I recently listened to NVIDIA CEO Jensen Huang and Elon Musk talk about where artificial intelligence is going.
Their message was pretty amazing:
We may still be at the very beginning of the AI revolution.
And one reason is something most of us don't think about — machines are beginning to become users of computers too.
Today, Most Internet Traffic Starts With Us
Think about how you use technology.
You send an email. You search Google. You watch a video. You ask AI a question.
Then you go to lunch, watch TV or go to sleep.
The computer waits for you.
Artificial intelligence changes this completely.
Imagine eventually having 100 AI assistants working for you. One researches investments. Another schedules appointments. Another reads documents. Another checks information. Others communicate with other AI assistants.
They could potentially work 24 hours a day, seven days a week.
Every time an AI thinks through a problem, searches for information, creates something, checks its work or communicates with another computer, it uses computing power.
That is where the enormous new demand could come from.
Now Imagine Millions of Self-Driving Cars
A self-driving car isn't simply driving.
Its cameras and sensors are constantly watching the world around it. Its computers have to recognize cars, people, roads, traffic lights and unexpected situations — and then decide what to do next.
Now imagine one car encounters something unusual, such as a construction worker giving a confusing hand signal.
That experience can become valuable training information.
Powerful computers can analyze what happened and use it to improve the AI. An improved version of that software could eventually be distributed across the fleet.
One car encounters the problem.
The system learns from the experience.
Millions of cars can eventually benefit from what was learned.
Robots Could Do the Same Thing
Now take that idea and apply it to robots.
Suppose a robot in a factory is learning how to pick up an oddly shaped box.
It tries.
It fails.
It tries again.
Eventually, the system figures out a better way.
With humans, teaching one worker doesn't automatically teach every other worker.
AI can be different.
Thousands of robots could gather experience. That information can be used to train a better AI system. Improvements could then be sent back to thousands—or eventually millions—of machines.
In other words:
Robots could learn from robots. Cars could learn from cars. And AI agents could learn and communicate with other AI agents.
That could create an enormous learning network.
This Is Why Computing Demand Could Explode
Today there are roughly 8 billion people on Earth.
But imagine a future with not only billions of people using computers, but also enormous numbers of AI agents, robots, autonomous vehicles, smart factories and other intelligent machines using computers alongside us.
And machines can create much more computer activity than people.
A person might ask an AI 20 questions during the day.
An AI agent might perform hundreds or thousands of individual computing steps to complete just one assignment.
A robot could process information from cameras and sensors continuously.
A self-driving vehicle could make decisions every fraction of a second.
And millions of machines could communicate and learn from one another.
This doesn't mean Internet traffic or the number of computer chips will literally increase exactly 1,000 times.
The important point is that the amount of computation being performed could become dramatically larger than it is today.
Think of It This Way
The first Internet primarily connected:
People → Information
The next generation of the Internet may increasingly connect:
People → AI
AI → AI
Cars → AI
Robots → AI
Robots → Robots
Factories → AI
And unlike us, machines can operate 24 hours a day, 365 days a year.
What Does That Mean?
If this happens, we won't simply need better AI software.
We may need an enormous physical infrastructure behind it:
More chips. More data centers. More electricity. More power plants. More networking. More fiber. More cooling. More memory. More factories. And eventually, millions of robots.
Jensen Huang describes these enormous data centers as “AI factories.”
A traditional factory turns raw materials into products.
An AI factory takes electricity and computing power and turns them into something very different:
intelligence.
Could AI Surprise the Entire Economy to the Upside?
This may be the most exciting part of the story.
Most economic forecasts assume that economies grow relatively slowly because there are only so many workers, factories and hours in the day.
AI could begin to change that equation.
Imagine a doctor with an AI assistant that can analyze information in seconds. An engineer who can design something in one day instead of one week. A small business with 20 AI assistants working around the clock. A factory where robots continuously become more efficient. Or scientists using AI to discover medicines and new materials much faster.
That means the same number of people could potentially produce much more.
Economists call this productivity, and over long periods of time, productivity is one of the biggest reasons living standards improve.
AI could also create an enormous investment boom along the way. Building the AI economy requires data centers, electrical grids, power generation, semiconductor factories, networking equipment, construction workers, engineers and many other industries.
So we could potentially get two economic benefits at the same time:
First, trillions of dollars could be invested building the infrastructure.
Second, once that infrastructure is operating, AI could help millions of workers and businesses become more productive.
That is why the economic results could eventually surprise the world to the upside.
Nobody knows how large the benefit will be, and there will certainly be disruptions and mistakes along the way.
But imagine if artificial intelligence eventually allowed the average worker or business to produce even 20%, 30% or 50% more with the same amount of time.
Multiply that across millions of workers and businesses around the world and the numbers become enormous.
We May Still Be at the Beginning
We tend to look at the hundreds of billions of dollars already being spent on artificial intelligence and wonder:
“Haven't we already spent too much?”
Jensen Huang's argument is almost the exact opposite.
If billions of AI agents, robots, vehicles and intelligent machines eventually become computer users themselves, then today's infrastructure may only be the foundation for what comes next.
The Internet gave billions of people access to information.
Artificial intelligence could eventually give billions of machines the ability to reason, communicate, learn and work.
If that happens, the demand for computing could be unlike anything we've experienced before.
And more importantly, all of that new intelligence could allow the world's economy to produce things faster, cheaper and better than we ever thought possible.
Maybe the biggest surprise of the AI revolution won't be how much technology we have to build.
Maybe it will be how much bigger the economy can become because we built it.