Why Can't the Power Grid Keep Up With AI Data Centers?
AI looks like a race to build better chips. But as AI gets bigger, an older technology is becoming just as important: electricity.
AI needs computers.
Those computers use powerful chips from companies such as Nvidia.
Thousands of those chips are packed into data centers.
But building the data center is only part of the problem.
You also have to power it.
And AI is beginning to need an extraordinary amount of electricity.
According to the International Energy Agency, data centers consumed about 485 TWh of electricity worldwide in 2025.
By 2030, that could rise to roughly:
950 TWh
Nearly double today's level.
Electricity consumption from AI-focused data centers alone grew 50% in 2025.
But the biggest problem isn't simply producing more electricity.
It's getting enough electricity to the right place, at the right time.
A Data Center Can't Simply Plug Into the Grid
Electricity doesn't travel directly from a power plant to an AI server.
It moves through a huge physical system:
Power plant
↓
Transmission lines
↓
Substations
↓
Transformers and distribution equipment
↓
Data center
Think of the power grid as a giant road network.
A power plant may produce the electricity, but transmission lines and substations are the highways and intersections that move it.
Building more cars doesn't solve anything if there aren't enough roads.
Electricity works in a similar way.
The IEA estimates that grid constraints could put around 20% of planned global data-center capacity through 2030 at risk of connection delays.
That's a surprisingly large number.
So why can't we simply build the grid faster?
AI Moves Fast. Power Infrastructure Doesn't.
A new generation of AI chips can arrive quickly.
A data center can also be developed relatively quickly.
The power grid operates on a very different clock.
Building new transmission lines requires land.
Permits.
Engineering.
Substations.
Transformers.
Cables.
Community approvals.
And years of construction.
In advanced economies, the IEA says new transmission lines can take four to eight years to build.
Wait times for critical equipment such as transformers and cables have roughly doubled over the past three years.
Four years is an eternity in AI.
Several generations of chips can arrive while a power project is still working its way toward completion.
That creates a strange mismatch:
Digital infrastructure is moving at software speed.
Electricity infrastructure is still moving at infrastructure speed.
And that gap is becoming a bottleneck.
One AI Rack Could Use as Much Power as 65 Homes
The problem is also happening inside the data center.
An advanced server rack is roughly the size of a large refrigerator.
But the IEA estimates that by 2027, one advanced AI server rack could have peak power demand comparable to:
65 households.
Between 2020 and 2025, the power density of AI servers increased 11-fold.
By 2027, the IEA expects it to increase roughly another four times.
Packing that much computing power into a small space creates another problem.
Heat.
The electricity going into AI chips eventually produces heat that has to be removed.
So an AI data center doesn't only need chips.
It needs electrical equipment.
Power-management systems.
Backup power.
And increasingly sophisticated cooling.
AI infrastructure is becoming an energy infrastructure business too.
Data Centers Are Starting to Follow the Electricity
Traditionally, many data centers were built near large cities and internet users.
Being close helped reduce latency.
But large AI facilities have another priority:
Where can we get power?
That question is beginning to change where data centers are built.
According to JLL analysis reported by Reuters, newer European AI-focused hyperscale data-center projects are being built an average of about 175 kilometers from major cities.
The average for projects built in recent years was around 46 kilometers.
Developers are moving farther away partly in search of cheaper land and faster access to electricity.
That's an important shift.
For some of the biggest AI infrastructure projects, proximity to abundant power can matter more than proximity to people.
Who Actually Builds the Power Behind AI?
When people think about the AI boom, they often think about Nvidia.
But a working AI data center requires a much larger industrial supply chain.
Here are some of the companies operating at different parts of it.
GE Vernova — Generation and the Grid
GE Vernova makes equipment used to generate electricity and move it through the grid.
That includes gas turbines as well as grid infrastructure.
If AI increases electricity demand, both sides of that business become relevant.
Siemens Energy — Turbines and Power Infrastructure
Siemens Energy supplies gas turbines and other energy infrastructure.
Demand from data centers has become one of several forces increasing demand for new power-generation equipment.
Eaton — Electrical Infrastructure
Eaton makes electrical equipment used to manage and distribute power.
Producing electricity isn't enough.
Data centers need equipment that can safely deliver enormous amounts of it to servers.
Vertiv — Power and Cooling
Vertiv supplies power-management and thermal-management infrastructure for data centers.
As AI chips consume more electricity, removing the resulting heat becomes increasingly important.
Generac — On-Site and Backup Power
Generac is widely known for backup generators.
But data centers have become a major growth market.
Generac reported that its backlog of products serving the data-center market reached approximately:
$1.6 billion
in July 2026.
Bloom Energy — Power Without Waiting for the Grid
Bloom Energy uses fuel-cell systems that can generate electricity near a data center itself.
That represents another response to the same problem:
If connecting to the grid takes years, can the data center bring some of its own power?
This Is Not a List of Stocks to Buy
These companies are not here because their share prices are guaranteed to rise.
They are useful because they reveal the physical supply chain behind AI.
Look at AI this way:
AI chips
↓
Data centers
↓
Power generation
↓
Grid and transformers
↓
Power management
↓
Cooling
↓
On-site power
Suddenly, AI doesn't look like only a semiconductor or software industry.
It touches turbines, transformers, generators, cooling systems and electricity networks.
The more AI becomes physical infrastructure, the larger that supply chain becomes.
Why Not Just Build Power Plants Next to the Data Centers?
Some developers are trying versions of exactly that.
On-site generators, fuel cells, batteries and dedicated power plants can reduce dependence on slow grid connections.
But this doesn't make the problem disappear.
Power equipment still has to be built.
Fuel has to be supplied.
Permits may be required.
Environmental rules apply.
And local communities increasingly care about who pays for the new infrastructure.
The issue has become serious enough that PJM, the largest U.S. grid operator, recently proposed a system that could require some data centers to switch to backup power during grid emergencies.
PJM's system serves about 67 million people.
That shows how far the issue has moved beyond Silicon Valley.
AI infrastructure can now affect utilities, regulators, local communities and potentially household electricity bills.
The Next AI Bottleneck
For the past few years, the AI race has focused heavily on chips.
Who can get the most GPUs?
Who can build the biggest models?
Who can build the biggest data centers?
But as AI infrastructure expands, the competition is moving outside the computer.
AI needs electricity.
Electricity needs transmission lines.
Transmission lines need substations.
Data centers need transformers.
Servers need cooling.
And none of those things can be copied instantly like software.
The IEA already estimates that grid constraints could delay around one-fifth of planned global data-center capacity through 2030.
So one of the most important questions in the next phase of AI may not simply be:
Who can build the smartest AI?
It may also be:
Who can get enough electricity to run it?
The next bottleneck in AI may not be intelligence.
It may simply be power.
Sources
International Energy Agency — Key Questions on Energy and AI
https://www.iea.org/reports/key-questions-on-energy-and-ai
Reuters — Europe AI Data Centres Seek Cheaper, Quicker Energy and Land
https://www.reuters.com/business/europe-ai-data-centres-seek-cheaper-quicker-energy-land-2026-08-19/
Generac — Second Quarter 2026 Results
https://investors.generac.com/news-releases/news-release-details/generac-reports-second-quarter-2026-results
BEYOND THE OBVIOUS.