---
title: "AI Data-Center Electricity Demand & the Grid"
slug: ai-power-demand
type: data-index
sector: energy
canonical_url: https://deepstoryresearch.com/data/ai-power-demand
series: [global_consumption, ai_share, us_shortfall, pjm_capacity_price, per_site]
series_count: 5
data_point_count: 22
data_as_of: 2026-06
source_quality: curated_snapshot
tier: free
generated_at: 2026-08-11
---

# AI Data-Center Electricity Demand & the Grid
_AI data-center electricity demand and the grid_

> Deepstory Research context file · **Free tier** · <https://deepstoryresearch.com/data/ai-power-demand>
> Self-contained AI briefing. Drop into an LLM window or RAG pipeline.

## AI research use contract

- Treat this file as source-grounded context, not a live database. Quote `source_ref` and `data_as_of` when using numbers.
- Separate `FACT`, `ESTIMATE`, `FORECAST`, `TARGET`, and `INFERENCE`. If a row basis is unclear, say it is unclear.
- Prefer T1/T2 sources for hard facts; use T3/T4 sources as context that should be checked before money, legal, medical, operational, or policy decisions.
- Keep answer structure clean: first say what the data says, then what you infer, then what would change the conclusion.

## Research upgrade checklist

- Add regional grid/interconnection data because AI load is a local bottleneck before it is a global electricity story.
- Separate measured demand, announced campus capacity, and analyst scenarios in every table.
- Track generation source, PPA status, and energization timing for named hyperscaler campuses.
- Add a downside scenario where chip supply or model-efficiency gains reduce projected power draw.

## TL;DR — index summary

AI is now a grid-scale load: global data centers used ~415 TWh in 2024 (~1.5% of world electricity), AI workloads alone drew ~44 GW in 2025, and PJM capacity-auction prices climbed steeply off a $50/MW-day (2022/23) base. A single hyperscale site pulls ~100 MW — the electricity of ~100,000 households. Demand is arriving faster than capacity, and the grid is repricing.

## What this index covers

Five series: global data-center electricity consumption, the AI vs non-AI share of that load, PJM capacity-auction clearing prices, the US demand-vs-capacity shortfall, and per-site power draw.

**Series in this index:**

- `global_consumption` — Global data-center electricity use, TWh.
- `ai_share` — AI vs non-AI data-center power (GW).
- `us_shortfall` — US data-center demand vs available capacity (GW).
- `pjm_capacity_price` — PJM capacity-auction clearing price, $/MW-day.
- `per_site` — Power draw of a representative site/entity.

**Entities tracked:** IEA, McKinsey, Morgan Stanley, PJM (largest US grid operator), JLL, hyperscalers.

## Key findings

- Global data centers used ~415 TWh in 2024 — ~1.5% of world electricity (iea-electricity).
- AI workloads drew ~44 GW in 2025 of ~103 GW total DC capacity (McKinsey / JLL) (mckinsey-aipower).
- PJM capacity-auction clearing prices rose steeply off a $50/MW-day (2022/23) base (pjm-capacity).
- US data-center demand was ~40 GW in 2024 with supply roughly matched — a gap that widens ahead (morgan-stanley).
- A typical hyperscale data center draws ~100 MW — roughly 100,000 households (iea-electricity).

## Evidence basis map

The free tables above show `source_ref` but not the row-level `value_basis`. This section mirrors the visible rows only, so AI tools can separate observed values from estimates, forecasts, targets, and derived values.

| series | row | source_ref | value_basis |
| --- | --- | --- | --- |
| global_consumption | year=2024; twh_historic=415 | iea-electricity | IEA Energy & AI: ~415 TWh in 2024 (~1.5% of global electricity); corroborated by S&P Global |
| global_consumption | year=2030; twh_scenario=945 | iea-electricity | IEA base case ~945 TWh by 2030 (S&P Global: roughly doubles) |
| global_consumption | year=2035; twh_scenario=1200 | iea-electricity | IEA base case ~1,200 TWh by 2035 |
| ai_share | year=2025; ai_gw=44; non_ai_gw=59 | mckinsey-aipower | McKinsey “AI power”: AI workload power ~44 GW in 2025; total DC capacity ~103 GW (JLL) ⇒ non-AI ~59 GW (deduction) |
| ai_share | year=2028; ai_gw=100; non_ai_gw=75 | mckinsey-aipower | Interpolated along McKinsey base-case trajectory (AI 44→156 GW; total ~103→219 GW, 2025→2030) — labeled estimate |
| ai_share | year=2030; ai_gw=156; non_ai_gw=63 | mckinsey-aipower | McKinsey “AI power” base case: AI ~156 GW of ~219 GW total DC capacity by 2030 ⇒ non-AI ~63 GW (deduction) |
| us_shortfall | year=2024; demand_gw=40; capacity_gw=40 | morgan-stanley | Morgan Stanley: US data-center demand ~40 GW in 2024 (supply roughly met) |
| us_shortfall | year=2028; demand_gw=100; capacity_gw=55 | morgan-stanley | Morgan Stanley: ~100 GW demand by 2028 vs ~45 GW shortfall ⇒ ~55 GW available (Deloitte corroborates 80→150 GW total DC) |
| pjm_capacity_price | delivery_year=2022-23; clearing_price_usd_per_mw_day=50 | pjm-capacity | PJM BRA 2022/23 RTO clearing price $50.00/MW-day |
| pjm_capacity_price | delivery_year=2023-24; clearing_price_usd_per_mw_day=34.13 | pjm-capacity | PJM BRA 2023/24 RTO clearing price $34.13/MW-day |
| pjm_capacity_price | delivery_year=2024-25; clearing_price_usd_per_mw_day=28.92 | pjm-capacity | Topics_Content/07_New_Trending_Topics_2026.md §1.3 — PJM 2024-25 clearing price $28.92/MW-day |
| pjm_capacity_price | delivery_year=2025-26; clearing_price_usd_per_mw_day=269.92 | utility-dive | PJM BRA 2025/26 RTO clearing price $269.92/MW-day (Utility Dive; PJM report) |
| pjm_capacity_price | delivery_year=2026-27; clearing_price_usd_per_mw_day=329.17 | pjm-capacity | Topics_Content/07_New_Trending_Topics_2026.md §1.3 — PJM 2026-27 clearing price $329.17/MW-day (>10× the 2024-25 print) |
| per_site | entity=Typical hyperscale data center; draw_mw=100 | iea-electricity | A typical hyperscale DC ~100 MW (~100,000 households) |
| per_site | entity=City of New Orleans; draw_mw=1667 | sp-global | New Orleans load ~1.67 GW (Meta Hyperion ≈ 3× New Orleans) |
| per_site | entity=Meta Hyperion AI campus (planned); draw_mw=5000 | sp-global | Meta Hyperion campus (Louisiana) needs ≥5 GW (~3× New Orleans) |

## The series (per-chart briefings)

### Global DC consumption — line
**Direct answer:** Global data-center electricity use, TWh.
**Time bracket:** annual

| year | twh_historic | twh_scenario | source_ref |
| --- | --- | --- | --- |
| 2024 | 415 |  | iea-electricity |
| 2030 |  | 945 | iea-electricity |
| 2035 |  | 1200 | iea-electricity |

**Read:** A steep climb — AI is bending the curve up.

### AI vs non-AI share — stacked bar
**Direct answer:** AI vs non-AI data-center power (GW).
**Time bracket:** annual

| year | ai_gw | non_ai_gw | source_ref |
| --- | --- | --- | --- |
| 2025 | 44 | 59 | mckinsey-aipower |
| 2028 | 100 | 75 | mckinsey-aipower |
| 2030 | 156 | 63 | mckinsey-aipower |

**Read:** AI is the fast-growing slice inside a growing whole.

### US demand vs capacity — line
**Direct answer:** US data-center demand vs available capacity (GW).
**Time bracket:** annual

| year | demand_gw | capacity_gw | source_ref |
| --- | --- | --- | --- |
| 2024 | 40 | 40 | morgan-stanley |
| 2028 | 100 | 55 | morgan-stanley |

**Read:** Balanced in 2024; the gap is the forward risk.

### PJM capacity price — bar
**Direct answer:** PJM capacity-auction clearing price, $/MW-day.
**Time bracket:** by delivery year

| delivery_year | clearing_price_usd_per_mw_day | source_ref |
| --- | --- | --- |
| 2022-23 | 50 | pjm-capacity |
| 2023-24 | 34.13 | pjm-capacity |
| 2024-25 | 28.92 | pjm-capacity |
| 2025-26 | 269.92 | utility-dive |
| 2026-27 | 329.17 | pjm-capacity |

_+1 more row on the live page and in the full working dataset._

**Read:** The grid repricing capacity as AI load arrives.

### Per-site draw — bar
**Direct answer:** Power draw of a representative site/entity.
**Time bracket:** snapshot

| entity | draw_mw | source_ref |
| --- | --- | --- |
| Typical hyperscale data center | 100 | iea-electricity |
| City of New Orleans | 1667 | sp-global |
| Meta Hyperion AI campus (planned) | 5000 | sp-global |

**Read:** Contextualises the abstract GW figures at the facility level.

## Cross-series synthesis — why this matters

The five series stack into one causal chain: rising consumption (global_consumption) is increasingly AI-driven (ai_share), that demand is outrunning capacity (us_shortfall), and the market clears the imbalance through capacity price (pjm_capacity_price) — which is why household power bills are rising in data-center-heavy states (see the ai-datacenter-power-bills index). Per-site draw makes the scale legible: each new hyperscale campus is a small city's worth of load.

## Entities & relationships

| Entity | What they do | Key stat | Relationship |
| --- | --- | --- | --- |
| PJM | Largest US grid operator | Capacity prices up off $50/MW-day base | Where AI load repricing shows up first |
| IEA | Global energy agency | ~415 TWh DC use (2024) | Primary for consumption + per-site |
| Hyperscalers | AI data-center builders | ~44 GW AI load (2025) | Drive demand; link to ai-capex-vs-defense |
| Morgan Stanley / McKinsey / JLL | Demand-capacity analysts | GW demand/supply estimates | Provide the shortfall + share splits |

## Timeline of key events & decisions

- **2022/23** — PJM capacity clears ~$50/MW-day → the low base before AI load repriced capacity
- **2024** — ~415 TWh global DC use; US demand ~40 GW ≈ supply → the balanced starting point
- **2025** — AI workloads ~44 GW → the fast-growing slice pushing the shortfall open

## Cross-industry ripple

- **Household utilities:** Capacity-price increases pass through to residential bills (ai-datacenter-power-bills).
- **Nuclear / gas / renewables:** The shortfall pulls forward new generation, including nuclear PPAs. _(inference)_
- **Semiconductors:** AI capex and power demand are two views of the same buildout (ai-capex-vs-defense, frontier-ai-race).

## Non-obvious reads (interpretation)

_This section is interpretation, not sourced fact — each item names the observation it is built on and a confidence level._

- **Observation:** US demand roughly matched supply in 2024, yet capacity prices are climbing.
  **Read:** Markets are pricing the forward shortfall, not the current balance — so the capacity-price series leads the physical-shortfall series, and power, not chips, may be the binding constraint on AI scaling. _(confidence: medium)_

## Glossary / key terms

- **TWh** — Terawatt-hour — annual electricity-consumption unit.
- **GW** — Gigawatt — instantaneous power-capacity unit.
- **PJM** — The largest US regional grid operator (13 states + DC).
- **Capacity auction** — A forward market paying generators to be available; $/MW-day is its clearing price.
- **Hyperscale** — A very large cloud/AI data center (often ~100 MW+).

## How to use this with AI

Paste this file into an LLM context window (or a RAG store) and ask cross-series questions. The tables carry a `source_ref` per row; the source registry below maps each ref to a named source and a trust tier.

### Suggested prompts (multi-series)

```text
Using global_consumption, ai_share and us_shortfall, argue whether power or chips is the binding constraint on AI scaling.
```

```text
Trace how the pjm_capacity_price series connects to household bills in the ai-datacenter-power-bills index.
```

```text
Relate this index to ai-capex-vs-defense: is the capex or the power buildout the leading indicator?
```

## Sources & trust

| ref | source | trust tier | url |
| --- | --- | --- | --- |
| eia-opendata | US EIA Open Data | T1 · Government / regulator / central bank | https://www.eia.gov/opendata/ |
| iea-electricity | IEA — Energy and AI / Electricity | T3 · Reputable press / research org / OWID | https://www.iea.org/reports/energy-and-ai |
| pjm-capacity | PJM Capacity Auction Results (BRA reports) | T2 · Primary filing / official body | https://www.pjm.com/markets-and-operations/rpm |
| sp-global | S&P Global — data-center power demand | T3 · Reputable press / research org / OWID | https://www.spglobal.com/ |
| morgan-stanley | Morgan Stanley Research — data-center power | T3 · Reputable press / research org / OWID | https://www.morganstanley.com/ |
| deloitte | Deloitte — data-center power demand insights | T3 · Reputable press / research org / OWID | https://www.deloitte.com/ |
| utility-dive | Utility Dive — PJM capacity prices | T3 · Reputable press / research org / OWID | https://www.utilitydive.com/ |
| ember | Ember Climate — global electricity | T3 · Reputable press / research org / OWID | https://ember-energy.org/ |
| owid-energy | Our World in Data — Energy | T3 · Reputable press / research org / OWID | https://ourworldindata.org/energy |
| mckinsey-aipower | McKinsey — AI power: Expanding data center capacity to meet growing demand | T3 · Reputable press / research org / OWID | https://www.mckinsey.com/industries/technology-media-and-telecommunications/our-insights/ai-power-expanding-data-center-capacity-to-meet-growing-demand |
| jll-dc | JLL — Global Data Center Market Outlook | T3 · Reputable press / research org / OWID | https://www.jll.com/en-us/insights/market-outlook/data-center-outlook |
| crs-dc-energy | Congressional Research Service — Data Centers and Their Energy Consumption (R48646) | T1 · Government / regulator / central bank | https://www.congress.gov/crs-product/R48646 |
| fortune-dc-costs | Fortune — Data centers could hike power costs in some states over 50% by 2030 | T3 · Reputable press / research org / OWID | https://fortune.com/2026/05/19/data-centers-electricity-costs-us-public-opinion/ |
| ec-dc-map | Electric Choice — US Data Center Power Consumption Map by State (2026) | T3 · Reputable press / research org / OWID | https://www.electricchoice.com/datacenters/ |

## Caveats & what this index cannot answer

- Exact future DC demand (series mix historicals with analyst estimates).
- Per-utility retail rate impacts (see the power-bills index for state-level).
- Estimates and forward targets are labelled in the working dataset; never read a labelled estimate or forecast as a settled figure.
- This is an observational data index, not investment, legal, or medical advice.

## Data freshness & methodology

- **Last updated:** 2026-08-11
- **Data as of:** 2026-06
- **What changed most recently:** PJM capacity prices repricing sharply above the 2022/23 $50/MW-day base as AI load arrives.

**Methodology (as seeded):**

> Source-backed values are seeded for all seven charts: global data-center
> electricity consumption (IEA Energy & AI, historic vs base-case scenario,
> corroborated by S&P Global), the AI vs non-AI share of data-center power
> capacity (McKinsey base case, corroborated by JLL), the PJM capacity-market
> clearing price by delivery year (PJM Base Residual Auction reports,
> corroborated by Utility Dive), the US data-center demand-vs-supply gap to
> 2028 (Morgan Stanley; Deloitte), per-site power draw vs a city, US
> data-center power demand (2025 vs 2028), and the projected electricity-price
> impact by region/scenario.
> 
> Every numeric point carries a sources[].ref and a value_basis. The 2028
> available-capacity figure is derived from Morgan Stanley’s ~45 GW shortfall
> estimate against ~100 GW demand (labeled in the value_basis). Demand figures
> trace to CRS data-center energy analysis; the Virginia (+57% by 2030) and
> national wholesale (+6% to +29% by end of decade) price impacts trace to
> Fortune / public-opinion reporting on data-center power costs.
> 
> ESTIMATE: the AI-share chart uses McKinsey’s published AI-vs-total capacity
> trajectory (AI ~44 GW in 2025 → ~156 GW of ~219 GW total by 2030); the 2028
> point is interpolated along that trajectory and the non-AI slice is a
> deduction (total − AI). It is a published-estimate split, not a measured
> per-year megawatt count.
> 
> CAVEAT: price-impact figures are forward scenarios, not realized prices, and
> depend on how fast new generation is added. Ranges are shown as separate
> low/high bars rather than a single point. Re-verified 2026-07-17.

---

_Free tier. The full row-level dataset and per-source detail live on the live page and in the working file. Canonical: <https://deepstoryresearch.com/data/ai-power-demand>. Deepstory Research · https://deepstoryresearch.com_