Money: Why An AI Slowdown Is Likely Regardless of Safety Concerns

Forget AI-driven extinction for the moment - a big ask, we know - but maybe not the greatest immediate concern. Alot is being written about the threat to humanity from AI, while the fact that those who are pooh-poohing the danger - Jensen Huang, Mark Zuckerberg and, lord have mercy - Donald Trump, suggests that self-interest is playing a larger role than objective analysis.

But the reason that an AI slowdown is increasingly likely is because the numbers underpinning the projections of AI growth have gone from Panglossian to delusional. Among the basic problems now emerging is the fact - yes, fact - that there is not enough copper wire, enough construction workers and enough energy to build all the data centers being proposed. And without the data centers there is not enough 'compute' capacity to generate the back end calculations required for those growth predictions. Add to that another fact, that the big tech companies are now issuing debt at levels never before seen. And the thing about debt is that the issuer is contractually required to pay interest as well as paying it off. Which savvy debt investors are now beginning to question tech's ability to do. So the fundamental issue is less about extinction than about money - always the issue in tech - and that means the financial spigots are about to turn off, perhaps even for Nvidia, which has become circular lender of last resort. JL

Spencer Jakab reports in the Wall Street Journal:

The market reaction to a slowdown by industry leaders for safety reasons was telling. Shares of Microsoft, Alphabet and Meta jumped on an awful day for tech. Worst-hit were companies providing chips, power or data centers. It was an overdue reality check. The physical world has limits. Hyperscalers’ plans were starting to bump up against them. Data centers will reach 12% of total U.S. electricity demand by 2030, the equivalent of 30 and 60 nuclear plants. Public backlash has meant going from tax breaks to opposing data centers. Supplying copper to wire them would lead to a 25% shortfall by 2040. (And) After exhausting their own cash flows, hyperscalers' a borrowing binge is worrying investors. In one year, their share of the leading bond index grew 78%. Making enough profit to justify the biggest infrastructure splurge ever is a tall order. With open source competition rising and token prices falling, investors having doubts. “How is all this going to be monetized?”

Are investors dying to invest in a technology that might literally kill them? 

After an Anthropic employee put the chances of AI doing that at more than 10%, economist Owen Lamont quipped that it was “like Robert Oppenheimer doing an IPO for the Manhattan Project in 1945.” 

Factoring human extinction into stock values is silly, so let’s hope for the best and just stick to stocks. The market reaction Monday to a proposed slowdown by industry leaders for safety reasons was telling. 

Shares of hyperscalers Microsoft, Alphabet and Meta all jumped on an otherwise awful day for tech. Worst-hit were companies providing chips, power or infrastructure to hundreds of planned data centers.

It may have been an overdue reality check. Trees can grow to the sky on analyst spreadsheets, but the physical world has limits. Hyperscalers’ plans were starting to bump up against them.

Take power needs. Data centers are set to reach 12% of total U.S. electricity demand by 2030, according to Rystad Energy. That’s the equivalent of between 30 and 60 nuclear plants of additional capacity. Politicians facing a public backlash have gone from doling out tax breaks to opposing the facilities in a matter of months.

And maybe it’s a moot point: Supplying all the copper to wire them would have led to a 25% shortfall of supply for the vital metal by 2040, according to S&P Global. A major new copper mine takes 15 to 20 years to get up and running.

Paying for data centers is starting to create strains too. After exhausting their own mighty cash flows, hyperscalers have gone on a borrowing binge that’s starting to worry credit investors. In just one year, their share of the leading investment grade bond index grew by 78%, according to Capital Group.

Making enough profit to justify the biggest infrastructure splurge ever is a tall order ahead of Anthropic and OpenAI’s hotly anticipated IPOs. With open source competition rising and token prices falling, Ryan Kelley, chief investment officer at Hennessy Funds, is among those having doubts.

“How is all this going to be monetized?”

One hopes that the safety concerns from AI researchers are overblown. They certainly seem sincere, but some investors suspected ulterior motives when CEOs were so quick to consider tapping the brakes.

Might it be a form of collusion, trimming cash needs while charging more—sort of like a technological OPEC? Or maybe CEOs just realized investment plans were unrealistic, needing too much of the world’s energy and savings to be feasible, much less profitable? 

The fact that the companies committing their shareholders’ cash rallied on the news suggests that investors were already nervous about more than just safety. That could be bad news for the companies selling the picks and shovels of this gold rush.