Everyone is arguing about one question.
Will there be enough demand for AI?
I think there is a different question.
Can we even build enough infrastructure if the demand is real?
I came across someone doing some rough math on AI demand. The numbers are obviously estimates, but the idea got me thinking.
Start with people.
Assume 1 billion people use AI every day.
They are not all online at the same time, so maybe around 350 million are active during any given period.
Add another 50 million power users running AI agents all day.
Now you’re around 400 million active human users.
Humans probably aren’t the biggest market anyway.
Think about self driving cars.
Robots.
Factory automation.
AI assistants running inside every computer.
Smart cameras.
Medical devices.
Industrial machines.
Those systems don’t sleep.
They don’t use AI for a few minutes a day.
Many of them would be running all day.
Even using conservative assumptions, you could end up with 2 to 3 billion machines constantly asking for AI inference.
Now the numbers get crazy.
That is roughly 3 billion active AI users, human and machine combined.
Whether the exact number is right doesn’t matter.
The point is the demand could become much bigger than what humans alone create.
Then reality shows up.
Can we actually build enough GPUs?
Can we generate enough electricity?
Can we build enough transmission lines?
Can we expand enough substations?
Can we build enough semiconductor fabs?
Can we produce enough advanced packaging?
Can ASML make enough EUV machines?
Then there are the raw materials.
Copper.
Gallium.
Germanium.
Tantalum.
Indium.
Rare earth processing.
Every one of those has become part of the AI supply chain.
Every one of them has limits.
Even if every company wants to build faster, physics doesn’t care.
Neither do permitting delays.
Neither does the power grid.
Then there is another question almost nobody asks.
Who pays for all this?
If AI replaces millions of jobs faster than it creates new ones, where does future demand come from?
People without income don’t buy software subscriptions.
They don’t buy robots.
They don’t buy new products.
Money has to keep moving through the economy.
That doesn’t mean AI fails.
It means the business model becomes much harder if productivity rises while consumers lose purchasing power.
Maybe AI becomes the biggest technology revolution in history.
I think it probably will.
But history also shows that every industrial revolution eventually runs into real world limits.
Not because people stop believing.
Because reality gets a vote.
Power plants take years.
Transmission lines take years.
Data centers take years.
Chip fabs take years.
Mining projects take years.
You can’t compress all of that into one earnings cycle.
That is why I keep wondering if the next AI slowdown won’t come from weak demand.
It may come because the world simply can’t build fast enough.
Or because we build faster than the economy can afford.
Either way, investors should stop thinking AI is only a software story.
It is becoming the biggest infrastructure project the world has ever tried to build.