Most AI talk right now is about a bubble. The worry is too much cash chasing too many data centers.
One big investor sees the opposite risk.
Raj Agrawal runs real assets at KKR, the group that invests in power and data centers. He thinks the market is guessing too low on how much power AI will need.
The Real Limit Isn't Chips. It's Power.
For two years, AI was all about chips like Nvidia's GPUs. The race was about who could grab the most.
Agrawal flips that view. The hard limit may be the power to run all those chips.
He told Bloomberg that AI is growing too fast for today's power plans. And power is slow to build.
A new plant can take years. A new power line can take just as long.
Demand can spike in months. Supply takes years to catch up.
That gap is the whole problem.
Every morning, Market Briefs breaks down where smart money is moving in five minutes. You also get a free investing masterclass when you join.
Why The Power Math Keeps Growing
Even the safe forecasts are huge. The International Energy Agency says data center power use will more than double by 2030.
AI is the main reason. Its chips run all day and all night, and they pull a ton of power.
A single big AI data center can use as much power as a small city. Now picture hundreds of them.
Cooling those chips takes power too. Data centers run hot, and heat is the enemy.
Agrawal's point is that it could go higher still. The demand keeps getting bumped up, not down.
Think of a town that doubles overnight. The houses go up fast, but the pipes and wires can't keep up.
KKR is backing this with real money. It owns data center stakes in the US, the Middle East, and Asia.
The US and China will drive most of that growth, the IEA says. In the US alone, data center power use could more than double by 2030.
Where The Power Comes From
Adding power is slow for a reason. New plants and lines need permits, land, and years of work.
So old sources are back in style. Gas and nuclear can run around the clock, which is just what data centers need.
Solar and wind help, but they don't run all the time. Backup still has to come from somewhere.
This is the same bet behind AI's power problem trade. More investors keep circling it.
The firms that deliver steady power fastest will hold the upper hand.
What To Watch
The question isn't whether AI needs power. It's who gets paid to supply it.
Look at utilities, gas, nuclear, grid gear, and the firms that build and cool data centers.
Power is the slow part of the build-out. So watch how fast new supply comes online.
The chips get the headlines. The power bill sets the limit.
If you want this kind of read every weekday, join 350,000+ readers of Market Briefs and get a 45-minute investing course as a bonus.
