The Quiet Bull Market in Tin
and other AIAIAIAIAI Musings
I saw 3 tweets on AIAIAI that triggered me to write this piece:
Tweet #1 from tech analyst Patrick Moorhead highlights that Google disclosed that it does its Gemini training on its TPUs. He wasn’t sure if it was “all of it” or “part of it”. He later got confirmation from Google that 100% of Gemini was trained on the TPU.
For the Boomers among you (like me), that means that Google is NOT using Nvidia GPUs for training Gemini. This was apparently out in December 2023 so maybe some of you knew this, but I didn’t. After all, the prevailing knowledge on Wall Street is that Nvidia is the God-like company that will win 100% of the AI market. I wrote “A Boomer's review of Nvidia's results - Part 1” a month ago and I think it’s time for Part 2 soon.
Tweet #2 claimed that Druckenmiller sold his NVDA position (unverified).
I can’t say I’m surprised! Back on March 10th, I wrote “Contagion” where I laid out the checklist to identify the end of a “Monster Rally”. Nvidia met all the criteria and had the ugliest weekly candle I recall. Druck isn’t one to fall in love with positions when the setup changes. Here’s what I wrote:
Tweet #3 is from Gavin Baker (by now you know that I’m a Gavin fan, I mean the guy named his fund “Atreides”. Respect). Gavin says that “AI exposure is most overvalued in Industrials by a wide margin” and that “AI exposure is cheapest within Tech”
Gavin is absolutely right about “AI exposure” being overvalued in certain sectors. The monkeys are trying to play 3rd derivative plays in AI and there are many “fake AI” plays with stupid multiples (In my piece “Tale of Two Countries”, I was making fun of the rivalry between Databricks and SNOW, where the Databricks CEO accused SNOW of “not doing AI”. Given one trades at 27x Sales and other one at 17x Sales, they should probably both keep quiet and enjoy their AIAIAI premium!).
Gavin adds that “AI exposure is cheapest with Tech”.
This is where I come in to add my 2 cents, because there’s a tiny segment of “AI” that I wrote about that was dirt cheap at the time and remains cheap :)