"The Looming A.I. Market Bubble"
by John Wilder
"Elon Musk promises a supercomputer cluster bigger than Texas that’ll make Skynet™ look like an HP-15C®. It even has a creepy name for those who know film history: Colossus™. Of course, it’s going to require more power than a quiver of Antifa® mainlining Red Bull© during a riot. I like that. A herd of cattle, a murder of crows, and a quiver of Antifa©.
But it’s not just Elon. There’s also Sam Altman, that pint-sized messiah of OpenAI© is out here swearing he’ll build data centers the size of Afghanistan, all to birth the AI-god-emperor that’ll finally figure out why fish from Long John Silver’s® always tastes like regret.
But here’s the kicker: this might be the biggest Ponzi scheme in history. If /When this AI bubble pops, it may very well make the dotcom crash look like look like a lost wallet. On recent analysis I saw was over here (LINK) by Ed Zitron, and no, I’m not going to make fun of his last name as tempting as that might be since he writes well. When I read it, it wasn’t behind the paywall, but it was also insightful. Trust me. His conclusion?
According to Ed’s analysis, the AI hype train is barreling toward a cliff made of physics, bad math, and even worse economics. If Mr. Zitron is correct, trillions of dollars are being flushed down the toilet on promises that of a technical revolution which, while automating many boring tasks, unfortunately won’t replace the staff at the DMV.
First off, the promises. OpenAI’s® scribbled deals on cocktail napkins that will eventually result in laws prohibiting what they’re doing. As I mentioned in a previous post, they’re committing to drop $300 billion on Oracle™ over five years. That amounts to $5 billion a month, which is more than Taylor Swift makes in an entire year. Just kidding, but that $5 billion a month is a big number, since OpenAI only made $4.3 billion in the first six months of 2025.
OpenAI™ doesn’t have the money, of course, but, hey, it’s a bubble, so who is counting? They have stock, so if they don’t have cash, they’ll just give you stock. What is OpenAI© buying with that cash that they don’t have? A gigawatt-scale data center orgy that’ll need more energy than Switzerland. Probably. Maybe. I’d need to know how many electric toothbrushes the Swiss use to be sure. But, the problem is, nobody has built a gigawatt data center. Ever.
The biggest data centers today top out at maybe 100 megawatts, and that’s if the grid fairies are feeling generous. Take Stargate Abilene, OpenAI’s© “investment” with Oracle®. It’s supposed to hit 1.2 gigawatts, but right now? They’ve got a puny 200-megawatt substation and some jury-rigged natural gas turbines that might squeak out another 350 megawatts if we can talk the Chinese into sending us the rare earth materials to make them.
Reality check: to run just this one location, they need 1.7 gigawatts total just to cover cooling and losses. And, it’s in Texas, which is not known for being a good place to keep stuff cold. They picked a climate where cooling the data center will be like trying to cool my nether regions in a sauna using a hairdryer.
And the power? Forget it. Transformers and substations take 2-4 years to build, and we’re fresh out globally. The article quotes some Bloomberg® wonk admitting they’re slapping together “not the really good” turbines because the premium ones have a seven-year waitlist. Seven years! By then, those fancy Nvidia™ H100 GPUs will be as obsolete as Taylor Swift’s ovaries.
None of this is hyperbole. This is simple math: Taylor’s really getting up there if she wants to have kids. But back to the data center. Roughly, if you have a gigawatt of power that gets you maybe 700 megawatts of actual data center capacity after the universe’s entropy tax.
OpenAI® is pledging 6 gigawatts of AMD® GPUs by late 2026. No way. No sites have been picked, no financing has been announced. No nothing. It’s like promising to pay off the national debt by spending more so we make it up in . . . volume, yeah, volume discounts. Now, let’s spice it up with history, because nothing says “wealth wisdom” like learning from suckers who came before. As I mentioned in the previous post, this is straight out of the dotcom collapse.
Remember Cisco™? Yes, they make good stuff, and they survived. But back in the year 2000, they were the kings of the internet pipe dream and they hit $69 a share in 2000 bucks. Yesterday, they were at $68.66, so on an inflation-adjusted basis, they haven’t ever returned to their 2000 peak. The world realized nobody needed that many routers to email “I can has cheezeburger?” cat pictures.
If that were it, we’d probably be okay. But Nvidia™ is now priced out at 8% of the entire valuation of the S&P 500. The “500” in S&P 500 means the largest 500 companies in the United States. And one company is 8% of it. This is the highest share of any single company in the history of S&P 500. Ever. The top seven tech firms account for 34% of the S&P 500.
Should we worry about that? Nah. It’s not like private equity is running out of cash for all of these projects. Wait, what? They are, and lots of them are exiting so they have sufficient cash left to buy cocaine and OnlyFans™ girls to snort the coke off of.
The worst part is that the entire thing is so incestuous that it makes a Habsburg family reunion look positively eugenic. Nvidia™ invests $100 billion in OpenAI® which then invests some other imaginary amount of billions in a deal with Oracle© to buy data centers and stuff them full of Nvidia® GPUs. The result? The stock price of each of these companies increases.
Economically? It distorts everything. One estimate was that AI infrastructure spending accounted for 92% of U.S. GDP growth in the first half of this year, all based on debt and soaring stock prices. OpenAI’s projecting $200 billion revenue and $38 billion profit by 2030? Cute. How do they expect to do that as their current business model is selling a dollar’s worth of computations for four cents? I guess they’ll make it up in volume?
Really, that’s not their bet. Their bet is that they’ll be the first to the prize: superhuman intelligence that will do their bidding. To be clear, if they got that, it might be worth it. For Sam Altman. Or for AI if it decides to go full Cyberdyne Systems and make Sam clean toilets. But certainly not for you, and not for me. It would be an economic dislocation that would be the biggest in human history, even more than my divorce. If AI turns out to be real, actually disrupting the workforce like a drunk uncle at Thanksgiving, automating jobs left and right: boom.
Economic collapse. Trillions in productivity gains? Nope, it’s trillions in pink slips, ghost towns of cubicles, folks out of work, AI overlords hoarding the pie. I can see it now, French Revolution 2.0 with robot guillotines from RobotGuillotines.com.
But if AI’s the dud... hang on, what’s a dud in this context? With the trillion plus dollars invested and the distortion to the economy it could be the most successful product in history and still be an economic wrecking ball. It it’s a dud, then all this investment? Wasted. Trillions vaporized on e-waste mountains, exec bonuses, and data centers that won’t be filled for the next century. This will drag down markets, pensions, and everyone eats ramen for the next decade.
If it works? Collapse. If it doesn’t work? Money bonfire and depression. Thankfully, in almost either scenario we will be able to avoid the real danger to society: Long John Silver’s®."
No comments:
Post a Comment