Empires Once Marched on Roads – AI marches on extension cords

Enrique

Jun 12, 2026

Part I : The Tent Economy

For most of the industrial age, infrastructure was built with the assumption that it would outlive the people who built it.

A railroad bridge might remain in service for a century. A hydroelectric dam could power cities for generations. Even the first wave of hyperscale data centers were designed as long-lived assets—concrete, steel, fiber, and redundancy engineered to survive decades of technological change.

The AI boom has quietly inverted that equation.

Meta’s rapid-deployment structures in Ohio are not merely an engineering curiosity. They represent a different philosophy of capital investment. Instead of building permanent infrastructure and upgrading the equipment inside, the industry is increasingly treating the building itself as disposable.

The reason is simple: the chips are aging faster than the concrete.

A traditional data center might take a year or more to permit, design, and construct. By the time the ribbon is cut, an AI company may already be planning its next generation of accelerators. In a market where every month matters, speed becomes more valuable than permanence.

The result is something that would have seemed absurd only a few years ago:

Billions of dollars of cutting-edge silicon operating inside structures that resemble military field camps.

Historically, the pattern looked like this:

Today, it increasingly looks like this:

The infrastructure no longer anchors the investment.

The investment anchors the infrastructure.

This is why the comparison to a Roman legion camp feels surprisingly appropriate. Roman armies didn’t stop to build marble cities every time they advanced. They erected temporary camps, secured the perimeter, completed the mission, and moved on.

Meta appears to be approaching AI infrastructure the same way.

Not because it lacks resources.

Not because it lacks confidence.

But because the future is arriving faster than permanent buildings can be completed.

The irony is hard to miss.

The most advanced computing systems in human history are increasingly being housed in structures whose defining feature is how quickly they can be assembled.

The age of cathedrals may be giving way to the age of tents.

Part II : The Sweet Spot

The most interesting possibility is that Meta isn’t building these structures despite their limitations.

They may be building them because of their limitations.

Traditional infrastructure is optimized for permanence.

AI infrastructure appears to be optimized for timing.

For decades, the industrial equation looked something like this:

The building was expected to outlive several generations of equipment.

If the servers became obsolete, you swapped the servers.

If the networking improved, you upgraded the networking.

The concrete stayed where it was.

AI changes the equation because the compute itself is now depreciating at extraordinary speed.

The cluster you install today may be dramatically less competitive in three years.

The cluster you install in five years may make today’s hardware look quaint.

The result is a strange race between three curves:

Curve 1: Equipment Value

The GPUs begin losing strategic value almost immediately.

Not because they stop working.

Because something faster appears.

Curve 2: Maintenance Cost

Every year the facility remains in operation, maintenance pressure increases.

Fabric ages.

Seals degrade.

Cooling systems accumulate wear.

Power equipment demands attention.

The environment begins collecting interest on every shortcut taken during rapid deployment.

Curve 3: Time-To-Market Advantage

This is the curve Meta appears to care about most.

The first months of operation may generate more strategic value than years of future operation.

Every week saved during construction is another week training models.

Another week attracting customers.

Another week gathering data.

Another week ahead of competitors.

The magic happens where those three curves intersect.

Not at year twenty.

Not at year thirty.

But near the beginning.

The goal may not be to build the perfect facility.

The goal may be to reach the peak of the value curve before depreciation and maintenance begin eating away at the advantage.

Viewed this way, the tents start making sense.

Meta may be treating AI infrastructure the way Formula 1 teams treat race cars.

Nobody expects the car to last fifty years.

The objective is to win the season.

After that, everything gets rebuilt.

This is why the discussion around durability may actually miss the point.

The question isn’t:

“Will these structures still be operating in 2050?”

The question is:

“Will they have generated enough value by 2028 that nobody cares?”

That is a very different investment philosophy.

A railroad is built for generations.

A factory is built for decades.

An AI tent city may be built for a window.

And in a market where competitive advantage is measured in months rather than years, that window may be all that matters.

Part III : The Roman Frontier

The more one studies these AI tent cities, the less they resemble modern infrastructure and the more they resemble something much older.

A Roman legion marching into unfamiliar territory did not begin by constructing marble temples.

It built a camp.

The objective was not permanence.

The objective was presence.

Get there first.

Establish control.

Secure supply lines.

Expand influence.

Move again tomorrow if necessary.

Meta’s rapid-deployment structures feel remarkably similar.

The company is not waiting for the perfect building.

It is establishing a foothold on the AI frontier.

Viewed through that lens, the tents stop looking temporary and start looking strategic.

The generators become supply depots.

The fiber becomes roads.

The GPUs become the modern equivalent of legionaries.

And the data center itself becomes a fort.

Gold rush towns.

Oil fields.

Railroad camps.

Military outposts.

The frontier arrives first.

The permanent structures arrive later.

AI may simply be following the same historical script.

The difference is that the resource being extracted is not gold, oil, or territory.

It is capability.

Every new cluster increases the amount of intelligence available to the organization operating it.

Every GPU brought online expands the frontier.

Every month saved in construction becomes another month of training.

Another month of experimentation.

Another month ahead of competitors.

This helps explain why so many decisions that appear irrational from a traditional infrastructure perspective suddenly make sense.

Roman generals did not ask:

“Will this camp still be here in fifty years?”

They asked:

“Will this camp help us win tomorrow?”

Meta appears to be asking the same question.

Not because permanence no longer matters.

Because the frontier is moving too quickly to wait for permanence.

The irony is that the AI industry often speaks in the language of the future:

Artificial General Intelligence.

Superintelligence.

Digital civilization.

The next stage of humanity.

Yet the physical infrastructure supporting those ambitions increasingly resembles one of humanity’s oldest patterns:

A frontier camp assembled in haste at the edge of the known world.

The AI race may not look like the future after all.

It may look exactly like every frontier that came before it.

The only difference is that instead of horses, roads, and grain, the empire now marches on GPUs, generators, and extension cords.