“Something Wicked This Way Comes”

It’s only mid-August, but I already see Halloween gear going up in stores everywhere. As a result, I’ve got the macabre on my mind.

Not anything as commonplace as gremlin costumes and ghost displays, mind you. It’s nothing that easy.

More along the lines of Ray Bradbury’s Something Wicked This Way Comes.

That novel, written in 1962, tells the tale of a mysterious carnival that rolls into a small Illinois town. At first, its show seems magical, promising people exactly what they want. Youth, power, excitement: It’s all there for the taking.

Yet it’s ultimately for a price.

Because there’s always eventually one to pay.

Most of us understand that’s true in theory. Yet it’s easy to ignore in practice when something seemingly awesome “this way comes.”

That’s why Something Wicked is such a classic. It tells a timeless truth that applies just as easily to the space race of the 1960s as to artificial intelligence today.

Make no mistake of it: AI is extraordinary. And it’s already transforming almost everything we do.

However, we’re going to eventually see the real price of this merry-go-round. In which case, we might not like everything we see as it continues to turn.

For the last three years, Silicon Valley has spent extraordinary amounts of money building AI infrastructure.

Data centers, power plants, transmission lines, cooling systems… Big Tech is sparing no expense on any of it. That’s why Morgan Stanley estimates these kinds of companies will spend some $2.9 trillion on support systems through 2028.

It also estimates that hyperscale cash flows will fund about $1.4 trillion. Which leaves $1.5 trillion to pay for through things like:

And, eventually, the public.

In short, hyperscalers may have financed the first phase of the AI boom. But Phase II will have to be increasingly backed by credit – and credit has a price.

Big Tech’s financial profiles are changing even now as free cash flow gets crushed by enormous capital expenditures. And no, corporate bonds can’t really cut it any longer. Nor can traditional debt in general.

That’s why we see Nvidia (NVDA) working with major financial institutions to finance hundreds of billions of dollars of AI infrastructure. As the chip giant announced on Monday, it’s secured:

Stop and think about that for a good, long minute: The company selling essential equipment is now trying to help its buyers afford those parts and pieces.

The vendor is now financing the customer. That’s where we are right now in the AI spending boom.

It’s more than enough to tell you just how enormous (and increasingly difficult to finance) this buildout has become.

Wall Street has a tried-and-true solution whenever an industry needs more debt than lenders will give:

Sure enough, that's exactly what's beginning to happen with data centers.

Back in February, for example, Compass Datacenters completed an $830 million securitization. About $500 million of that received Moody’s highest possible credit rating at AAA, with senior bonds priced at roughly 120 basis points above their benchmark.

If you've studied financial history – like, say, the subprime crisis – that should make you very, very uncomfortable. It means Wall Street is creating securities to move debt elsewhere, and ratings agencies are stepping up to certify the safest pieces.

Notice I didn’t say “safe,” but “safest.” As in the parts that are less likely to fail. That way:

The machine feeds itself that way over and over again… until it can’t anymore.

Knowing that, it’s important to take note of the SEC's Office of Structured Finance and its recent response to a request concerning data center securitizations by Latham & Watkins. This firm, for the record, “advises the businesses and institutions that power the global economy.”

Including major tech companies.

On July 29, the SEC concluded that the items in question were not “asset-backed” under Section 3(a)(79) of the Securities Exchange Act. And that’s a very big deal.

You see, after the 2008 financial crisis, the powers that be created an extensive framework governing securitized credit. This included investor protections like disclosure and risk retention.

As we’re now seeing, not every data-center deal meets those specifications – a fact that should concern everyone involved.

Which, like it or not, is absolutely all of us in this intensely interconnected AI world we’re living in.

History has shown us repeatedly how technology can be 100% real, 100% transformative, and yet 100% mishandled.

That was true with railroads, telecommunications, and the internet. Each one generated so much investment hype that bubbles were formed… and subsequently burst.

Moreover, we’re not in the early stages of that pattern. I’m not saying it’s all going to blow tomorrow, but it’s been literal years in the making.

In response to the Covid-19 lockdowns, the Federal Reserve added trillions of dollars to its balance sheet in a very short amount of time. By late 2021, the country was awash in short-term liquidity… with an enormous amount of that finding its way into the Fed’s overnight reverse-repurchase facility, or RPP.

This is a giant reservoir for money-market funds and other eligible institutions to park their cash. And at its peak near the end of 2022, the RPP contained roughly $2.5 trillion.

It’s been draining ever since though as Treasury yields grew more attractive, bill supply expanded, and investors looked to advance AI-driven profits… all while interest rates rose. In fact, as of late last year, that pool of liquidity was essentially gone.

Yet hyperscalers’ need for funding has only increased.

It’s clear that the math doesn’t add up. It subtracts.

Put another way, suppose that an asset produces $100 of annual cash flow at a required 5% return. That means it’s worth $2,000.

But if you raise that requirement to 6%, its value falls to $1,667% – even though nothing happened to the asset itself. Add another half-percentage point, and it falls further to $1,538: a 23% decrease from our original 5% level.

This basic scenario is playing out right now, with investors demanding higher and higher returns. As such, a 20%–30% decline in equity values isn’t unthinkable.

I hate to say it, but it’s true.

If anything, what we’re already seeing could snowball further. Falling assets could easily reduce collateral values, making lenders demand higher spreads still.

That, in turn, would make AI projects more difficult to finance. Some projects (perhaps many) would have to be canceled. Suppliers would lose orders. Equity investors would start demanding lower valuations…

Just like that, the credit cycle reverses.

That’s why I completely agree with Wall Street for growing more cautious about the AI revolution. While I might not always agree with its conclusions, it’s high time we question hyperscalers’ free cash flow – or lack thereof.

We also want to pay more attention to how interest rates affect them and what their credit spreads are. And, above all, we want to know how much securitization is happening.

Things simply get too dicey once an investment boom has to fall back on borrowed money. There’s very little room for missteps when debts are packaged, rated, and distributed throughout the financial system.

Banks, insurers, pension funds, private-credit funds, and ordinary investors: We’re all now officially involved whether we like it or not.

As such, it seems high time we get on Ray Bradbury’s page. He understood how easy it is to go all-in on our desires when easy opportunities present themselves.

But that can lead to frightening results, whether through the carnival from hell or from a giant financial bubble bursting all over the financial world again.

When the bill comes due, the bill comes due. And there’s no escaping when it does.

Happy SWAN investing,

Brad Thomas
Editor, The Wide Moat Daily