The infrastructure supporting advanced AI is becoming extraordinarily expensive.

Data centers require enormous amounts of electricity, land, semiconductors, cooling systems, networking equipment, and capital.

That creates a potential problem.

AI may give small teams unprecedented intellectual leverage while simultaneously requiring infrastructure that only the world’s largest companies and governments can afford.

If access to that infrastructure becomes closed, scarce, or controlled by a small number of institutions, we risk undermining one of the most exciting promises of AI.

Innovation could become bottlenecked not by ideas or talent, but by permission and access.

Imagine if the early internet had worked that way.

Imagine needing approval from a handful of telecommunications companies before launching a website.

Imagine computing power being reserved primarily for the largest corporations.

Imagine entrepreneurs being told that they could experiment—but only after the established institutions had taken what they needed first.

The internet probably would have developed very differently.

We should be careful not to build that version of AI.

Innovation Wants to Decentralize. Infrastructure Wants to Centralize.

This is the central tension.

AI makes small teams dramatically more capable.

But the infrastructure underneath AI rewards enormous scale.

Innovation wants to decentralize. Infrastructure wants to centralize.

Those two forces are going to collide repeatedly over the next decade.

The solution doesn’t necessarily require every entrepreneur to own computing infrastructure.

Quite the opposite.

A small company shouldn’t need to own a data center any more than it needs to own a power plant or a telecommunications network.

It needs reliable, affordable, competitive access to the output.

That distinction matters.

We don’t need millions of people building data centers.

We need millions of people capable of accessing what those data centers produce.

Access May Be as Important as Capability

Much of today’s AI conversation focuses on who will build the most powerful model.

That is obviously important.

But there is another question that may matter just as much:

Who gets to use it?

If extraordinary AI capability exists but is available primarily to governments and a handful of trillion-dollar companies, its economic impact could look very different from a world where millions of entrepreneurs, researchers, students, small businesses, and independent developers have access to comparable tools.

The history of the internet suggests that innovation often comes from places we don’t expect.

That’s precisely why broad access matters.

We don’t know who will create the next transformative company.

We don’t know which university student has the next breakthrough idea.

We don’t know which small laboratory will discover something important.

We don’t know which five-person company could reinvent an industry.

The point of open infrastructure isn’t knowing who the winners will be.

It’s making sure they have the opportunity to show up.

Capital Becomes the Bottleneck

If labor becomes dramatically more productive, another resource eventually becomes the constraint.

Increasingly, that resource may be capital—and the infrastructure capital provides access to.

A five-person company may soon possess the intellectual and operational capacity that once required 50 or 100 employees.

But those five people may still need enormous amounts of computing capacity to turn their ideas into reality.

That creates an opportunity not just for technology companies, but for investors, capital markets, utilities, infrastructure providers, and policymakers.

The financial system will have to figure out how to finance a world in which tremendous economic output can come from surprisingly small organizations.

What happens when a ten-person company can produce $100 million of economic output?

How should it be valued?

How much capital should it receive?

How should lenders evaluate it?

And perhaps most importantly, how do we ensure that access to the fundamental infrastructure of this new economy isn’t limited only to organizations that are already enormous?

The Return of the Owner-Operator

Technology may unintentionally bring us back toward a very old model of capitalism: the owner-operated business.

For much of the twentieth century, economic scale often required managerial scale. Businesses became enormous organizations because coordinating thousands of people was necessary to produce enormous amounts of output.

AI may loosen that connection.

The company of the future could be surprisingly small.

A founder.

A handful of highly skilled employees.

Specialized outside partners.

AI systems performing much of the repetitive intellectual labor.

And enormous computing infrastructure accessed on demand rather than owned outright.

That structure could allow founders and employees to retain significantly more ownership of what they create.

Instead of building a 2,000-person organization to produce a billion-dollar company, perhaps you build a 75-person organization.

But that future depends on those 75 people having access to the same fundamental technological infrastructure as the 75,000-person incumbent.

Without that access, AI could reinforce today’s largest institutions rather than challenge them.

The Infrastructure Is the Opportunity

The internet demonstrated what can happen when enormously expensive infrastructure becomes broadly accessible.

AI gives us the opportunity to do it again.

We should want massive investment in data centers, energy generation, semiconductors, networks, and AI models.

But the ultimate measure of success shouldn’t simply be how much computing capacity we build.

It should be how many people can build something with it.

The most interesting AI company of 2035 may not exist today.

Its founders may still be in school.

They may be working inside another company.

They may be sitting somewhere with an idea nobody else believes in yet.

Our job isn’t to predict who they are.

It’s to make sure they can plug in.

Because the great economic lesson of the internet wasn’t simply that connectivity was powerful.

It was that broadly distributed access to powerful infrastructure unleashed innovation from everywhere.

AI could do the same.

But only if we build it that way.