Artificial Intelligence

[Eldad]

Given the human propensity for mania and the horrible track record of almost all expert predictions, I think it makes the most sense to bet that the world will go on as it always has. 
 

I have harped on this before but all of these people are thinking about this in a vacuum.
 

What happens politically when the global populace turns on it? AI terrorist groups will pop up all around the world if the future is anything like these people claim. 
 

What happens when US based and Chinese based AI systems endlessly hack each other potentially making all cypher security worthless and the usefulness of any of this negated. 

 

How will AI ever have agency to do anything on its own? 

Edited October 27, 2025 by Eldad

[Milu]

Just now, Longnose said:

Whats truly magnificent is that despite the constant growing of CAPEX. The returns on said CAPEX stayed strong and grown with it. 

 

Yes I've been hearing about how all this capex money is being wasted and how some of the smartest CEOs of our generation are just blindly following setting billions on fire like lemmings, yet so far returns on capital are holding steady if not increasing. 

[rogermunibond]

I would bet that for Oracle and Softbank and probably Nvidia (vendor financing) the returns are very very ambiguous.

 

IMO, Google and Meta have the best returns on AI capex because they have the best businesses that can monetize AI currently.

 

Maybe in the future that changes some as the AI investment moves from the cloud to the edge, where latency is more of an issue, say for autonomous vehicles.  Or if more of the AI compute goes to the device.

[Dalal.Holdings]

45 minutes ago, Longnose said:

Whats truly magnificent is that despite the constant growing of CAPEX. The returns on said CAPEX stayed strong and grown with it. 

 

 

Looking only at gross margins ending in 2024 will not give you an answer. It's operating margins going fwd that matter relative to capex spend

[Longnose]

22 minutes ago, rogermunibond said:

Maybe in the future that changes some as the AI investment moves from the cloud to the edge, where latency is more of an issue, say for autonomous vehicles.  Or if more of the AI compute goes to the device.

Hmm... yea bit different graphs there. on  ORCL and NVDA 

[Spekulatius]

1 hour ago, Milu said:

I would suspect that many of these companies revenues and earnings are up 10 to 20x over the last decade too, so not sure how much signal you can get from these charts.

Well FCF for META is heading to zero next year.

 

These tech plays will have an income statement that resembles utilities.

[Dalal.Holdings]

Surely the more money you invest, your returns must go up, right? Isn't that why Buffett says Berkshire's returns should get better as BRK gets larger ? 

 

This is also why capital intensive businesses are superior to asset-light ones, amirite ?

 

Edited October 27, 2025 by Dalal.Holdings

[Longnose]

The problem with BRK and Utilities is they are spending CAPEX on non income producing assets.  Whereas a large portion of the CAPEX in GOOG, META world have potential to make significant returns. I feel every year their CAPEX has been at an all time high and it still continues to product more value for every $$ invested than before. 

 

If a business can invest a Dollar and produce 2-3. why would you stop? These businesses haven't proven that they've hit a wall on income producing CAPEX yet. Will it get there?? eventually? are we there yet? that's where uncertainty is at its highest. So pick your play.  UP, DOWN, Sit on the side and watch the show? 

 

I have a large bet on GOOG at the moment. They gonna continue to expand that TAM and the CAPEX is enabling that. That's my bet. 

[Milu]

54 minutes ago, Spekulatius said:

Well FCF for META is heading to zero next year.

 

These tech plays will have an income statement that resembles utilities.

Yes it could very well be, but if the people inside Meta see the ability to deploy that cash flow into things with high returns wouldn't you prefer that they deployed it all rather than doing buybacks or paying back debt etc. For example if chipotle determine they can make a 400k year one cash return on a 2 million dollar cost of opening a restaurant wouldn't it be smart for them to invest as much free cash as they have the sufficient bandwidth to handle?

 

I think many people in the value investing community seem to think not having any free cash flow for a year or two is some terrible thing whereas in reality if the company has ability to deploy cash into high return investments it is one of the best things.

 

Obviously this all depends on the actual returns that we end up with, but just assuming zero or negative free cash flow is bad isn't the best way to assess things in my opinion.

[Dalal.Holdings]

2 minutes ago, Longnose said:

The problem with BRK and Utilities is they are spending CAPEX on non income producing assets.  Whereas a large portion of the CAPEX in GOOG, META world have potential to make significant returns. I feel every year their CAPEX has been at an all time high and it still continues to product more value for every $$ invested than before. 

 

If a business can invest a Dollar and produce 2-3. why would you stop? These businesses haven't proven that they've hit a wall on income producing CAPEX yet. Will it get there?? eventually? are we there yet? that's where uncertainty is at its highest. So pick your play.  UP, DOWN, Sit on the side and watch the show? 

 

I have a large bet on GOOG at the moment. They gonna continue to expand that TAM and the CAPEX is enabling that. That's my bet. 

 

This is a wild take. BRK is spending capex on things that don't produce income?

 

So you think all the capex META and GOOG are spending will make significant returns. Well, your view is not very different from the market. A lot of those lofty expectations are priced in. Good luck cashing in on them

[weighingmachine]

Surviving the AI Capex Boom

https://blog.sparklinecapital.com/wp-content/uploads/2025/10/sparkline-ai-capex.pdf

[Longnose]

4 minutes ago, Dalal.Holdings said:

 

This is a wild take. BRK is spending capex on things that don't produce income?

 

So you think all the capex META and GOOG are spending will make significant returns. Well, your view is not very different from the market. A lot of those lofty expectations are priced in. Good luck cashing in on them

Eh... i massively oversimplified. BRK makes money on its CAPEX. But I would argue GOOG and META make more money on their CAPEX. 

 

Which IMO would be why Warren is sitting on such a mountain of cash. He wants to deploy it where he can make more returns. If he felt any of his current businesses could ingest more cash im 100% confident in Warren/Ajit/Greg/Management to deploy it into those businesses. 

 

But im not much of an arguer so ill move on from this discussion

[winjitsu]

On the recent Meta / Blue Owl $30bn datacenter, I thought this tweet was very insightful. You have Meta moving to FCF break-even and even enlisting SPVs to build even further, with what I consider to be larger and larger off-balance sheet liabilities. 
 

 

Edited October 27, 2025 by winjitsu

[benchmark]

3 hours ago, Spekulatius said:

Well FCF for META is heading to zero next year.

 

These tech plays will have an income statement that resembles utilities.

How did you get FCF to be near zero for META next year?

[Spekulatius]

13 hours ago, benchmark said:

How did you get FCF to be near zero for META next year?

It’s not zero next year but much much smaller. Tikr (which is some analyst consensus ) forecast for 2026 $126b in a cash from operations and ~$99B in Capex. I am not sure if this includes these off balance sheet vehicles or not.

 

I think datacenters have a structural component (building, utilities) and a tech component (servers etc). The latter depreciates fairly quickly - in four years or so, it’s worthless. So, the overall depreciation rate for the AI datacenters will be very high and they need to be refurbished completely every couple of years. This will permanently change how the income stats to looks for those tech companies that participate in the AI arms race.

 

So now the big question is how much incremental EBITDA will META generate from their ~$100B in annual Capex spent. Hopefully more than from the Metaverse. They need about $20B in incremental Ebitda just to break even on those investments (imo).

.

[Minseok]

I didnt run the numbers myself and i forget the podcast episode but i think in aggregate the cloud providers are using a conservative ratio of cashflow for capex. Enough margin to pay for buybacks and dividends with spare. Something like 15-30% of cashflow.
 

If anything this suggest they are saving the full throttle for later when they really feel the fomo.

 

it could have been from one of:

The real eisman playbook

compound and friends

[rogermunibond]

Sam Altman now saying that OpenAI IPO most likely given capital needs.  Hold on to your hats.

 

 

[Spekulatius]

Quote

Based on recent announcements, here is the power consumption information for OpenAI's current and planned data centers:

• Total planned capacity: OpenAI CEO Sam Altman has floated a vision for as much as 250 gigawatts (GW) of compute capacity by 2033. For comparison, this amount is equivalent to about one-third of the peak power consumption of the entire U.S..
• Initial large-scale project: In September 2025, OpenAI and Nvidia announced a partnership to build and operate AI data centers with an initial capacity of 10 GW. According to a Cornell University professor, this is more than the peak power demand for countries like Switzerland or Portugal.
• Stargate projects: The OpenAI and Oracle Stargate project includes additional data centers. Recent reports suggest five new sites may total 7 GW of power.
• Existing large data center: The Stargate data center in Abilene, Texas, already uses enough electricity to power half a million homes. 

This immense power usage, which is driven by AI's high computational demands, is a significant leap from traditional data centers. 

 

[Spekulatius]

Quote

Based on recent announcements, here is the power consumption information for OpenAI's current and planned data centers:

Total planned capacity: OpenAI CEO Sam Altman has floated a vision for as much as 250 gigawatts (GW) of compute capacity by 2033. For comparison, this amount is equivalent to about one-third of the peak power consumption of the entire U.S..

Initial large-scale project: In September 2025, OpenAI and Nvidia announced a partnership to build and operate AI data centers with an initial capacity of 10 GW. According to a Cornell University professor, this is more than the peak power demand for countries like Switzerland or Portugal.

Stargate projects: The OpenAI and Oracle Stargate project includes additional data centers. Recent reports suggest five new sites may total 7 GW of power.

Existing large data center: The Stargate data center in Abilene, Texas, already uses enough electricity to power half a million homes. 

This immense power usage, which is driven by AI's high computational demands, is a significant leap from traditional data centers. 

Wo sounds like openAI alone has announced datacenters that are going to consume 17 GW (there are more vague announced  but they are further out). To generate 17GW, one needs 17 medium sizes but else power plants. 17GW is enough to power about 12.75M homes. Thats just the OpenAI and there is Amazon, xAI, Anthrophic, Google, Meta etc. 

 

So when I joke that big tech soon looks like I am not joking because basically they will need to build a giant utility. 
 

Does it really make sense to build that much infrastructure for AI. What if China or the players will create AI that is 80% s good but 5 or 10% of the cost ? Who then really will be the winner?

 

Also what will power prices do with to the real economy including household power bills? OpenAI is not going to build 17 nuclear power plants. If they don’t run out of money they will contract for power crowding out other consumers (business or consumers)

 

How is all this going to work?

Edited October 29, 2025 by Spekulatius

[Eldad]

29 minutes ago, Spekulatius said:

Wo sounds like openAI alone has announced datacenters that are going to consume 17 GW (there are more vague announced  but they are further out). To generate 17GW, one needs 17 medium sizes but else power plants. 17GW is enough to power about 12.75M homes. Thats just the OpenAI and there is Amazon, xAI, Anthrophic, Google, Meta etc. 

 

So when I joke that big tech soon looks like I am not joking because basically they will need to build a giant utility. 
 

Does it really make sense to build that much infrastructure for AI. What if China or the players will create AI that is 80% s good but 5 or 10% of the cost ? Who then really will be the winner?

 

Also what will power prices do with to the real economy including household power bills? OpenAI is not going to build 17 nuclear power plants. If they don’t run out of money they will contract for power crowding out other consumers (business or consumers)

 

How is all this going to work?

Exactly. The brain can run on a cheeseburger for probably 72 hours, it’s a lot better than this junk. 
 

But seriously, this should be enemy number one of the greens. I guess we now know why Bill Gates is no longer talking about sequestering cow farts. 
 

This is going to blow up spectacularly. I can’t wait. 

[adventurer]

Interesting take shifting the risk from a "normal" stock market bubble to a more geopolitical risk. 

 

Also: https://www.wsj.com/economy/jobs/white-collar-jobs-ai-324b749c?mod=hp_lead_pos1 I am in my early 30s and I of course did know the risk for these layoffs. But the pace is scary and this does look like a Loolapalooza with many different perceptions involved. The outcome appears to be very hard to predict if you want to account for political counter measures too. I doubt people will let that go through just like that. 

Edited October 29, 2025 by adventurer

[Red Lion]

15 hours ago, Spekulatius said:

So when I joke that big tech soon looks like I am not joking because basically they will need to build a giant utility.

 

I suspect this all gets financed by private equity in one form or another. Brookfield, Blackstone or KKR is going to step in and build generators and have these datacenters/AI companies sign long term take or pay contracts. 

 

Or those same companies will buy the assets off the data centers/AI companies and do sale/leaseback type transactions. 

 

It doesn't make a ton of sense to create these utilities on the the tech firms' balance sheets, but it makes plenty of sense that LPs in infrastructure funds (endowments, universities, sovereign wealth funds, and maybe even 401ks the way things are rolling out) end up financing the majority of this buildout (whether up front or by sale/leaseback).

[Broeb22]

16 hours ago, Eldad said:

Exactly. The brain can run on a cheeseburger for probably 72 hours, it’s a lot better than this junk. 
 

But seriously, this should be enemy number one of the greens. I guess we now know why Bill Gates is no longer talking about sequestering cow farts. 
 

This is going to blow up spectacularly. I can’t wait. 

Bill Gates wanted to sequester cow farts? Seems a worthwhile endeavor. 

 

If AI is really "successful", then there will be less humans and there should be less cows whose farts need to be sequestered, unless the AI successor race develops a taste for hamburgers, but then again they'll probably figure out how to make cows who don't fart as much because AI is superior to humans. So many worthwhile questions, not enough time.

[Eldad]

2 minutes ago, Broeb22 said:

Bill Gates wanted to sequester cow farts? Seems a worthwhile endeavor. 

 

If AI is really "successful", then there will be less humans and there should be less cows whose farts need to be sequestered, unless the AI successor race develops a taste for hamburgers, but then again they'll probably figure out how to make cows who don't fart as much because AI is superior to humans. So many worthwhile questions, not enough time.

Yes he invested in a company to bottle their farts and he bemoaned all of their methane production constantly and wanted us to eat fake meat. 
 

Now With AI data centers set to make cows look like a drop in the greenhouse gas bucket, he has predictably pivoted. 
 

Meanwhile you have Peter Thiel giving lectures on why Greta and other anti-AI greens are the “antichrist”. 
 

It never occurred to me to be green before but I might have to join them. Seems like the more rational team at this point. 

[NnnnotSoSmart]

1 hour ago, Red Lion said:

I suspect this all gets financed by private equity in one form or another. Brookfield, Blackstone or KKR is going to step in and build generators and have these datacenters/AI companies sign long term take or pay contracts. 

Watch CNBC’s full interview with Brookfield CEO Bruce Flatt

 

Bruce Flatt, Brookfield CEO, joins ‘Squawk on the Street’ to discuss the company’s $80B deal with the U.S. government, the future of nuclear and much more.

 

Flatt comments on the AI data center build out.

(starts talking about data centers at approximately 4:00 min)

"We're not in an AI bubble" 

"We can't build enough of them."

"This can't be done by everyone." - Huge amounts of capital required.

To countries outside US: "If you don't build AI infrastructure in your countries, your companies will leave."

 

https://www.cnbc.com/video/2025/10/29/watch-cnbcs-full-interview-with-brookfield-ceo-bruce-flatt.html

 

 

Edited October 29, 2025 by NnnnotSoSmart