Francis Turner

Meet CoreWeave, the AI industry’s ticking time bomb The debt-laden AI compute company has been boosted by Nvidia.

@edzitron.com I feel like I've read this article before. I mean you have written about coreweave multiple times. Yet weirdly your name fails to appear
www.theverge.com/ai-artificia...

The official filing documents are going to be fascinating

This Is How the AI Bubble Will Pop The artificial intelligence boom is the most important economic story in the world. But the numbers just don't add up.

Another person has figured out that LLM AI is a bubble - www.derekthompson.org/p/this-is-ho...

This is insane. @edzitron.com has repeatedly pointed out that OpenAI is a money incinerator. There is no way that anyone can justify that valuation

Or one that decides to remove your right to do/get something for posting the wrong thing on social media....

What happens when your Digital ID gets hacked (or updated by a nasty government) and your right to work, benefits etc. is removed?

The fundamental problems with Digital ID all revolve around the failure modes. What happens when there's a security breach? when there's a bug in the system? when your phone (and its associated digital ID token) is stolen? etc. Including of course, what happens when the government is evil?

More likely the total global pot of money that can be spent on LLM tops out at around $100B/year. And $100B/year won't let any one be actually cashflow positive based on @edzitron.com's numbers

And it depends heavily on everyone globally seeing value in replacing all applicable intern/entry level jobs with LLM. There's a lot of IFs in that.

In other words, it ain't going to happen.

If all companies and governments do indeed fire all their relevant workers and use all the money saved to pay $300B to OpenAI then OpenAI's numbers work, just. BUT that means that everyone else gets to throw away their AI investments.

Extend that to Europe/Japan etc. Add in the academic market and the translation market etc. It's really hard to see how LLM can get more than ~$300B/year total in revenue because there aren't enough salaries that they can replace to get more

The back of the envelope sums suggest that LLM can replace about 4M US jobs at an average of $40k/year salary levels. That works out at a total available pot of money of ~$160B/year for the US.

Excellent (long) article that digs into the incestuous money burning world of LLM AI. About the one thing it doesn't mention (good thing considering the length) is, assuming LLM AI is useful, calculating how much can people pay for it. Not will. Can.
Something I tried to answer in this substack