Post

Newsletter: We need to talk about CoreWeave, the first big IPO of the generative AI bubble, with a speculated $35bn valuation, burdened by ruinous debt, impossible-to-meet capacity obligations, and a data center partner that's never built any AI compute. www.wheresyoured.at/core-incompetency/

wheresyoured.atCoreWeave Is A Time BombSoundtrack: EL-P (ft. Aesop Rock) - Run The Numbers In my years writing this newsletter I have come across few companies as rotten as CoreWeave — an "AI cloud provider" that sells GPU compute to AI c...

137 reposts729 likes

Their primary moat appears to be “we got dibs on the new chips”. And the services are so amazing OpenAI needs $350m of free diluted stock to get a deal signed. The founders clearly have figured out that the music is stopping, and to cash out now. Bag holders line up.

In particular: "In 1997 and 1998, and for the six months ended June 30, 1999, 39%, 79% and 87% of our revenue, respectively, was generated under our master [...] agreement with Microsoft. [...] We do not believe that we could replace the Microsoft revenue in the short- or medium-term."

Securities disclosures (like an S1) can be very revealing of problems with a company, but more with the numbers than with the "risk" commentary. Every company (even Apple) tells you they're on the verge of catastrophe, it's not so much "honesty" as ass covering. "You can't say we didn't warn you!"

Just extending the Apple example (because they're the easiest "very healthy and rich company"), their latest annual report linked below goes on and on about "risk factors" from pages 5 to 17. It buries anything that might be a genuine risk, and makes it functionally useless to investors.

sec.govaapl-20240928

I'm saying this is not them "dirty laundry", it's just generic risk language. People writing the disclosures literally just copy those paragraphs from others in the same industry . One of the quotes below is from an Astera Labs S1 from Feb last year, the other from CoreWeave's S1.

If genAI was the huge business that boosters keep claiming, CoreWeave ought to be making money hand over fist even if it's run by a tribe of drunken baboons. The fact that it's not means that either genAI is a bust as a business or NVIDIA is using its monopoly to pocket 100% of the profits.

Whats interesting is that their VP of operations says on linkedin he's been working there for 7 years. Do you think they were always building AI infrastructure? What have they been doing all this time?

There's an old movie, I can't remember which one... there's a part where the narration chronicles the euphoria of the Roaring 20's and then Black Friday and the Depression, and it's illustrated by people watching a tower of coins grow toward the sky... and then suddenly the tower of coins collapses.

Beautiful and well written as usual. Can I be a pain and ask to correct “Principal” instead of “Principle”, as I don’t know why, but so many spell it wrong, and makes things confusing for non-native english readers? Ta! 😅

"By comparison, Microsoft bought 485,000 GPUs in 2024 [..] Meta likely has somewhere in the region of 600,000 GPUs, [..], Amazon has hundreds of thousands of its own." and Google was proud to not be in this paragraph

That part about core """scientific""" is like wild, like it doesn't make any sense to even do. Is the American economy just a bunch of assholes that know each other passing debit around like it's a hot potato? I'm starting to think so