Fuck AI

If we know it is coming, can we short it ourselves and get rich I wonder?

It's different this time.

... says every bubble, ever.

Having lived through the dot-com bubble in real time as a professional, allow me to draw the following comparisons:

You know all those huge figures about investments you hear? That’s not debt and low interest blending. That’s real cash investment.

This couldn't be farther from the truth. Here's a great read, but the long and short of it is that this is entirely based on debt and smoke-and-mirrors.

they’re booked out before the ground is broken

On the premise that they will be needed, and that the companies ordering them will be solvent when they're completed. These big AI companies have limited runway, and the data centers they need won't be available until after the runway is gone.

Data centers and power generation are IRL investments that are needed whether it’s for AI use or not

Power generation, maybe. That's like all the fiber that MCI laid before they went tits-up in an accounting scandal of their own making, due entirely to creating a thing that nobody needed yet. Yeah, it's useful now (several decades later) but it was speculatively created without a way to pay for it. These ai-specific data centers cannot be repurposed for regular servers, short of gutting the whole thing. Add to this that the chips used for LLMs cannot be repurposed for anything else, and are practically obsolete by the time they're installed.

Also, there aren’t a smaller set of goods everyone is after, like paying $1 million for toys.com, a URL never worth anything close to the hype.

There are dozens (probably way more) of new "AI" companies being created every day in the hopes of being scooped up by the big players. This is exactly the same as the dot-com boom. Create a cool-sounding company with an idea and a shell of a copycat idea, sell to rube investors, rinse and repeat. This is just as bad, if not worse than it was before.

2 of them are solid and so far 2 are not anything like a bubble ready to pop.

Which 2 of them are solid? Even if that were the case, one leg will come out from the stool (or 2 or 3) and the whole thing will topple. They're all dependent on each other with fake money and promises, and it's a matter of time. I'm saying sooner rather than later, but who knows.

The datacenters being built are designed bespoke for this generation of LLM based ai. They're not really useful for anything else like serving webtraffic or normal compute nodes. Once the next generation of AI comes out that shows LLMs are just next-best-word-guessers, those datacenters will be worthless.

LLMs are useful for some things. I think they're great as a replacement for googling "what do the reddit comments say about this product" or "how do I use this api". But it doesn't think. It cannot logically reason about things because it doesn't actually know anything. It just has probabilistic representations of data embedded in its weights and biases.

Your point about the bubble bursting on debt vs real capital is correct though. The money is already spent and it already existed - it wasn't financial products that were based upon debt that were turned into a capital product. So when it bursts, it will be more of a fizzle and it won't tear down the entire economy. The failure will still result in many bankruptcies and layoffs but not all at once.

I can't say I agree with most of your details, but your second and last sentence are on point. This bubble (or bubble conglomeration that you describe) are not like any previous bubbles and in the end we are most certainly going to have a couple survivors still standing and take everything. The main thing I disagree with is that it will be the only thing that happens. We're going to see a massive gutting of the financial markets and although datacenters are real things that get used, we don't need this many specifically power hungry GPU/AI designed datacenters if the AI market pops. All these AI services are massively lossleading hoping to be one of the survivors. I think that once we get even close to the "pop" we're going to see desperate price hikes and even the bigger companies are going to raise their pricing along with it. I suspect that many of those customers are going to have to admit that their AI plans aren't living up to what they anticipated and will heavily curtain their usage once the prices jump.

Data centers and power generation are IRL investments that are needed whether it’s for AI use or not.

Well said, and I think a good comparison (and something Doctorow has brought up before) is that the dot-com bubble built a ton of infrastructure and fiber networks that were later able to be utilized. He sort of touched on this when he said "Plan for a future where you can buy GPUs for ten cents on the dollar".

That all said, as Ed Zitron is oft to highlight, something like 30% of the value of the SNP 500 is in "AI" right now. Even a relatively small pop could lead to more widespread repercussions.