AI may be running into a surprisingly old-fashioned problem: electricity.
Building more AI models requires more GPUs.
More GPUs require more data centers.
And more data centers require enormous amounts of:
- electricity;
- transformers;
- transmission capacity;
- cooling;
- grid connections.
The AI race is therefore becoming a power-infrastructure race.
The key question is:
Can electricity supply expand quickly enough to keep up with AI demand?
Educational research only. This article is not investment advice.
Why AI Needs So Much Power
Modern AI systems require large clusters of high-performance chips.
Those chips consume significant electricity.
They also generate heat, which means data centers need additional power for:
cooling, networking and backup systems.
As AI moves from model training into widespread inference, total computing demand could continue increasing.
Reuters reports that AI workloads could account for around half of global data-center capacity by 2030.
That turns electricity availability into a major constraint.
The Grid Cannot Expand Overnight
A new data center can sometimes be built faster than the power infrastructure needed to support it.
Reuters reports that hyperscalers may want facilities completed within months, while grid connections can take:
up to 24 months in some emerging markets
and:
more than eight years in some developed markets.
That gap creates a bottleneck.
A company may have:
- land;
- financing;
- GPUs;
- customers.
But without enough electricity, the project cannot operate at full capacity.
Data Centers Are Moving Toward Power
This is already changing where AI infrastructure gets built.
Developers in Europe are increasingly choosing sites farther away from traditional technology hubs because land and electricity are easier to secure.
The average new European data-center site planned for 2026–2028 is about 175 kilometres from a major city, compared with just 46 kilometres for projects delivered between 2022 and 2025.
The logic is simple:
Data centers are moving toward available power rather than waiting for power to come to them.
The Scale of Investment Is Enormous
The world’s largest cloud companies are spending aggressively.
JLL estimates the four biggest hyperscale cloud providers will spend about $725 billion in 2026, up from roughly $410 billion in 2025.
McKinsey expects global data-center investment to approach $7 trillion by 2030.
That creates opportunities for:
- utilities;
- grid equipment;
- transformers;
- power generation;
- cooling systems;
- electrical infrastructure.
But it also creates risk.
Why Power Can Become an AI Risk
Electricity shortages can affect AI economics in several ways.
Higher Costs
Scarce grid capacity can make power more expensive.
Construction Delays
Projects may wait years for connections.
Capital Risk
Billions can be committed before facilities are fully operational.
Grid Stress
Very large data centers can create new stability challenges.
In July, more than 3 gigawatts of data-center demand suddenly disconnected from the PJM grid in the United States after a transmission-line failure, creating a widespread voltage disturbance.
That shows how large AI infrastructure is becoming relative to the power system itself.
South Korea Shows the Scale of the Problem
The semiconductor industry is facing similar pressure.
On September 14, Samsung Electronics and SK Hynix rejected a proposal from Korea Electric Power Corp that would have required about $18.7 billion in upfront payments to secure electricity for future semiconductor mega-clusters.
That illustrates how power availability is becoming a major financial issue—not just an engineering issue.
How the TSL Macro Model Fits
TradingSimuLab’s Macro Model helps place the power shortage inside the wider economy.
Important questions include:
Net Score
Is massive infrastructure spending supporting growth or creating inflation pressure?
Confidence
Are power prices, capital spending and industrial demand moving in the same direction?
Scenario Probabilities
Is the economy moving toward stronger investment, higher inflation or tighter financial conditions?
We are not assigning a live TSL Macro score here.
The goal is to organize the forces behind the AI buildout.
Why Risk Simulation Matters
The power crunch also creates investment risk.
TradingSimuLab’s Risk Simulation would ask:
Expected Return
Does the potential growth justify the capital required?
VaR and CVaR
How severe could downside become if projects are delayed?
Max Drawdown
How badly could valuations fall if AI infrastructure spending slows?
Probability of Gain
Across many possible outcomes, how often does the investment still finish positive?
The central issue is:
High demand does not eliminate execution risk.
What Could Ease the Power Crunch?
The bottleneck could improve through:
- faster grid expansion;
- new gas generation;
- nuclear power;
- renewable energy;
- battery storage;
- better chip efficiency;
- more efficient cooling;
- locating data centers near existing power.
Some of those solutions take years.
Others can scale faster.
That timing difference will matter.
Final Takeaway
The AI boom is no longer just about better chips.
It is increasingly about whether the physical infrastructure can support them.
The chain is:
More AI → More Compute → More Data Centers → More Electricity → More Grid Pressure
The key question is not simply:
“How much AI demand exists?”
It is:
“Can power infrastructure expand fast enough to serve it economically?”
That may become one of the most important constraints on the next phase of the AI boom.
For more market research tools, macro analysis and risk simulations, sign up to TradingSimuLab and explore the platform.