Great article as always. I think there is a great amount of truth in the phrase "shortages become gluts". History certainly supports this as you wrote. However, I really question this study referenced in the Odd Lots story. Everything I read and hear about the AI rollout is we are heading fast into an energy bottleneck. This is due to regulatory and capability headwinds. Having worked in the energy business for 40 years, I understand how hard it is to stand up huge projects. All I see from the CreditSights study is a lot of hopes and promises. I agree we are someday headed to a glut, I just don't see it by 2030.
very likely you have read Michael Burry's thoughts on AI cycle by now. if not you should check it out. you shall find Dr. Burry very much agrees. stock peaks will before the capex cycle peaks.
munger and buffett spoke in many instances of investment/capex into efficiency do not always translates into more profit for the company or its shareholders. in many cases it goes directly to the customers in the form of better, faster or cheaper product. one example they often use is when one department store in downtown Omaha installed an escalator and then its competitor across the street was forced to install one, neither benefitted.
the other is ofc is textile. year after year the original berkshire invested into capex but all the improvement translated into cheaper products. year after year it was forced to go back to "GO".
Sorry about that. I should have said next week. It's coming soon, but I had a few things pop up in the research bucket that required attention. Thanks for your patience
Thanks for taking the time to write this, John. A few things to ponder : (i) is there any technological moat that is unsurmountable for new potential competition regardless of money and therefore might justify a longer runway here for the semiconductor folks (ii) have to agree with the other comments here in terms of passthrough of benefits to customers due to competitive dynamics, but does this make AI a must have down the line to stay competitive and critical for basically everyone (like some SAAS businesses today) (iii) Where would you put the market size of all this? copilot charges ~$30 / month / user for enterprise clients. Even at a high multiple on revenues and assuming 100% margins and assuming only 1/3rd of the Capex figures is actually AI related, its still tough to see even a mediocre ROIC on all this spend.
Thanks for the comment. I think AI is going to grow much larger and become a bigger part of what we do. My question is more about where does the cash flow come from at the end of the day (the law firm getting more clients because of it… example from my post). You can keep going down the line: if the law firm does get more clients, the clients then must be getting additional value somehow. It’s hard to know how it shakes out, but I’m skeptical that there will be enough cash flow to justify what might end up being trillions spent on the infrastructure. It doesn’t mean AI isn’t going to grow, but if supply grows faster, then it becomes a problem for the picks and shovels players. I think supply is going to grow faster than people expect, based on what we’re seeing happening now. On MSFT and Co-pilot, I don’t know. I do have a friend at a software company that suggests they are using it so much that there is no way MSFT is earning profits on their $30 per seat charge. That would imply MSFT would need to raise rates, at least in this case. And then the question becomes does a higher fee still make sense? Hard to say…. will be interesting to see how this plays out next few years. I think there are a lot of risks building up in the providers, but to be clear, I’m very bullish on the future of the technology
John,
Great article as always. I think there is a great amount of truth in the phrase "shortages become gluts". History certainly supports this as you wrote. However, I really question this study referenced in the Odd Lots story. Everything I read and hear about the AI rollout is we are heading fast into an energy bottleneck. This is due to regulatory and capability headwinds. Having worked in the energy business for 40 years, I understand how hard it is to stand up huge projects. All I see from the CreditSights study is a lot of hopes and promises. I agree we are someday headed to a glut, I just don't see it by 2030.
Brilliant in many ways.
very likely you have read Michael Burry's thoughts on AI cycle by now. if not you should check it out. you shall find Dr. Burry very much agrees. stock peaks will before the capex cycle peaks.
munger and buffett spoke in many instances of investment/capex into efficiency do not always translates into more profit for the company or its shareholders. in many cases it goes directly to the customers in the form of better, faster or cheaper product. one example they often use is when one department store in downtown Omaha installed an escalator and then its competitor across the street was forced to install one, neither benefitted.
the other is ofc is textile. year after year the original berkshire invested into capex but all the improvement translated into cheaper products. year after year it was forced to go back to "GO".
Tomorrow will come up the article with the easiest hurdles?😀
Sorry about that. I should have said next week. It's coming soon, but I had a few things pop up in the research bucket that required attention. Thanks for your patience
Excited to see the article with the easiest hurdles.
Thanks for taking the time to write this, John. A few things to ponder : (i) is there any technological moat that is unsurmountable for new potential competition regardless of money and therefore might justify a longer runway here for the semiconductor folks (ii) have to agree with the other comments here in terms of passthrough of benefits to customers due to competitive dynamics, but does this make AI a must have down the line to stay competitive and critical for basically everyone (like some SAAS businesses today) (iii) Where would you put the market size of all this? copilot charges ~$30 / month / user for enterprise clients. Even at a high multiple on revenues and assuming 100% margins and assuming only 1/3rd of the Capex figures is actually AI related, its still tough to see even a mediocre ROIC on all this spend.
Thanks for the comment. I think AI is going to grow much larger and become a bigger part of what we do. My question is more about where does the cash flow come from at the end of the day (the law firm getting more clients because of it… example from my post). You can keep going down the line: if the law firm does get more clients, the clients then must be getting additional value somehow. It’s hard to know how it shakes out, but I’m skeptical that there will be enough cash flow to justify what might end up being trillions spent on the infrastructure. It doesn’t mean AI isn’t going to grow, but if supply grows faster, then it becomes a problem for the picks and shovels players. I think supply is going to grow faster than people expect, based on what we’re seeing happening now. On MSFT and Co-pilot, I don’t know. I do have a friend at a software company that suggests they are using it so much that there is no way MSFT is earning profits on their $30 per seat charge. That would imply MSFT would need to raise rates, at least in this case. And then the question becomes does a higher fee still make sense? Hard to say…. will be interesting to see how this plays out next few years. I think there are a lot of risks building up in the providers, but to be clear, I’m very bullish on the future of the technology