The current AI industry stack:
1. Mostly monopoly or oligopoly compute companies, from ASML to NVDA to MU etc.
who mostly sell to,
2. Three serious hyperscalers.
who mostly sell to,
3. Two serious model companies.
who sell to,
4. All sorts of companies, large and tiny.
What seems clear:
1. Is most vulnerable to a boom/bust cycle, since revenue can collapse 80%+. But stands to win the most if compute demand stays strong. If demand doubles from 2029 to 2032, they are still very cheap. If demand maintains 2029 level long term, they are reasonably priced or a bit cheap. If demand collapses, they will collapse.
2. Has good upside if compute demand stays high, low downside given long term contracts. Heads they are worth double+, but don't benefit more than that from an ever increasing demand scenario. Tails they tread water for a few years as they work through over investment.
3. Are an extremely binary bet which comes down to, "Is it AI hardware improvement driving most AI gains, or will rearchers/other going to drive it?" If the former, these are going to be terrible commodity businesses. If the latter, they will become the most valuable companies in the world and it's not close. I think the bullish odds are very low, but if someone disagrees these are a *screaming* long.
4. Seem screwed generally.
Importantly: it's very possible that there's both:
1. Gets hit badly in a boom/bust cycle
and
3. Researchers end up driving the big improvement in AI, in which case they'd actually beneift from the boom/bust cycle, as the game would likely be winner takes all. We can already see this where the very best model has huge pricing power ... I just don't think this is durable at all for lots of reasons.
I guess also: 3 could actually cause 1.
If one frontier company starts getting really far ahead, and it is self improving that way, everyone else gives up. Now this frontier company sets the terms for everyone else in the stack, and prices crater big time.
The super cyberpunk future is 3 wins and their spin offs wind up becoming all the most valuable companies. But yeah, it just seems far fetched.
Say that Anthropic pulls ahead big time in the next 6-12 months though... I'll start changing my mind and get long even if the stock is already $3 trillion market cap or whatever.
Thus far, all evidence points to these being awful commodities though IMO.
Hardware demand will eventually collapse if model companies wind up being a crappy competitive business.
Hardware companies will also eventually collapse if model companies wind up being a hardcore monopoly or duopoly because they are self improving.
So what is the bull outcome for hardware? A moderate outcome, I suppose, where models end up competing in smaller niches longer term.
Maybe each path isn't *that* narrow...?
Share this Scrolly Tale with your friends.
A Scrolly Tale is a new way to read Twitter threads with a more visually immersive experience.
Discover more beautiful Scrolly Tales like this.