Artificial Intelligence

[Spekulatius]

If memory prices stay up for 2 years and then crash back to earth, the memory chip stocks are not worth their current valuations. The value depends on how high memory prices go , how long it lasts and how far they mean revert.

 

I actually think NVDA has some resilience in terms of margins and it’s the better bet here, if you even want to make one. I think for token usage to explode from here, the tokens need to get cheaper which is the way technology works anyways.

Edited Monday at 02:12 AM by Spekulatius

[DooDiligence]

Another banger from Irrational Analysis.

H/T to @gfp for unearthing this guy.

 

https://irrationalanalysis.substack.com/p/the-king-bleeds-long-live-the-king

[Dalal.Holdings]

 

Apple gets to keep making Free Cash Flow while the rest burn it on capex

 

[DegenerateGambler]

what's the total revenue from AI supporting all this capex?  Anthropic + Gemini + OpenAI + optimization from AI for ad revenue maybe somewhere between 150-300 billion dollars?  And the Capex is set to surpass 700 billion this year.  I guess AI could become a somewhat decent margin business someday but Chinese models are already catching up and 10-30% of the token costs.  Is there a trillion dollar business here just based on LLMs?

[mattee2264]

Buffett has given his endorsement to the AI trade claiming he is behind the Google investment.

 

I do not know if that is a contrarian indicator or whether it should be a source of assurance. Presumably he's fine with Big Tech becoming more like utilities able to put large sums of capital to work and earn decent returns.

 

I don't think returns will necessary be from LLMs. There are all kinds of software applications likely to emerge in the coming years and cloud monopolies should prosper in the same way they did when everyone went online during COVID. 

 

[Malmqky]

Buffett is a very smart man, but he's had many misses. Airlines, IBM, etc. etc.

 

I think the worst thing you can do is factor Buffett buying into your investment thesis and decisions. 

 

Google is a double digit portion of my portfolio fwiw.

[Gregmal]

Buffett owns OXY though. 

[Spekulatius]

On 7/14/2026 at 12:32 AM, DegenerateGambler said:

what's the total revenue from AI supporting all this capex?  Anthropic + Gemini + OpenAI + optimization from AI for ad revenue maybe somewhere between 150-300 billion dollars?  And the Capex is set to surpass 700 billion this year.  I guess AI could become a somewhat decent margin business someday but Chinese models are already catching up and 10-30% of the token costs.  Is there a trillion dollar business here just based on LLMs?

The AI model will never be as good of a business than regular software or even cloud, The reason is that the incremental cost to foreach query is actually quite high - it’s the computing cost for tokens. I think I saw that Anthrophics gross margins are only about 40% according to some posters on X. We will find out with the S-1.

It will extremely Important for them to have the most efficient computing stack possible or the economics might look pretty grim otherwise. It’s clear now that the model itself has very low switching  cost at least for current applications like coding.

 

 

Edited yesterday at 05:49 PM by Spekulatius

[Saluki]

CME and ICE are each working on a futures index contract for compute capacity at data centers that use NVDA chips. One index is being created by someone who used to work at Bloomberg, the ICE one is by ORRN (sp?) which is a broker for data centers and people who want to rent computing power for AI startups.

 

If you think AI or chip companies are overvalued but you have an opinion on the amount of Chips and energy that is going to be needed by data centers. This might be a way to play it. Both of them have submitted the contracts for approval by the cftc but they are both still pending 

 

Kalshi lets you bet on computing capacity now, but in a really stupid way. You have to buy a bunch of yes, no prediction market contracts at different prices and for your own forward pricing curve. 

[Peregrine]

I doubt any of the hyperscalers are even thinking about ROI for AI. They're doing so because they think that they have no choice cause everyone else is spending with their heads cut off. Well, except for Apple which may end up in better shape than everyone else in the aftermath of this.

[mattee2264]

I think compute is the way to look at it. It is a commodity so demand and supply apply. 

 

At the moment demand is pretty much infinite because AI is so compute intensive in its current guise and it is being massively subsidised for its users (with OpenAI Anthropic losing billions a year). Training the LLMs and developing new AI related software requires a lot of compute. Companies are also encouraging their employees to use AI i.e. tokenmaxxing. There also hasn't been a shakeout yet so there are a lot of startups using compute in many cases being subsidised by Nvidia or the hyperscalers and this is all supporting demand. 

 

Also many of these datacenter projects are in the pipeline so haven't gone live and aren't yet affecting supply. Nor are they hitting P+L of hyperscalers because they are accounted for as construction in progress. Their legacy cloud businesses are doing incredibly well because the AI buildout is creating a lot of extra cloud demand but supply. 

 

A lot of people are arguing that AI is a FUNDAMENTAL bubble. In other words people are extrapolating fast growth which is unsustainable. Valuations aren't as crazy as during the dot-com bubble and that is giving people a false sense of security. 

 

It is a build it and they will come approach. 

 

The problem could arise a few years down the line when supply/capacity is greatly expanded as a result of these massive investments but demand hasn't increased as fast as expected. 

 

Most of the ROI figures only really make sense if you assume either massive productivity improvements or significant job displacement (but in the latter case if that happens as fast as the hyperscalers need it to happen then the economy is in trouble because consumption represents the majority of GDP and people without jobs generally don't spend much)

 

 

[Spekulatius]

WTH is Kimi moonshot?

 

 

[LC]

https://www.thedrive.com/news/inside-the-flock-dragnet-how-systemic-errors-led-to-police-ambushing-me-for-no-reason
 

Story sort of represents the approach society has towards AI