America is spending staggering amounts of money on artificial intelligence. That does not mean America is rebuilding its industrial base.
That distinction is becoming increasingly important.
The United States is experiencing a genuine AI investment boom. Hyperscalers have increased R&D and capital investment roughly 50-fold in two decades, from about $15 billion in 2005 to $750 billion in 2025. By the end of 2026, the figure could approach $1 trillion.
Those numbers sound like an industrial revolution.
They aren’t.
The uncomfortable reality is that America is pouring capital into computation while failing to generate a comparable surge in the physical assets that make an economy more productive: factories, machinery, infrastructure, energy systems, and industrial capacity.
Productive investment as a share of U.S. GDP has barely moved.
That is the paradox at the heart of America’s AI boom: the country is becoming extraordinarily good at financing intelligence without becoming proportionally better at building things.
And that matters.
Investment determines where future production, jobs, productivity, and strategic power will reside. Since the global financial crisis, the United States has generally outperformed other advanced economies on investment. But China operates on an entirely different scale. It adds roughly $4.4 trillion in net productive assets every year—around four times the comparable U.S. amount.
America has the world's most powerful technology companies.
China is building the world's industrial infrastructure.
Those are not the same achievement.
The missing factories
If the United States wants to reduce its dependence on foreign manufacturing, the scale of the challenge is enormous. Addressing its most critical import dependencies could require roughly $2 trillion in additional manufacturing investment—about 6% of GDP.
Yet the physical investment needed to make that happen is moving in the wrong direction.
Investment in factory structures fell 6% in 2025 after peaking in 2024. General industrial equipment investment was essentially flat, despite a modest improvement in machinery and equipment spending early this year.
The much-discussed reshoring wave that began in 2022 has therefore produced something of a statistical plateau.
Announcements are not factories.
Groundbreakings are not production.
And press releases are not supply chains.
There is inevitably a lag between announcing a semiconductor plant, pharmaceutical facility, battery factory, or other industrial project and actually producing something at scale. But the broader problem is more fundamental: America remains an extraordinarily expensive place to build.
America has a cost problem
The United States cannot simply declare an industrial renaissance into existence.
Across construction, labor, materials, equipment, and time-to-market, American projects routinely carry enormous cost disadvantages.
Building semiconductors in the United States can cost roughly 40% more than in the most competitive locations. Pharmaceuticals can be around 60% more expensive. Developing a new antibody medicine can cost roughly 2.7 times as much as doing so in China.
And subsidies do not magically eliminate the underlying problem. They merely transfer part of the bill to taxpayers.
Two structural disadvantages account for much of the gap.
The first is capital delivery.
American construction is expensive and slow. Construction costs for semiconductor facilities can be roughly twice those in Asia, while timelines can also be twice as long. Nuclear power offers an even more brutal illustration: projects in the United States can take a decade or more, while comparable Chinese projects can be completed in substantially less time.
The second problem is labor.
American industrial labor can cost two to five times as much as labor in China or Taiwan. That might be acceptable if American workers were dramatically more productive.
Increasingly, they aren't.
In advanced semiconductor fabs, Taiwanese engineers can reportedly produce about a quarter more per worker than their U.S. counterparts, even as U.S. wages are more than 2.7 times higher.
That is not a wage problem.
It is a productivity problem.
And it is ultimately a systems problem.
AI won't save an inefficient factory by itself
This is where the AI story becomes particularly interesting.
The same technology driving America's extraordinary investment boom could help solve some of the industrial problems that currently make American production uncompetitive.
But only if companies use AI as an operating technology rather than treating it as a financial asset class.
Modular construction and off-site manufacturing can dramatically shorten project timelines and reduce capital costs. Robotics can reduce dependence on expensive labor. AI can improve planning, maintenance, engineering, quality control, scheduling, and production processes.
Done properly, these technologies could eliminate perhaps half to two-thirds of America's industrial cost disadvantage.
But there is a catch.
You cannot automate your way out of every structural disadvantage.
If permitting takes too long, capital remains tied up. If construction costs are excessive, robotics won't solve the entire problem. If electricity is unreliable or expensive, AI infrastructure becomes more expensive. If supply chains remain dependent on foreign inputs, a domestic factory may still be strategically vulnerable.
AI is an amplifier.
It is not a substitute for competent industrial execution.
America doesn't need to make everything
There is also a danger in responding to China's industrial scale with indiscriminate protectionism.
The United States cannot—and should not—try to manufacture everything domestically.
The economics simply do not work.
Instead, policymakers should focus on the industries where dependence creates genuine strategic vulnerability: products that are critical to national security, highly concentrated among foreign suppliers, or sourced from geopolitically distant partners.
Roughly 25% of U.S. manufactured imports fall into this category.
That should be the starting point.
The objective should not be autarky.
It should be resilience.
America does not need to win every manufacturing contest. It needs to make sure that losing one does not become a national-security crisis.
That requires selective trade policy, targeted financial support, industrial incentives, infrastructure investment, faster permitting, and—perhaps most importantly—an acceptance that strategic redundancy sometimes costs more than pure economic efficiency.
The harder question
The United States has already demonstrated that it can mobilize capital at extraordinary speed.
AI proves that.
What remains unproven is whether it can mobilize physical capital with the same urgency.
That means building factories faster.
Approving infrastructure faster.
Training industrial workers faster.
Deploying energy faster.
Installing machinery faster.
And accepting that some strategically important production will be more expensive in America than it would be elsewhere.
This is the uncomfortable trade-off.
For decades, the United States optimized for efficiency, low prices, financial returns, and global specialization. Now it increasingly wants resilience, domestic capacity, and geopolitical independence.
Those objectives are not free.
America can have cheap goods, maximal efficiency, and minimal redundancy—or it can pay to maintain industrial capacity that might otherwise disappear overseas.
It cannot indefinitely pretend it can have all of them at once.
The AI boom has demonstrated that America still knows how to create enormous pools of capital around a new technology.
The industrial challenge is different.
Can America turn that capital into steel, factories, machines, power plants, supply chains, and skilled workers—or will its AI revolution remain largely confined to servers, software, and balance sheets?
That is the real test.
Because an economy does not become industrially powerful by becoming better at thinking about production.
It becomes powerful by becoming better at producing.