It's Giving Enron

On the AI bubble, and the various echoes of the dotcom crash

There are three big stories that one can tell about the dotcom crash.

The first is an overvalued-startup story. Think Pets.com. Startups were massively overvalued. New companies were going public with no revenues and no business model. Everyone pretty much knew it was a bubble. But for years it kept inflating anyway.

Pets.com had the misfortune of being the company that signaled the end. The company IPO’ed, the stock didn’t pop, and that was the bad omen that the extended period of mass investor psychosis was over.

There wasn’t anything especially wrong with the company. They had a market niche and a great ad campaign. Decades later, Chewy.com would get the timing right on basically the same business concept. But when people think “dotcom crash,” the first image that usually springs to mind is that Pets.com sock puppet.

The second is a telecom story. Think Global Crossing. It was clear to a whole lot of companies that the future of the internet would involve fast connections over fiber optic cables. The telecom industry flooded the market with cheap fiber, underestimating the costs and overestimating demand. Figures like George Gilder were convinced that there would be a “Moore’s Law for bandwidth.” Gilder was catastrophically wrong. Part of the dotcom crash involved telecom firms burning a ton of investor cash, going bankrupt, and then selling off all that bandwidth at bargain basement prices.

The difference between the telecom story and the startup story is what they left behind in the aftermath. In November 1999, a startup called Compter.com received $5 million in its first funding round, and immediately spent all that money on a Super Bowl ad. When that company fails, it leaves nothing but a punchline in its wake. But when Worldcom and Global Crossing go bankrupt, they had tangible assets to sell.

And then there’s the Enron story. Ben Riley wrote an excellent piece earlier this year titled EnronAI, revisiting Enron’s role in the crash. The Enron story is fundamentally about accounting fraud, but it was very complicated accounting fraud. Enron played fancy math shell games. The complexity was kind of the point. Make your books complicated enough and you can convince people you’ve invented a newfangled valuation model, at least for awhile. The same dynamics played out during the great financial crisis, expect no one went to prison that time.

So those are our three potential narratives: (1) a startup bubble, (2) unrealistic capital expenditures, and (3) way-too-fancy financial chicanery. All three of these phenomena happened simultaneously, but the lessons we take from the dotcom crash vary depending on which story we emphasize.

---

You can see echoes of all three stories in the current AI bubble. We have AI startups valued at $12 billion before announcing any product. Capital expenditures on AI now account for something like 40% of U.S. economic activity. And the big players are all engaged in (ahem) “circular deals” so complicated you need to subscribe to Matt Levine’s newsletter to have the faintest chance of understanding them.

Even Sam Altman says we’re in an AI bubble right now. But what he has in mind is the first version of this story. The dotcom bubble had Pets.com, but also Amazon. Altman figures there are a lot of Pets.coms out there right now, but his company is the equivalent of Amazon.

For the past year, the AI data center construction boom has given off strong Global Crossing vibes. Microsoft, Meta, Google, OpenAI, and X.ai are all spending billions to build massive data centers. I’m just a simple political scientist who reads old tech magazines, but I cannot fathom how the costs of data center construction are ever supposed to be recouped from a mass user base that pays between $0 and $20/month for the products. (Read Ed Zitron for much more on how little sense these numbers make).

The Global Crossing version of the story tends to elicit righteous shrugs. Sure, the specific companies went bankrupt, but look at how valuable the infrastructure they left behind was! If you think we’re in a telecom-style AI bubble, then that spells trouble for the specific companies along the road to a bright and transformative AI future.

But with the latest wave of multibillion- and trillion-dollar dealmaking among the largest AI players, the vibes are turning decidedly Enron-like. Nvidia announced it is investing $100 billion in OpenAI, which OpenAI will then use to purchase Nvidia products. OpenAI announces a deal to buy $78 billion in chips from AMD, and is awarded 10% of the company in the deal, effectively offsetting the purchase.

None of these astronomically high-dollar deals are necessarily accounting fraud.¹ It wasn’t clear in 2000 that Enron was accounting fraud either, though. Read Ben Riley. Enron was doing cutting-edge financial valuation something-or-other. Low-level and mid-level bankers couldn’t make sense of it, but they were overruled by their bosses who saw a feeding frenzy and wanted a seat at the table. It created a ton of shareholder economic value, right up until the whole thing vaporized itself. The same was true of FTX in 2022. If the crypto market hadn’t precipitously dipped, the fraud at the center of FTX never would’ve been exposed.

What we can say for certain right now is that these massive deals are based on increasingly complex financial shell games. OpenAI — a company that constantly needs fresh injections of investor cash, because its costs are bigger than its revenues — keeps announcing deals to spend billions of dollars to acquire stakes in chip manufacturing companies. Unless they invent digital god and completely transform the entire global economy, like, really pretty soon, they’re inevitably going to run into some hard accounting realities.

This is, obviously, not financial advice. I’m not a finance guy. It’s also not legal advice. I’m not a lawyer. All I am is a guy who pays close attention to the history of the digital future and constantly finds himself muttering “history doesn’t repeat itself but it sure does rhyme” under his breath.

But I’ll say this: the AI bubble isn’t predominantly giving off Pets.com or Global Crossing vibes anymore. It’s giving Enron vibes. When the dust settles, we’ll probably see a whole lot of books written about the fancy math games that convinced investors that two companies passing $100 billion back and forth were creating $200 billion in value instead of having a net financial impact of $0.

1. Also, it’s only fraud if it violates a law that the government still enforces. Even in normal times, that’s a complicated question when you’re dealing with sophisticated and well-lawyered actors. And we aren’t in normal times. I wrote almost a year ago that the Trump economy would “all go pretty well for tech executives and the VC class, at least at first. You can make a lot more money through financial fraud than you can through selling products in a competitive marketplace.” So long as the VCs are running the government, its possible that none of this will be treated as fraud. Much more likely that we see bailouts when the collapse comes.