Data center energy and AI

How **data-center electricity** relates to AI growth: servers, accelerators, cooling, networking, and workload mix. IEA *Energy and AI* (10 Apr 2025) provides global modelling and policy framing for AI electricity demand and energy-syste...

What it is

How data-center electricity relates to AI growth: servers, accelerators, cooling, networking, and workload mix. IEA *Energy and AI* (10 Apr 2025) provides global modelling and policy framing for AI electricity demand and energy-system impacts. IEA *Key Questions on Energy and AI* (16 Apr 2026) revisits the nexus amid surging investment. LBNL publishes US bottom-up usage reports (2024 report; 2025 Update).

HIGH PRIORITYDIAGRAM

◷ IN PRODUCTION

Visual Spec & Architecture Diagram

Data-center energy stack: IT load (servers/accelerators) + cooling + power provisioning overhead; PUE annotation; AI training/inference as portion callout with 'methodology caution—cite IEA/LBNL-style sources, no invented kWh'.

Educational Focus: Energy literacy needs a stack, not a single scary number.

Why it matters

Policy, grid planning, and product cost all depend on facility-scale electricity—not chip marketing slides. Literacy includes reading scenario ranges, not single headlines.

How it works (plain)

Bottom-up models (LBNL) combine IT equipment shipments, per-device energy, cooling performance, and facility types. Top-down / global models (IEA) address international demand and supply. AI accelerators and inference utilization assumptions can swing totals (LBNL sensitivity scenarios).

Verified figures (label source + year)

LBNL 2024 report (via Berkeley Lab News Center, 15 Jan 2025):

US data centers ~4.4% of total US electricity in 2023
Usage rose from 58 TWh (2014) to 176 TWh (2023)
Projected 325–580 TWh by 2028, ~6.7–12% of US electricity depending on broader load growth

Source: https://newscenter.lbl.gov/2025/01/15/berkeley-lab-report-evaluates-increase-in-electricity-demand-from-data-centers/ ↗ Report page: https://eta.lbl.gov/publications/2024-lbnl-data-center-energy-usage-report ↗

LBNL 2025 Update (publication abstract pages):

Reference Case ~649 TWh in 2030; compounded uncertainty 521–843 TWh
Share of US electricity ~11.8% reference; scenario band ~9.5–15.3% by 2030
2024 report’s earlier band was 6.7–12.0% by 2028—different horizon; not a direct apples-to-apples “gotcha” without noting the year change

Sources: https://eta.lbl.gov/publications/united-states-data-center-energy-2025 ↗ · https://seta.lbl.gov/publications/united-states-data-center-energy-2025 ↗

IEA: read the reports for global projections and caveats rather than inventing a single kWh here: https://www.iea.org/reports/energy-and-ai/ ↗ · https://www.iea.org/reports/key-questions-on-energy-and-ai ↗

Everyday example

Comparing “your city’s power plant” to “one laptop charger” without saying which year or which city—is how bad charts go viral.

Try it

Write a 4-line citation card for any energy claim: geography, year, metric (TWh or % of electricity), method (bottom-up/top-down), link.

Myths

⚠️ Myth: One number from a social post beats IEA/LBNL.: ✓ Reality: Prefer primary reports + methodology footnotes.
⚠️ Myth: 2024 vs 2025 LBNL bands “prove” each other wrong.: ✓ Reality: Horizons (2028 vs 2030) and updated AI shipment/power assumptions differ—teach the discrepancy.