Bubble, bubble, toil and trouble…
Hello, my friends. Have you been feeling too sane lately? Have I got something for you! It is a company called CoreWeave.
You may not have heard of it because it’s not doing the consumer-facing part of AI. It’s a data center company, the kind people talk about when they say they want to invest in the “picks and shovels” of the AI gold rush. At first glance, it looks impressive: it’s selling compute, the hottest resource in the industry; it’s landed a bunch of big-name customers such as Microsoft, OpenAI and Meta; and its revenue is huge — $1.4 billion in the third quarter this year, double what it was in the third quarter of 2024. The company has almost doubled in share price since its IPO earlier this year, which was the biggest in tech since 2021. So much money!
But as I began to look more closely at the company, I began feeling like I’d accidentally stumbled on an eldritch horror. CoreWeave is saddled with massive debt and, except in the absolute best-case scenario of fast AI adoption, has no obvious path toward profitability. There are some eyebrow-raising accounting choices. And then, naturally, there are the huge insider sales of CoreWeave stock.
The current AI craze is looking like the dot-com bubble of 2000. It’s not that the technologies are bogus. It’s that the investors are abandoning traditional value metrics, and driving up prices for AI stocks far beyond what can be justified in even the most optimistic scenarios.
Does this mean a reckoning is coming? No idea, but I’m keeping clear of the tech sector. I’ll sleep better.
My next band will be an easy-listening combo with a girl singer, called Dot Calm and the Bubbles.
The technologies are in fact bogus, in the same way cold fusion is bogus. Absent a massive improvement in IC design (which could well come about in ten years or so) it is physically impossible to run any LLM without spending orders of magnitude more money than it is possible to make from the result.
I have no idea how economic it is to run a LLM, but I can’t call it bogus. The technology does work, although the precise extent it will be useful remains to be seen.
I come from an industry (geophysical exploration) whose vision often exceeded its compute capabilities. As an example, prestack 3D migration was a pipe dream in the early 1980s. We knew how to do it, but it took years of computation. But with time, those hardware restrictions gave way.
I figure the cost of LLM will also slowly melt away as hardware and algorithms improve.
LLM’s currently have a roughly 30% hallucination rate, rendering their output unreliable for many purposes. Right now they are search assistants at best. Hard to justify trillions worth
I’ve had good results with Grok.
You don’t want to automatically trust everything it says. You want to check its claims from time to time. But I find it rarely goes seriously off track, and is always informative in some manner.
I bought $60 puts on Coreweave a week ago, I like the direction of things. A forensic accountant I follow summed it nicely: Jensen’s (NVIDIA CEO) customer’s customer is Jensen. Coreweave is the customer. This stinks as bad as Nortel ever did. My broker called when Nortel hit $60 on the way down from $120. I looked at the financials and said call me back at $6. He said I was crazy. By $6 he wasn’t a broker anymore.
I’d do NVIDIA puts, but they are so expensive, smart money already knows trouble’s a comin.
I read somewhere [Bloomberg] that data centers need to be redone every 3-4 years to keep up.
In the case of AI, 3 years is your best case, and that’s assuming that your center has sufficient power to support the upgrades. The new B200 cards consumes 1000W compared to 700W to the old H200/H100 cards, and that was a period of about 6 months.
META changed their depreciation on the GPU’s from 3 years to 6 years, halfing the run rate on ammortization. This is a $US12 Billion per year inflation of earnings over the next 3 years. The stinky stuff is everywhere.
Weird choice, but whatever works for them… Typically you want to depreciate a quickly as possible for the benefits.
L – Don Ho sings tiny bubbles. Every investor’s computer should sing this every morning.
https://www.youtube.com/watch?v=mlCiDEXuxxA
Do they have a sock puppet spokesperson like Pets.com? Probably.
So many people are desperate to get in on the ground floor … that they end up diving from the 14th floor …
https://youtu.be/5Uocajmnjgw?si=JX7O7DbKfVb4d5r7
When this bubble pops, it’s going to devastate the tech sector orders of magnitude worse than the dot-com collapse wiping out Nortel. And since every company is a tech company now the ripple effects are going to be massive.
Much worse, all the Nortel gear and CISCO gear eventually got used, long life and long obsolescence, and the obsolete bits can be switched out without damage to the rest. The data centers will be obsolete in 3-4 years.