THE WAR MACHINE IS GETTING CHEAPER—AND SMARTER

THE WAR MACHINE IS GETTING CHEAPER—AND SMARTER

Why drones, AI, and compute are dismantling the military advantages of the past

For most of the modern era, military power followed a familiar equation: bigger platforms, better weapons, superior electronics, more expensive machines.

The United States became exceptionally good at that equation.

Stealth aircraft. Aircraft carriers. Precision-guided missiles. Satellites. Nuclear submarines. Global logistics. Networked command systems.

The result was a military machine so technologically advanced that many potential adversaries were effectively playing a different game.

But the game is changing.

The most consequential weapons of the next war may not cost hundreds of millions of dollars. They may cost thousands. They may not require a pilot. They may not need GPS. They may not even need a human operator making decisions one machine at a time.

They may arrive in swarms.

And behind those swarms will be something even more important than the hardware: artificial intelligence, data and computing power.

That is the uncomfortable message emerging from the analysis of Paul Scharre in the July/August 2026 issue of Foreign Affairs. The United States still possesses the world's most powerful military. But the technological advantage that once made that dominance relatively predictable is becoming harder to sustain.

The paradox is extraordinary: the world's most technologically sophisticated military may be entering an era in which sophistication itself becomes a liability.

THE $300,000 PROBLEM

Consider the economics of a drone.

A conventional warship can cost hundreds of millions—or billions—of dollars. A sophisticated combat aircraft can cost tens or hundreds of millions.

Now imagine an adversary attacking that platform with a relatively primitive autonomous vehicle costing a fraction of one percent of its price.

That changes the arithmetic of war.

Ukraine demonstrated the principle dramatically in the Black Sea. Relatively inexpensive unmanned boats and anti-ship missiles helped devastate Russia's Black Sea Fleet. According to Scharre's account, 13 Russian ships were sunk after two years of war, with dozens more damaged. One drone boat costing approximately $300,000 could threaten a warship worth hundreds of millions.

The important innovation is therefore not simply “the drone.”

It is the cost-exchange ratio.

For centuries, militaries competed by building increasingly sophisticated weapons. But when a $10,000 or $100,000 system can destroy or disable a million-dollar or billion-dollar asset, the economics begin to favor whoever can manufacture the cheaper weapon faster.

This is almost the opposite of the Pentagon's traditional industrial model.

American defense procurement has spent decades moving toward what military planners call “exquisite” systems: highly sophisticated, extremely capable, enormously expensive and produced in relatively small quantities.

Drones attack that logic at its foundation.

The new question is not:

How powerful is the weapon?

It is:

How many can you build before your opponent destroys them?

That distinction could determine the outcome of future wars.

THE DRONE FACTORY BEATS THE AIRCRAFT FACTORY

The scale problem is already visible.

Ukraine produces approximately four million drones annually, while the U.S. Army was acquiring roughly 50,000, according to Scharre.

The comparison is almost absurd.

The United States has vastly greater economic resources. Yet its defense-industrial system struggles to manufacture inexpensive autonomous systems at the speed demanded by modern warfare.

Why?

Because the American military procurement system was optimized for something else.

It was designed to identify requirements, establish specifications, solicit bids, test systems, negotiate contracts, manufacture platforms and maintain them for decades.

That works reasonably well for aircraft carriers.

It is disastrously slow for technologies that evolve every few months.

A drone designed today may become obsolete next year. An AI model that represents the state of the art today may be surpassed within months.

The battlefield is becoming a software environment.

The procurement bureaucracy is still behaving like a hardware industry.

That mismatch could be more dangerous than technological inferiority itself.

THE SWARM IS COMING

Today's drone is often little more than an aircraft with a camera, communications link and explosives.

Tomorrow's drone could be something fundamentally different.

Imagine hundreds—or thousands—of autonomous machines operating as a distributed system.

They communicate.

They adapt.

They search.

They identify patterns.

They redistribute tasks when individual machines are destroyed.

They operate despite jamming.

They may navigate without GPS.

And they may coordinate attacks at speeds impossible for human operators to match.

That future is already being prototyped in Ukraine.

Some Ukrainian systems can autonomously navigate toward a target after communications with their human operator are disrupted. Long-range drones have demonstrated the ability to navigate without GPS by comparing camera imagery with previously stored satellite imagery.

Scale those capabilities.

Then remove the assumption that every drone needs a human pilot.

The military operator of the future may not “fly” a drone at all.

The operator may command a swarm.

That is a profound organizational transformation.

Military command has historically been hierarchical: commander, subordinate commander, unit, vehicle, soldier.

Autonomous swarms imply something more decentralized.

The human establishes objectives and constraints.

Machines decide how to distribute themselves.

The battlefield becomes an ecosystem of autonomous agents.

And that raises an uncomfortable question:

What happens to a military organization built around controlling individual platforms when its weapons begin controlling themselves?

THE $4 MILLION INTERCEPTOR TRAP

The economics become even more troubling when defense enters the equation.

The United States and its allies may successfully intercept an incoming drone or missile—but success can become economically meaningless if the interceptor costs vastly more than the weapon it destroys.

Scharre highlights the extreme example of a relatively inexpensive Shahed drone being intercepted by a Patriot missile costing millions of dollars.

That is a victory on the tactical scoreboard.

It can be a defeat on the spreadsheet.

If the attacker can launch ten cheap weapons for the cost of one interceptor, the defender eventually faces a production problem.

The attacker manufactures threats.

The defender manufactures answers.

And if answers are more expensive than threats, the defender is playing the wrong game.

This is why low-cost interceptors matter as much as low-cost drones.

The emerging battlefield may require a new generation of defensive systems whose primary characteristic is not maximum sophistication but economic sustainability.

The objective is no longer to build the perfect shield.

It is to build enough shield.

AI CHANGES THE EQUATION AGAIN

Drones transform the physical economics of war.

AI transforms the cognitive economics.

Modern militaries generate enormous quantities of information: satellite imagery, radar signals, communications, sensor feeds, intelligence reports and battlefield observations.

Humans cannot process all of it.

AI can.

Large language models and other machine-learning systems are already being incorporated into military intelligence and operational planning. Scharre describes the use of AI-enabled systems to synthesize large amounts of battlefield information and assist planners in constructing strike packages in dynamic environments.

But here's the catch:

AI is not a secret weapon anymore.

Commercial technology spreads.

A capability developed by one company can rapidly appear in competitors' systems. Models can be copied, distilled or reproduced using different technical approaches.

Scharre argues that Chinese AI models are approaching leading American systems and discusses “adversarial distillation,” in which competitors extract capabilities from advanced models.

The consequence is strategically important.

The military advantage may no longer belong to the country that invents an AI capability first.

It may belong to the country that deploys it first, integrates it best and scales it fastest.

That is a very different race.

COMPUTE IS THE NEW INDUSTRIAL CAPACITY

There is another layer beneath AI.

Compute.

The next military competition may depend less on who owns the most aircraft and more on who controls enough processors, data centers, energy and networks to operate intelligent systems at scale.

Scharre makes a striking analogy: computing power in the AI era is becoming analogous to manufacturing capacity during the industrial age.

During the Industrial Revolution, factories transformed national power.

In the AI revolution, data centers may play a similar role.

The country with more compute can train more capable models, run more simulations, process more intelligence and deploy more AI agents.

That means semiconductors are no longer merely components inside computers.

They are strategic infrastructure.

This explains the importance of U.S. export controls on advanced chips and semiconductor manufacturing equipment. Washington has attempted to restrict China's access to leading-edge processors and the machinery required to manufacture them.

But hardware restrictions have a fundamental limitation.

You can restrict chips.

You cannot easily restrict ideas.

If an AI capability can be copied, distilled or independently reproduced, the technological lead becomes temporary.

The objective therefore changes from maintaining an unassailable technological monopoly to maintaining a lead in adoption.

That may be the only durable advantage available.

THE REAL ENEMY MAY BE THE PENTAGON

This is where the story becomes less about technology and more about organizational psychology.

Militaries frequently resist technologies that threaten their institutional identity.

The U.S. Navy resisted the transition from sail to steam.

Army leaders debated the proper role of tanks even during World War II.

Different services developed different attitudes toward drones depending on how those systems challenged their organizational cultures.

AI creates an even deeper problem.

It threatens not merely equipment.

It threatens jobs, expertise and identity.

What does it mean to be a pilot when the aircraft flies itself?

What does it mean to be an intelligence analyst when an AI can synthesize thousands of reports?

What does command mean when autonomous systems can make tactical decisions faster than humans?

Technology can be adopted physically while being rejected culturally.

That may be the most dangerous form of technological failure.

The machine is available.

The organization simply refuses to use it properly.

SILICON VALLEY IS NOW PART OF THE DEFENSE INDUSTRY

There was a time when military technology was developed primarily inside military laboratories and traditional defense contractors.

That era is disappearing.

The most important advances in AI are coming from commercial companies.

The Pentagon therefore needs Silicon Valley.

But Silicon Valley does not necessarily need the Pentagon.

That creates a new strategic vulnerability.

Scharre describes growing tensions between the U.S. defense establishment and AI companies over the military use of advanced models, including disagreements surrounding Anthropic and concerns among technology workers about applications such as mass surveillance and fully autonomous weapons.

The lesson is bigger than one corporate dispute.

If engineers increasingly see military work as incompatible with their values, the Pentagon could lose access to precisely the talent it needs most.

The future military advantage may therefore depend on something surprisingly nontechnical:

trust.

The military needs AI companies.

AI companies need engineers.

Engineers need confidence that their work will be used within credible boundaries.

Break that chain, and no amount of defense spending will instantly replace the lost expertise.

STOP COUNTING SHIPS

Perhaps the strangest implication of this new era is that the traditional measures of military power are becoming obsolete.

Navies count ships.

Air forces count aircraft.

Armies count soldiers.

Those numbers still matter.

But they increasingly fail to describe the actual computational nervous system connecting modern military forces.

How much compute does a military possess?

How many AI systems are actually being used?

How much data is accessible?

How many personnel interact with AI every month?

How many autonomous systems can be produced?

How quickly can new models be tested and deployed?

Scharre proposes that the Pentagon begin tracking precisely these kinds of metrics, including computing capacity, model usage, data availability and the number of equivalent high-performance GPUs available to the department.

This is more than bureaucratic reform.

It represents a new definition of military power.

In the twentieth century, industrial capacity translated into tanks, aircraft, ships and missiles.

In the twenty-first, industrial capacity increasingly translates into compute, algorithms, data, autonomy and manufacturing scale.

THE SPANISH ARMADA LESSON

The most unsettling historical analogy comes from 1588.

Spain possessed an enormous and powerful fleet.

England possessed a different technological and tactical approach.

The Spanish Armada was organized around closing with enemy ships and boarding them. English forces exploited cannon technology and fought differently.

Spain had ships.

England had a better way of using technology.

The Armada was defeated, and although the war continued for years, Spain's trajectory as a global power had changed.

That is the real warning for the United States.

It does not need to become technologically backward to lose its advantage.

It only needs to become organizationally slower than its competitors.

The decisive military technology of the future may already exist.

The question is who learns to use it first.

THE NEW ARMS RACE IS SPEED

The twentieth-century military competition was dominated by industrial scale.

The next one may be dominated by adaptation speed.

A military that can deploy a new drone fleet in months may defeat one that requires years.

A military that can integrate a new AI model into operational systems in weeks may outperform one that spends years approving it.

A country that can manufacture millions of autonomous systems may overwhelm an opponent whose weapons are individually superior but numerically scarce.

This produces a strange reversal.

For decades, American military strategy was built around maintaining a technological gap.

The emerging battlefield may make technological gaps much harder to sustain.

Instead, the decisive gap may become the adaptation gap.

And adaptation is not something that can simply be purchased.

It requires changing procurement systems, organizational structures, training, doctrine, industrial capacity and institutional culture.

It requires accepting failure.

Experimenting.

Iterating.

Replacing expensive assumptions with uncomfortable evidence.

Most importantly, it requires recognizing that the battlefield is no longer simply a contest between machines.

It is a contest between systems that learn.

The United States still has extraordinary advantages: economic scale, world-class universities, advanced semiconductor companies, enormous defense resources and a vast technology ecosystem.

But none of those advantages automatically becomes military power.

They become military power only when an organization can transform them into capability.

That may ultimately be the real meaning of the coming AI war.

The winner will not necessarily be the country with the smartest algorithm.

Nor the largest aircraft carrier.

Nor the most sophisticated fighter.

Nor even the most advanced drone.

The winner may be the country that can build a new weapon on Monday, test it on Tuesday, deploy it on Wednesday, discover that it is obsolete on Thursday—and replace it on Friday.

The future of war may belong not to the strongest machine, but to the fastest learning system.

And that is a very different kind of arms race.

Contextual Diagram