The brutal math behind AI: it’s either you or Zuckerberg

Back in the mid-90s, the world was abuzz with a new technology called the Internet.

Technically the Defense Department had invented it back in the late 1960s. But by the mid-90s, Internet use was growing at an an astonishing rate among American consumers, triggering a major need for more digital infrastructure.

So in March 1997, a company called Global Crossing was born. Their entire reason for existence was to build the backbone of the Internet— thousands of miles of cables and fiber optic lines to carry all of our bits and bytes at the speed of light.

Global Crossing was easily able to raise billions from eager investors who understood that the Internet was the future; they believed that Internet usage would explode, and that, eventually, Global Crossing would generate substantial returns on its massive infrastructure investments.

They were right. Partially. Internet usage exploded. Unfortunately the part about generating substantial returns on their massive infrastructure investments… never happened.

In fact, within a few years, Global Crossing was losing billions. They had spent everything on cables and fiber optics, but they were never able to generate enough revenue to recoup their investment.

So in January 2002, less than five years after its founding, Global Crossing filed for bankruptcy.

What was left of the company was eventually sold for just $250 million during bankruptcy proceedings, representing a 99.5% loss in value from its $47 billion peak.

And most of Global Crossing’s investors suffered extraordinary losses.

Coincidentally, though, consumers ended up benefiting. Global Crossing did provide a valuable service to Internet users. They just weren’t able to make enough money from that service to remain a viable business.

So ultimately Global Crossing was a transfer of wealth from investors to consumers. Deep-pocketed, overly enthusiastic investment funds essentially subsidized all the digital infrastructure for Internet users.

(And as an early Internet user from the 90s, I appreciate them doing that!)

This type of outcome wasn’t an isolated incident. In fact it’s fairly common.

Remember WeWork? Customers were able to enjoy cheap office space at below-market rents (not to mention free tequila), courtesy of big investors like SoftBank and Goldman Sachs.

Uber was a similar story—customers enjoyed cheap rides at below-market rates thanks to investors who kept funding the company’s enormous losses.

Again, this is not unusual. In a burgeoning industry, investors often get so excited about some new technology or idea that they end up subsidizing it in the name of developing the market.

They throw insane amounts of money at something, even when there’s not really a valid business or revenue model behind it.

And it’s becoming pretty clear we’re starting to see this again with AI.

Even Jeff Bezos acknowledged it over the weekend. At a technology conference in Italy, he said that while the social benefits of AI will be huge, the investment side is starting to feel a lot like the dot-com bubble. Perhaps that’s why he recently cashed out billions of dollars of Amazon stock.

Plus Sam Altman of OpenAI said back in August that investors are too “overexcited” about AI.

Ironically he’s a big part of that bubble. OpenAI is leading the charge, spending absurd amounts of money on data centers.

And they’re not alone. Every few weeks it seems there’s another tech company announcing that they’re going to build the biggest data center— Meta, xAI, etc.

They even give these projects lofty names like Stargate and Colossus—sexy sci-fi branding to match the eye-watering price tags that go into the hundreds of billions.

But when you sit back and actually look at the numbers, the math just doesn’t work.

$500 billion for Stargate? Come on. What a money pit.

A big chunk of that capital— well north of $150 billion— will be earmarked for chips and other digital components. And, sure, that’s great for semiconductor companies… not to mention their suppliers, certain commodity producers, and even utility companies.

But the tech companies themselves will likely lose a lot of money.

Look at OpenAI. They’re reportedly pulling in $12 billion in annual revenue. That’s a lot. But they also have enormous costs—electricity, cooling, maintenance, payroll.

Altman himself has said that users politely saying “please” and “thank you” to ChatGPT costs the company tens of millions of dollars in extra electricity. So if they’re really, really lucky, they’ll have a few billion dollars left over in operating cash flow.

A few billion dollars is a pretty tiny return on a $500 billion investment. What’s more, that ~$150 billion in chip costs is not a one-time thing.

Remember, those chips will be obsolete in five years. Tops. Probably more like 3-4.

So they’ll need to replace those chips (and hence spend ANOTHER $150 billion) in a few more years. This means that, on average, OpenAI’s annual capital expenditure, just on chips, is at least $30 billion. Per YEAR. For a company that only makes $12 billion in revenue, and (maybe) a few billion in Operating Cash Flow— if they’re even cash flow positive at all.

You can start to see how the math simply doesn’t add up… and why, at least for now, these investors are essentially subsidizing AI for consumers around the world.

The only way the investors can actually recoup their costs—let alone generate a return on all that data center money—is if AI revenue expands rapidly, i.e. 50x growth in a very short period of time.

Is that possible? Of course it’s possible. But it’s hardly a slam dunk.

Most consumers right now don’t pay for AI. Or if they do it’s small dollars ($8/month).

What the AI evangelists are hoping for is that the big revenue will come from major companies— large corporate clients who replace expensive, salaried human beings with $299/month AI subscriptions.

Theoretically those mass layoffs could generate enough revenue for AI investors to eek out a modest return on investment.

McKinsey estimates that there would need to be $1 trillion to $1.5 trillion in AI revenue by 2030 (up from about $30 billion today) in order to justify all the data center investment.

But that amount of revenue would require laying off over 400 million people worldwide— more than twice the size of the entire US labor force— and replacing them with AI subscriptions.

There are other possibilities, of course. Maybe a global productivity boom is able to unlock enough economic growth that the AI investments ultimately pay for themselves. I hope so. But given all the anti-AI regulation in various governments (I’m talking about you Europe!) I’m not holding my breath.

It’s also possible that these big tech companies and AI investors take a huge bath and lose hundreds of billions of dollars. And this scenario would carry adverse consequences for pension funds, retirement programs, etc. who are invested in these big tech companies.

At a minimum, while it’s pretty clear that AI is obvious the future (no to mention the present…), the financial outcome carries a lot of risk and uncertainty. So definitely take the evangelism with a grain of salt.

Source

Simon Black is an international investor, entrepreneur and permanent traveler. His daily letter is both educational and entertaining, and we suggest that those who want unbiased, actionable information about global opportunities sign up for Sovereign Man’s free, actionable newsletter at http://www.SovereignMan.com.

From Simon Black of SovereignMan.com

Source: https://www.schiffsovereign.com/trends/the-brutal-math-behind-ai-its-either-you-or-zuckerberg-153669/