Last weekend, I noted that it’s not realistic to expect America’s largest hyper-scalers to fund the AI buildout with free cash flow. They’re going to lever the balance sheet (i.e., borrow) to foot a portion of the bill, the same way you or I borrow to fund big purchases even when, theoretically, we don’t have to. It’s an LBO — a leveraged buildout.

In principle, there’s nothing wrong with that. Notwithstanding the notion that leverage is everywhere and always positively correlated with a given borrower’s sensitivity to the vagaries of life (i.e., the more leverage you have, the more sensitive you are to volatility and vice versa), show me someone with good credit who brings a shoebox full of fifties to buy a car when the dealership’s offering well-qualified borrowers below-prime-rate financing, and I’ll show you a moron. (Or a rapper. Or Morgan Housel.)

If you’re, say, Meta or Alphabet (both of “whom” are borrowing heavily in part to fund AI), you’re not only a well-qualified borrower, you’re among “the most creditworthy customers in existence,” as The New Yorker put it late last month, in a good piece which begins by noting that if you “drive in almost any direction from almost any American city, soon enough you’ll arrive at a data center — a giant white box rising from graded earth, flanked by generators and fenced like a prison yard.”

So, if the question’s whether anyone should be concerned in principle about the best companies on Earth borrowing huge sums to create and access computing power as opposed to plowing every last penny of spare cash into the effort out of some pathological aversion to debt, the answer’s a hard “no.”

As I wrote in “40 Acres,” when you set out to borrow money and someone lends it to you, that’s an expression, by both parties, of confidence: You’re confident you can ultimately pay back the loan, and the lender shares that faith. Credit (debt) gets a bad rap. In essence, it’s a shared expression of faith in the future on the part of both the borrower and the lender. Whether that faith’s misplaced for any given loan is an entirely different discussion, and that’s where the AI debate is currently.

With all of that in mind, the figure below, updated to account for the latest bottom-up estimates (i.e., projections from company analysts post-earnings season), gives you a sense of what it would mean if the hyper-scalers tried to fund the arms race with cash.

Could they do it? Well, yes. Technically. If you assume i) capex estimates aren’t too low (they’ve been revised higher every quarter looking back more than a year), ii) cash flow estimates aren’t too high and, more to the point, iii) these companies are willing to plow more or less every spare dollar into the AI buildout at the expense of, most notably, buybacks.

Not only is that latter proposition unrealistic, I’d argue it’s inadvisable and certainly undesirable for those of us who depend on buybacks to artificially inflate stock prices.

Of course, the relative wisdom of leveraging your balance sheet (and this is true not only for companies but for households too) is everywhere and always a function of the cost to borrow, which is in turn a function of supply. And as BofA’s Michael Hartnett noted, issuance is up sharply over the past two months and spreads for the hyper-scalers have almost surely seen the lows / tights.

There’s the chart. Spreads are 50bps wider just since September, albeit still very, very forgiving on a 10-year lookback.

“We short hyper-scaler bonds,” Hartnett wrote, in his latest. “Cash flow’s insufficient to fight the AI capex arms race [and] US tech bond prices fell 8% in the 12 months prior to the March 2000 bubble peak.”

In the same note, Hartnett referenced the OpenAI “bailout” banter. “Even the kings of AI [are] hinting at a government backstop to lower the cost of capital,” he said.