Data Center Power Is Becoming the Real AI Scarcity Asset

Bottom line

If compute is the input, power is the throughput constraint.

The biggest mistake investors make about AI infrastructure is to stop at GPUs. The second-order bottleneck is electricity: how fast it can be procured, permitted, connected, transformed, and delivered to a site that is already booked out by hyperscalers and colocators.

The IEA and EIA are now saying this plainly. Data centers are not a niche power user anymore. They are a new category of load growth with enough scale to affect regional grids, transmission planning, and utility capex.

My view: the next wave of AI winners will be defined less by model quality than by time-to-power. If you cannot secure megawatts quickly, the best training plan in the world is just a slide deck.

The demand curve

The load path is steep enough to change utility planning.

Global data-center electricity trajectory

IEA data show the scale of the demand shock. This chart uses the IEA's projected electricity supply to data centers in 2024, 2030, and 2035.

단위: TWh

2024

IEA base year

460

2030

IEA base case

1,000

2035

IEA base case

1,300

What the official projections say
Source	Published figure	Why it matters
IEA Energy and AI	Data-center electricity consumption set to more than double to around 945 TWh by 2030.	This is the clearest global statement that AI load is becoming a structural power trend.
IEA energy supply for AI	Electricity generation to supply data centers projected to grow from 460 TWh in 2024 to over 1,000 TWh in 2030 and 1,300 TWh in 2035.	That path implies more generation, more transmission, and more equipment orders.
IEA key questions on energy and AI	AI-focused data-center electricity demand surged 50% in 2025.	AI workloads are growing faster than the broader digital economy.
EIA AEO2026	U.S. server electricity consumption could reach 446-818 billion kWh by 2050.	The U.S. grid has to plan for a very wide demand band.

Regional bottlenecks

The grid is already showing where the pressure will land first.

Where the stress is appearing
Region / system	Published signal	Why it matters
Virginia / Dominion zone	EIA says it has the largest concentration of data centers in the world and expects the largest absolute increase in summer peak demand through 2030.	This is the clearest live example of data-center load reshaping a regional grid.
ERCOT / Texas	EIA expects ERCOT demand to increase 7% in 2025 and 14% in 2026 as large data centers and crypto facilities come online.	Fast load growth turns power procurement and interconnection into strategic assets.
U.S. overall	IEA says data centers account for nearly half of electricity demand growth between now and 2030.	That is a direct policy and utility planning problem, not just a private-sector one.
Global system	IEA says renewables, gas, coal, and eventually nuclear will all play a role in meeting incremental demand.	The growth mix determines which fuel and infrastructure vendors benefit.

Transmission is slow, so the power shortage shows up before the new generation does.
Substations, transformers, breakers, and switchgear can become as important as the generation source itself.
A site with available land and a fast interconnect can be worth more than a cheaper site that is stuck in a queue.

Investment implications

The beneficiaries are broader than just utilities.

Who gets paid when AI load rises
Layer	Likely beneficiary	Mechanism
Utilities	Regulated rate base growth	New load can justify more transmission and distribution capex.
Independent power producers	More contracted demand	Hyperscalers want long-term, predictable power supply.
Gas turbines and grid equipment	More equipment orders	Load growth creates a backlog in physical infrastructure.
Nuclear / firm power	Longer-term strategic value	24/7 load wants reliable baseload, not just cheap energy.
Cooling and data-center infrastructure	Higher demand for thermal management	Power density raises the value of cooling efficiency.

The stock market usually tries to compress a new theme into a narrow basket. That is too small here. If the IEA is right, the real trade extends into transmission, substations, gas, grid software, and equipment, not just into the obvious utility ETFs.

In other words: this is not merely a power story for utilities. It is an industrial policy story for the whole grid stack.