Artificial Intelligence Is Becoming an Infrastructure Question

The AI debate is still dominated by the wrong question.

We keep asking what artificial intelligence will be able to do. Increasingly, the more useful question is whether the physical and economic infrastructure required to run it can keep up with what companies want it to do.

The AI industry is consuming enormous quantities of computing power, data-center capacity, electricity, and capital. The next stage of the technology race will therefore be decided partly in places that have little to do with chatbots: power plants, transmission lines, semiconductor factories, and industrial construction sites.

That shift matters.

For years, AI was primarily presented as a software story. A clever model could suddenly perform a task that previously required a human. The remarkable part was the software.

Now the bottleneck is increasingly physical.

Investment in AI infrastructure is accelerating as companies build the computing capacity required for increasingly sophisticated systems. Recent investment trends show that AI-related demand is reshaping the technology and financial sectors at the same time.

There is an obvious upside. Infrastructure investment creates jobs, supports technological innovation, and can improve American competitiveness. The United States has strong incentives to remain at the center of the AI economy rather than allowing the most important computing infrastructure to migrate elsewhere.

But infrastructure creates costs as well as benefits.

Data centers require enormous amounts of electricity. Communities have to decide whether they want new facilities nearby. Utilities have to determine who pays for grid upgrades. Investors have to decide whether today's enormous valuations will eventually be justified by actual profits.

None of those questions can be answered by saying that AI is "the future."

That phrase is becoming a substitute for analysis.

The most interesting possibility is that AI's largest economic effect may not come from replacing workers directly. It may come from increasing the productivity of existing workers. A lawyer who can review documents faster, an engineer who can test designs more quickly, or a small company that can automate routine administrative tasks may become substantially more productive without eliminating the human role altogether.

That outcome is hardly guaranteed. Some jobs will change dramatically, and some will disappear. But economic transitions rarely occur in the simple manner predicted by either technology enthusiasts or technology pessimists.

The government's challenge is similarly nuanced. It should establish rules for safety, privacy, liability, and transparency without attempting to freeze a technology that is still developing.

Europe's AI Act illustrates the scale of the regulatory shift now underway. Major portions of the European Union's AI framework became applicable this month, creating new obligations for companies operating in the region.

America will have to decide how much of that model it wants to emulate.

The answer should not be "all of it" or "none of it."

The most sensible AI policy will probably be less dramatic than either side wants. It will involve targeted rules, technical standards, enforcement mechanisms, and considerable room for experimentation.

That may sound boring.

For a technology this consequential, boring might be exactly what we need.