For this week’s Must Read–now that we are past World Cup and mostly past America250–here’s another bracing essay from Ed Zitron’s Where’s Your Ed At in The Subprime Data Center Crisis. Get a couple of cuppas and some time–it’s dense.

Mr. Zitron compares how data centers are financed today to the pre-2008 boom in collateralized debt obligations (CDOs) that insured mortgage bonds. CDOs, by raising money around these bonds near-endlessly, subsidized an artificial boom in housing demand stimulated by  historically low interest rates. Those of us around (nearly all our Readers) know how badly that wound up in 2008, with the circular grift crash taking down Lehman Brothers and other financial institutions and putting us into a four-year recession. It also made financing nearly unavailable for health tech companies just getting out of their Series A or B.

CDOs aren’t financing data centers, but Special Purpose Vehicles (SPVs) are. The SPV raises debt financing for a data center and sells it to institutional investors, asset managers or banks. The SPV makes the payments to the contractors and suppliers such as Nvidia for chips. How they pay interest on the debt is via a special account. Finally, when the data center starts to generate revenue, the SPV uses those funds to first pay for the operating expenses of the data center, then creditors (based on their seniority in the debt), then whatever is left goes to the holding company, such as CoreWeave. It reveals how companies like Meta, Google, Amazon, and Microsoft are financing their data center builds. (It does appear on this Editor’s reading that Oracle, by directly taking on debt, is taking a different route.)

It’s a complicated scenario. The analogy he draws is that the financing for data centers is equivalent to the subprime mortgages of 2008. It’s  compounded by the apparent fact that capacity being built in the pipeline far outstrips demand by a factor of 15, based on an industry report cited by Mr. Zitron. In addition, the demand is not from profitable companies. He cites that “70% to 90% of that demand is from Anthropic and OpenAI’s unprofitable services.”

It also assumes three factors govern AI expansion in a grab for world savings looking for safe bets:

1. AI data center demand is infinite and all compute will be used.
2. AI data centers all have “locked-in customer demand.”
3. That these are “safe” investments, backed by the richest companies in the world.

Exactly what demand is out there? Companies and countries are guessing and making huge financial bets. Mr. Zitron basically calls it an AI bubble inflated by AI companies and abetted by the media because no one is accurately measuring the demand versus capacity.

It’s a scenario that screams ‘red flags’ and ‘Danger, Will Robinson’.

Read, you decide. 

But the scenario Mr. Zitron outlined raised the goose pimples on this Editor’s neck.

Those of us in healthcare have been to this rodeo before. And the bronco riders were all taken to the ER. Or the morgue.

If the term SPV sounds a little familiar to Readers, to this Editor, the term and the structure strongly resemble Special Purpose Acquisition Companies (SPACs) which for a time in 2020-2022 were the It Girl of getting around the typical IPO process for many a health tech company, such as Babylon Health. Admittedly, SPACs and SPVs work differently. SPACs raised money from investors, IPO’d themselves to raise more money, and then in an ‘blank check’ transaction, sometimes with additional investment, acquired a private company, thus taking it public. All in 3-6 months! Simple, right? None of the muss and fuss, due diligence, or SEC scrutiny of an IPO. And as a device, it’s still being used.

Yet look at the outcome in healthcare. Nearly all of the healthcare SPACs ‘cracked’ after 2023 with stock values cratering within months or a year, Hims being one of the very few exceptions and maybe the only one. 30% went bankrupt. 26% were acquired well below their IPO price. And the remainder survive, some having flirted with the Devil of Demise, all below their IPO and valuation. For a tidy summary of the rolling SPAC collapse, read TTA 10 April 2025 and 26 June 2024. 

But what CDOs, the current AI bubble, and the 2020-2022 health tech/telehealth bubble have in common is a complicated way of financing designed to skirt regulations and proper market analysis, coupled with a healthy dose of illusions. Each bubble is based on a set of assumptions that envisioned endless geometric growth and future profitability, coupled with a “desire to believe” that negated logic and the entry of outright fraud. In 2020, we had a real lack of accurately gauging demand for health tech and telehealth services, especially in direct-to-consumer and telementalhealth services. The healthcare SPAC bubble also misread the viability of companies and their business models, many of which had more in the realm of hopes and founder dreams than validity in the marketplace. SPACs evaded scrutiny and created great headlines for this Editor. It was endless. Apprehensions were brushed away. It was also unsettling to this Editor who had a very small role in a long-ago iteration of a financing and business boom/bust–airline deregulation–at two airlines.

When the 2023 crash came, wiping out billions in investment, it was pretty much confined to healthcare and health tech. It’s still being worked through in the consolidation of health tech (not all bad and to be expected), in bankruptcy court (23andMe), and the outright frauds in Federal courts (Done Global). To quote myself from June 2024: The investment scene in health tech and AI strongly resembles the Wild West days of airlines post-deregulation 30 years ago. Investor money in, now fleeing for the exits, whether the bankruptcy court or passing the hot potato to others with money.

AI, of course, dwarfs the SPAC-driven health tech boom by a 1000X factor. And has the capacity to take down a world economy.