, repair, refurbishment, and then recycling within a whole-lifecycle approach to electronics.
How it shaped this resource
Read the original source ↗We use it to expand sustainability beyond electricity and water to procurement, equipment life, reuse, and end-of-life decisions.
07 · Mississippi public recordsCurrent state records
Permits and public records
Mississippi Department of Environmental Quality
What we learned
MDEQ provides environmental permit programs, applications, public notices, permit-board information, and guidance for groundwater and surface-water withdrawals.
How it shaped this resource
Read the original source ↗We use it to give residents a direct place to investigate a Mississippi proposal and identify the permits, water questions, and public participation opportunities that may apply.
08 · Mississippi utility recordsCurrent state records
Utility dockets
Mississippi Public Service Commission
What we learned
The Public Service Commission publishes current and archived utility dockets that document matters before the commission.
How it shaped this resource
Read the original source ↗We use it to help readers move from general energy claims to the actual utility record for a Mississippi project or proceeding.
Sources and requirements can change. Check the publication date, read the original material, and confirm the rules that apply locally before making a consequential decision.
Understand the choices
Follow the evidence from the data center to the community.
Use these six ideas to compare environmental claims, understand what is known, and spot what is still uncertain.
What AI consumes
AI runs on physical infrastructure. Electricity powers computing, storage, networking, and cooling. Water may be used directly for cooling and indirectly for electricity and chip manufacturing. Hardware depends on semiconductors, metals, minerals, global supply chains, and eventual reuse or disposal.
- Do not assign all data-center activity to AI.
- Include equipment lifecycles—not only day-to-day operation.
- Ask whether serviceable equipment can be reused before it is recycled.
Why one prompt has no universal footprint
Model size, type of task, server efficiency, utilization, cooling, local climate, electricity mix, and facility design all change the result. A fixed energy or water claim for every prompt conceals this variation.
Electricity and emissions
Efficiency measures how much computing is achieved with a unit of electricity. Carbon intensity describes emissions associated with that electricity. A facility may become more efficient while total demand rises because the facility or workload grows.
Possible environmental benefits
AI is used for electricity forecasting, building management, maintenance, research, and process efficiency. These are potential wider benefits, not automatic net reductions. Compare measured results with a credible alternative.
Questions for Mississippi communities
A new data center can affect utility planning, infrastructure, water, land use, public costs, and employment. The Commons does not take a predetermined position; it helps residents investigate evidence.
- How much electricity is expected, how will it be supplied, and what could happen to customer rates?
- Where will cooling water come from, and what are projected withdrawals and consumption?
- What upgrades are needed, who pays, and how are costs allocated?
- What construction and permanent jobs are expected?
- Which air, water, and land-use permits apply?
- What performance information will be reported publicly?
Actions by audience
Students can compare AI with other approaches; educators can teach facts versus projections; universities can evaluate vendors and hardware lifecycles; small businesses can demand measurable value; startups can design for model efficiency; nonprofits can balance mission benefit with resources; and residents can use public records to participate in local decisions.
Make a plan
Take these 5 steps.
0 of 5 completeSaved on this device
Use AI to support the work.
✓ Teaching data literacy with real energy and water evidence
✓ Comparing model or hosting efficiency during product design
✓ Using public records to ask informed local planning questions
Do not hand over the judgment.
× Repeating one fixed water or energy number for every AI prompt
× Treating all data-center demand as AI-specific
× Assuming a projected benefit has already produced a measured reduction
Use the tools
Check environmental claims for yourself.
Use these tools to compare evidence, ask better questions, and discuss tradeoffs without overstating what the numbers prove.
Environmental footprint fact sheet
A printable guide to the system behind the headline.
0 of 4 checked
Source-evaluation worksheet
Test the quality and boundaries of an environmental claim.
0 of 4 checked
Mississippi public-information guide
Find records that can answer questions about a specific proposal.
0 of 4 checked
Practice together25 minutes
Fact, Estimate, or Unknown?
Sort environmental claims by the evidence they actually provide.
- 1
Read each claim and underline the number, place, date, system, and source.
- 2
Label it measured fact, estimate, modeled projection, company-reported figure, misleading statement, or unknown.
- 3
Check whether it covers AI alone or all data-center activity.
- 4
Name the assumptions or missing information that could change the conclusion.
- 5
Rewrite the claim so a community member can understand both the finding and its limits.
Talk about it
Which missing detail most changed how you interpreted the claim?
Before you act
Share the plan with the people responsible for the work and the people who may be affected. Confirm local rules, protect private information, and name the person who can answer questions or correct a problem.