Artificial intelligence may very well transform the global economy. It may increase productivity, eliminate enormous amounts of routine work, create industries that do not yet exist, and fundamentally change how businesses operate. None of that, however, guarantees that today’s AI investments will produce tomorrow’s expected returns.
That distinction is becoming increasingly important.
The extraordinary investment boom surrounding artificial intelligence rests upon an assumption that receives considerably less scrutiny than the technology itself: demand for computing power will continue expanding almost without limit.
That is an extraordinarily large assumption upon which to build hundreds of billions of dollars of investment.
The AI revolution has unleashed an infrastructure race involving semiconductor manufacturers, cloud providers, data-center operators, utilities, energy companies and some of the largest technology corporations on earth. Companies are spending staggering amounts constructing the physical architecture required to train and operate increasingly sophisticated AI systems.
The conventional argument is straightforward. As artificial intelligence becomes cheaper and more capable, people will use more of it. Businesses will discover additional applications. Developers will create new products. Lower computing costs will stimulate additional demand, which will require still more computing capacity.
Economists recognize the underlying concept. When technology makes a resource cheaper and more efficient, consumption of that resource can increase rather than decrease. The phenomenon is commonly associated with the Jevons paradox, originally describing how improvements in coal efficiency contributed to greater coal consumption rather than conservation.
Applied to artificial intelligence, the proposition is seductive: cheaper intelligence creates more demand for intelligence.
Perhaps.
But markets become dangerous when an economic possibility quietly transforms into an investment certainty.
There is an enormous difference between believing AI usage will grow substantially and believing demand for computing capacity will grow rapidly enough, for long enough, to justify virtually every data center, semiconductor fabrication facility, electrical-generation project and capital expenditure program currently being contemplated.
AI does not have to fail for investors to discover that distinction.
Imagine artificial intelligence becoming enormously useful while simultaneously becoming dramatically more efficient. Models could require less computational power. Specialized chips could perform inference far more economically. Businesses could discover that relatively small models accomplish most commercially valuable tasks. Companies might also become considerably more selective about AI usage once executives begin demanding measurable returns rather than approving projects simply because competitors are doing so.
Under such circumstances, AI could succeed technologically while portions of the AI investment trade fail financially.
Capitalism has witnessed this before.
Railroads transformed America, but railroad investors repeatedly lost fortunes. The Internet changed civilization, yet countless Internet companies disappeared after the dot-com bubble. Telecommunications networks became indispensable even as investors suffered spectacular losses after excessive fiber-optic capacity was constructed.
Transformative technologies and profitable investments are not synonymous.
This is precisely where conservative economic thinking offers a useful corrective to technological enthusiasm. Markets work best when capital remains disciplined by profit, loss and genuine consumer demand. They become distorted when narratives replace price discovery and when investors begin treating projected demand as though it were guaranteed demand.
There is another concern. The sheer scale of AI infrastructure development increasingly intersects with government policy.
AI data centers require enormous quantities of electricity, land, water and transmission capacity. Politicians eager to attract technology investment will inevitably offer subsidies, tax incentives, accelerated permitting and other inducements. Washington will be tempted to classify AI infrastructure as strategically indispensable in the technological competition with China.
Some of those policies may be defensible on national-security grounds. But government involvement also introduces the familiar danger of privatizing profits while socializing losses.
If private companies believe Washington considers domestic AI infrastructure too strategically important to fail, investment decisions can become less disciplined. Capital that would otherwise demand convincing returns becomes easier to deploy when investors suspect government will eventually protect strategically important projects.
America should resist that temptation.
The proper response is not hostility toward artificial intelligence. Quite the opposite. The United States should want American companies to dominate AI, semiconductor production, robotics and advanced computing. Maintaining technological superiority over China is both an economic and national-security imperative.
But technological leadership does not require pretending that every investment carrying an AI label deserves an unlimited valuation.
The strongest AI industry will ultimately be one subjected to market discipline. Companies should have to demonstrate that customers will pay for their products, that data centers can earn acceptable returns, and that enormous capital expenditures correspond to economically productive demand.
Artificial intelligence probably represents one of the most consequential technological developments of this century. But that does not repeal economics.
The central question facing investors is therefore no longer simply whether AI will succeed.
It is whether AI will generate enough economically valuable activity to justify the gigantic infrastructure being constructed in anticipation of its success.
Those are two very different propositions.
And history suggests that whenever markets begin treating the second as an inevitable consequence of the first, skepticism is not pessimism.
It is prudence.