THE AI RISK WE ALREADY BUILT

There is a very large argument happening about artificial intelligence.

Could it become too powerful? Could humans lose control of it? Should development slow down? Should there be more guardrails?

Those are fair questions.

We have another one.

What are humans already changing in order to build it?

First, we made it easier to build.

In July 2025, President Trump signed an order designed to accelerate large data-center construction. It revoked President Biden's January 2025 AI-infrastructure order and directed the government to identify ways to speed environmental reviews and permitting.

The same order said qualifying projects could receive financial support including loans, loan guarantees, grants, tax incentives and offtake agreements.

Which makes what happened next particularly interesting.

Then we needed protection from the buildout.

In March 2026, the White House announced something called the Ratepayer Protection Pledge.

Its premise is remarkably straightforward: if hyperscalers create enormous new electricity demand, ordinary households should not get stuck paying for the new generation, transmission and grid upgrades required to serve it.

Amazon, Google, Meta, Microsoft, OpenAI, Oracle and xAI agreed to build, bring or buy new power and cover infrastructure costs associated with their data centers.

And the money got very large.

By early August 2026, AI-related debt issuance had reached nearly $500 billion for the year, according to Goldman Sachs analysts cited by Reuters Breakingviews.

Reuters also reported that lenders were becoming more cautious because projects were running into familiar physical-world problems: electricity queues, permits, supply chains and public opposition.

About all those jobs...

Data-center announcements frequently arrive carrying very large job numbers. Construction does create real work. But construction employment and permanent operating employment are not the same number.

One useful receipt comes from Washington State's own legislative auditor. It reviewed an urban data-center tax preference and reported that beneficiaries saved an estimated $42.4 million from 2023 through 2026.

The auditor also noted that it could not determine how much of the activity occurred because of the tax preference, and that none of the projects using it had produced a new urban data center. That's the state's report, not ours.

So naturally, we're considering space.

In April 2026, the U.S. Government Accountability Office examined proposals for data centers in space.

One proposed advantage?

Reducing the land, electricity and water demanded by data centers on Earth.

There are, admittedly, details.

GAO notes that space is surprisingly inconvenient for cooling enormous computers. Large systems could require solar arrays bigger than anything assembled in space so far, cooling at that scale remains unproven, and huge satellite constellations would introduce collision and debris concerns.

And then the risk conversation changed.

On September 14, 2026, President Trump pushed back against calls for additional AI guardrails and described opposition involving AI and data centers together as a “SICK conspiracy.”

He also called data centers the “oil of the next 20, 25 years.”

And that's where two very different arguments begin occupying the same sentence.

One argument asks what increasingly powerful artificial intelligence might someday do.

The other asks about the land, power and money already being spent to build it, right now.

You can be optimistic about artificial intelligence and still ask who pays for its power plant.

You can think predictions of AI extinction are wildly overstated and still ask whether a tax incentive produced the jobs promised.

You can want America to lead in AI and still ask why the government needed a pledge specifically designed to prevent ordinary Americans from paying for the infrastructure.

They're different questions.