Capital Wealth
WED CLOSE · SEP 16   S&P 500 7,551.81 ▼0.45%  ·  DJIA 51,461.90 ▼1.21%  ·  NASDAQ 25,978.42 ▼0.01%  ·  10-YR 5.003%  ·  2-YR 4.725%  ·  WTI $102.43 ▼3.2%  ·  GOLD $4,346.30 ▲1.3%  ·  VIX 17.71 ▲3.0%
Markets & The Fed · The AI Build-Out

An AI Slowdown Was Coming Anyway. Copper and Kilowatts Were Going to Force It.

When AI’s own leaders proposed slowing down, the hyperscalers rallied and their suppliers got hit. Heard on the Street reads that reaction as a reality check about physics, not safety.

By Sean Anees Saifi · Capital Wealth · Published Thursday, September 17, 2026 · Source: The Wall Street Journal, September 17, 2026 edition, plus Wednesday’s market close and prediction-market odds
Key Points
12%
of U.S. electricity demand from data centers by 2030
30–60
nuclear plants’ worth of new capacity required
25%
projected copper shortfall by 2040
78%
one-year growth in hyperscalers’ share of the IG bond index
An aerial view of a power plant under construction at sunset, cranes over steel framework beside a new substation and transmission lines.
Data centers are set to reach 12% of total U.S. electricity demand by 2030 — the equivalent of 30 to 60 nuclear plants of new capacity.
In one line: The market’s reaction to an AI slowdown proposal — buy the spenders, sell the suppliers — suggests investors were already worried the spending plans couldn’t physically happen.

Spencer Jakab opens his Heard on the Street column with the only joke available on the subject. After an Anthropic employee put the odds of AI destroying humanity above 10%, the economist Owen Lamont observed that investing in it was “like Robert Oppenheimer doing an IPO for the Manhattan Project in 1945.”

Then Jakab does the sensible thing and sets extinction aside, because it is not a variable you can put in a spreadsheet. What’s left is a market signal that is genuinely interesting.

Read the reaction, not the proposal

When industry leaders floated a coordinated slowdown for safety reasons, the stocks moved in a revealing pattern. Shares of the hyperscalers — Microsoft (MSFT), Alphabet (GOOGL) and Meta Platforms (META) — jumped, on an otherwise awful day for technology. The worst hit were the companies selling them chips, power and infrastructure: Nvidia (NVDA), Micron Technology (MU), CoreWeave (CRWV), Digital Realty Trust (DLR), Corning (GLW).

Think about what that means. Investors rewarded the companies that would spend less and punished the ones that would sell less. That is not how a market reacts to a safety announcement. It is how a market reacts to relief.

Three walls the build-out was already hitting

Power. Data centers are set to reach 12% of total U.S. electricity demand by 2030, according to Rystad Energy. That is the equivalent of somewhere between 30 and 60 nuclear plants of additional capacity. And the politics have turned fast — as Jakab puts it, politicians went from doling out tax breaks to opposing the facilities in a matter of months.

Copper. This is the one that stops the argument. Supplying all the copper to wire the planned data centers would have led to a 25% shortfall of the metal by 2040, per S&P Global. A major new copper mine takes 15 to 20 years to get up and running. You cannot schedule your way around geology.

Money. Having exhausted their own considerable cash flows, the hyperscalers went on a borrowing binge that is starting to worry credit investors. Their share of the leading investment-grade bond index grew 78% in a single year, according to Capital Group.

That last number deserves a moment from anyone who owns a bond index fund. A “diversified” investment-grade bond portfolio now carries substantially more exposure to the AI capital cycle than it did twelve months ago — and nobody sent a notice.

The question the spending has to answer

Ryan Kelley, chief investment officer at Hennessy Funds, asks it in five words: “How is all this going to be monetized?” With open-source competition rising and token prices falling, making enough profit to justify the largest infrastructure splurge in history is, as Jakab notes, a tall order — and two hotly anticipated public offerings are waiting on the answer.

Some investors suspected less noble motives in how quickly chief executives agreed to tap the brakes: a coordinated slowdown trims cash needs while supporting prices, which starts to resemble a technological cartel. Or, more charitably, the executives simply looked at the power, the copper and the borrowing and concluded the plans were never feasible.

Either way, Jakab’s conclusion is the one that matters for a portfolio: the fact that the companies committing their shareholders’ cash rallied on the news suggests investors were already nervous about more than safety. And that is bad news for the companies selling the picks and shovels.

What It Means For Your Portfolio

Hold — own the build-out through the builders, not the suppliers

A slowdown is good for the companies that would have spent the money and bad for the ones that would have sold the equipment. Which side of that you own matters more than a view on AI.

General planning principles, not advice for anyone in particular. The copper and kilowatt numbers are the most useful part of this column, because they are physical constraints rather than opinions. A plan that requires 30 to 60 nuclear plants and a quarter of the world’s copper supply gets revised whether or not anyone is worried about safety.

The AI exposure in the portfolios is deliberately held through platforms and builders — Microsoft (MSFT), Meta Platforms (META), Alphabet (GOOGL), and the power and grid names GE Vernova (GEV) and Quanta Services (PWR) — rather than concentrated in the equipment suppliers whose revenue is somebody else’s capital-spending line. Nvidia (NVDA) and Micron (MU) are held at weight, not added to. The one genuinely new item to check is passive: a bond index fund whose hyperscaler weight grew 78% in a year is carrying an AI position its owner didn’t choose.

Book a 15-Minute Review → Back to Edition No. 172 →