WATCH AI CAPEX

“At the current price this morning, the stock's trading at about $0.85 on the dollar of fair value. We had a chance to buy it a couple of times. So Berkshire stopped buying shares back in 2024. Greg announced a couple of weeks ago after the stock declined post the earnings release that he had initiated share repurchases again in consult with Warren and made sense in that the valuation relative to intrinsic value at least per my calculation was back down to where it was in 2024 when they stopped buying shares back.”

“To argue that the high current margins coupled with very high multiples, you can bake in scenarios for each of the five variables and it's really hard to get to more than a 5% return. And depending on where margins and multiples head from here, you can get to a loss for a decade, which is what happened after the 1999 peak.”

“I'm looking at Air Systems, AEHR, Air Test Systems. AEHR is the symbol. Small cap name, $2,600,000,000 market cap. It has boomed, Luke. It was trading just last April, one year ago. Down around $7 per share. Today, even after dropping 7% today, it's at $82 per share. It's moved, like, 800% in the past fifty two weeks. What about just a month? And then last month, it was trading at 30, and now it's at 82 in a in a month. I'll give you another number. 18.7%. What number do you think that is? Short interest. There you go. So this move is just probably a short squeeze.”

“One of the more peculiar of all the deals is, and this was disclosed in a core we've filing, was NVIDIA has promised to buy any of CoreWeave's service availability that they can't sell to anyone else. That is very unusual. That's not the same as making an investment. That could easily help CoreWeave with their debtors and and getting more debt financing. But it also means as a investor, we we don't know what's going on with real demand for CoreWeave because we probably won't be told if they start moving into the world where they're offloading to NVIDIA or not.”

“The capabilities of Mythos Preview, according to Anthropic and those that have used it are, it's basically like a hacking ray. You can point it at software and even if you don't have the source code, it can still detect vulnerabilities and write code that will exploit it. So a human doesn't need to do anything to break into that software.”

“We are so deeply afflicted by this idea if of, concept of disease, right, which, for people who are not familiar, it's basically this idea that, some sectors such as, you know, nannies or or or childcare or health care doesn't really progress with technological innovation as you would have, you know, with, say, manufacturing. We've always had, good, you know, disinflation through automation of plasma TV. The you track the price has gone up. You know, everyone has seen that chart where a college tuition hospital and and and and health care went up secularly and and and, you know, manufactured goods went down over time.”

“China has just blocked Meta's $2,000,000,000 Manus deal. Meta has already went ahead and acquired this company called Manus, which technically, to be fair, was it was started in China. The founders are Chinese. They realized that China was pretty restricted in this. And so they moved the company to Singapore. When they moved it to Singapore, Meta said, hey. We'd love to acquire it. They bought the company. They paid $2,000,000,000, and they have, like, deeply embedded it.”

“Currently, AI is too cheap, right? People are just out there consuming token because it's heavily subsidized, and they ask the most powerful model for cooking recipe. That's sort of your analogy, except that it's actually being used on the agent level as well, the agents are calling AI models inefficiently. Part of the reason why people run up these huge bills using open claw is exactly because the system has not been optimized for token deficiency, and part of the reason why is because price has not really been reflecting the genuine cost.”

“I'm having my roof repaired. I've got this 100-euro house with beautiful Spanish tiles that are also quite nettlesome when they break. And I've had, it's been a struggle for me since last fall. I've been looking for contractors, and it's quite difficult to find them, right? And when they are busy, they don't answer your calls. They've got terrible websites. When they show up, they don't speak your language.”

“In fact, the the most rate sensitive of the big four banks. Now over the past six months, stocks kind of been all over the place. It's been climbing since March, then had a pretty steep drop off in March. What really has been driving that? Well, a lot of it has been what's been driving the overall banking sector. They have recently had upgrades from an analyst perspective. And out of earnings, they have been performing pretty decently because every segment has grown revenue, grown earnings, deposits have grown, loans, all growing in q one. They had 9,300,000,000 in capital return to shareholders in a single quarter. But at the same time, they're also the most levered of the big four. Each 25 basis point Fed cut reduces Bank of America's NII by an estimated $225,000,000.”

“I mean, of course, it's a bubble. I mean, yes, in the bubble, it was always going to be a bubble. Just because something's a bubble doesn't mean that you couldn't have a very large sustained impact, and especially if you if one that's actually, you know, much larger and longer lasting than people had anticipated. The question is not what to to I think the interesting debate is not whether or not AI was a bubble, will become a bubble. It manifestly was, and have become one. The question is, how long is the bubble? How big could the bubble get, and at what stage are we?”

“Historical patterns in digital assets suggest that Bitcoin cycle lows are projected for late 2026.”

“Then if we look at factory expansion plans, the transition to take the factory that was making Model S and Model X vehicles and take that to produce the Optimus Humanoid Robots is still on track for to enter production by the end of this year. And so with Tesla's two big pillars in the future going to be self-driving cars and Humanoid Robots, both of those initiatives are on track with the positive free cash flow and that led the stock to rise after hours.”

“The capabilities of Mythos Preview, according to Anthropic and those that have used it are, it's basically like a hacking ray. You can point it at software and even if you don't have the source code, it can still detect vulnerabilities and write code that will exploit it. So a human doesn't need to do anything to break into that software.”

“Meta finds itself sort of racing to catch up. Right? And so they were giving out $100,000,000 offers to researchers trying to basically, like, rebuild this team and become competitive. Multi year deals worth $300,000,000. These are for scientists and engineers. This is wild. It's like NBA superstar money. With some of them receiving a $100,000,000 straight up in year one, a $100,000,000.”

“There's a lot to integrity and morality, and either you've got it or you don't. And there's no pay package. There's no compensation structure that's gonna alter behavior in how you are. I think it's unfortunate the way most compensation systems are structured with a modest salary performance bonus that may or may not have hurdles, and then you've got longer term and shorter term hurdles on your performance shares, your restricted shares. There's just too much short termism that comes with the way most comm structures are structured.”

“To argue that the high current margins coupled with very high multiples, you can bake in scenarios for each of the five variables and it's really hard to get to more than a 5% return. And depending on where margins and multiples head from here, you can get to a loss for a decade, which is what happened after the 1999 peak. Two interesting things. One, when Warren talked at length many times about the tailwind that he enjoyed from growth in real GDP per capita. The United States was the economic engine of the world.”

“The implementation of STRC fixes the yield gap with price-stable returns for most institutional holders.”

“Howard, everything you know about investing came from other people. Benjamin Graham taught you about margin of safety. Buffett taught you about quality. Charlie Munger taught you about mental models from multiple disciplines. John Kenneth Galbraith taught you about the psychology of financial manias. You read thousands of books, memos, case studies, and annual reports over fifty years. Every input was someone else's thinking.”

“These companies have gone from net cash in some cases to a little bit of net debt. You don't have enough cash flow from operations to support share repurchases, to support all of the things that companies spend money on if they're going to dedicate this vast amount of money to CapEx. So when you put CapEx on the balance sheet, regardless of the number of years over which you amortize it for depreciation, you're putting depreciation expense, which largely is maintenance on the balance sheet, and now you're putting interest on the balance sheet, which is interest expense.”

“In other words, the original ceasefire wasn't actually a ceasefire. So to extend it doesn't actually mean anything, because there was no ceasefire to begin with. So as Iran said, yes, it means nothing. And we already have proof of this, because just yesterday we learned that Iran attacked another three ships in the Strait of Hormuz. They fired at one ship that was passing just off the coast of Iran, and then another two that were in the Gulf of Oman.”

“Now you have AI calling itself, right? Which changes the magnitude of AI compute demand, probably by a hundredfold or potentially more, depends on just how crazy you let the AI agents go. And that obviously changes the picture completely. And I think one reason why a lot of people underestimated the most recent acceleration, and that you have pointed out yourself, you know, the compute is underestimated, because most people don't code, right?”

“Level three is autonomous agents. At this level, the user doesn't tell AI what to do. The user gives it a goal as well as the parameters of the desired output, things like length, time taken, content, and points covered. The agent does the work, checks it, and submits a finished product. This is labor replacement at the task level, not assistance.”

“Well, there's somewhat of a ceasefire and a blockade in the Straits Of Hormuz, And about 13% of the global oil supplies have been basically closed off to the rest of the world. And gulf producers shut in about 9,000,000 barrels per day of output, which, as you would imagine, is a bit of a problem for especially countries that are dependent on energy. Goldman Sachs has estimated the Persian Gulf crude output is down 14,500,000 barrels per day. That's 57% lower than prewar levels. And total US oil exports earlier this month hit an all time high of 12,900,000 barrels per day.”

“Anthropic is just my understanding is and there's even an article in this morning's journal. They're just killing OpenAI with the success in corporate in corporations with their clawed models. And so OpenAI share of of AI search has dropped from mid eighties to mid sixties, and it's falling fast. You've got regulatory risk. We're gonna find out on fair use. The Europeans are very aggressive in this room. And you've got a you've got a lawsuit from Elon when Sam A flipped from where this thing is a not for profit.”

“In other words, the original ceasefire wasn't actually a ceasefire. So to extend it doesn't actually mean anything, because there was no ceasefire to begin with. So as Iran said, yes, it means nothing. And we already have proof of this, because just yesterday we learned that Iran attacked another three ships in the Strait of Hormuz. They fired at one ship that was passing just off the coast of Iran, and then another two that were in the Gulf of Oman.”

“Tesla was able to launch its RoboTaxi in two new markets by April, where we see them starting in Dallas and Houston. And interestingly, they're launching immediately into the unsupervised RoboTaxi. So if you remember, when they started in Austin, there was a human safety monitor who would ride in the front passenger seat of the vehicle. But with the Dallas and Houston launches, they're going directly to no safety monitor, which tells me that the software is progressing well, testing is going well enough to enter new markets and immediately not need the safety monitor.”

“So first, I wanted to talk about Google DeepMind. They've just dropped Athletica this week. This is an autonomous math agent, and it's actually just built on top of Gemini three deeps DeepThink. And the results that it has been producing are really wild. So on the first proof challenge, which is basically a benchmark of unsolved or really complex kind of novel math problems, Athletica actually produce solutions that according to human expert evaluators were graded as, quote, publishable after minor revisions.”

“The CFO, Sarah Friar, reportedly told colleagues earlier this year that she doesn't believe the company is ready to go public in 2026. She's specifically worried about $660,000,000,000 in projected AI commute, compute commitments and whether the revenue ramp can support those contracts. Sam Altman is pushing for a q four IPO, and Friar is publicly opposed. She's the CFO inside of OpenAI. So we have some serious head butting from the top people inside of OpenAI.”

“And this is a political economy, you know, issue. I think more has given up on the idea of raising taxes. So certainly not on the middle class if you want to have any shot of electoral success, and cutting spending, you know, has also proven quite difficult. So the only option left is to grow the denominator, to grow the GDP.”

“The ongoing conflict involving Iran is the primary catalyst for the extreme volatility in oil markets.”

“We are so deeply afflicted by this idea of a concept of Bourbon's disease, right? Which for people who are not familiar, it's basically this idea that some sectors such as, you know, nannies or child care or health care doesn't really progress with technological innovation as you would have, you know, with say manufacturing, right?”

“In fact, for twenty five years, it's grown by 1.8%, but you're essentially if 30% to 40% of what every company makes goes to retire shares, you have not shrunk the share count, who got rich? The executives. You could say in the case of the shareholder, those repurchases supported the stocks and that's why we're trading at 26 times earnings today and that's probably the case. But those were dollars that didn't go into reinvestment in property, plant and equipment or acquisitions. That was money that was spent simply levitating a stock to make executives rich, driving up the stock price to higher and higher levels.”

“This is Mag five CapEx as a percentage of their operating free cash flow bill. And if you look at it in 2025, so that's this year, they'll spend about 66% of their operating cash flow on CapEx. In 2023, their total CapEx was a 156,000,000,000. And this year, it's 379,000,000,000. Right? So radical step up.”

“But I think to your point about a slow cognitive decline, we talk about tulipomania. This is my roundtable where I've got thirty, thirty two, really my best friends, colleagues, contemporaries, peers in the investment world that I've gotten over the years. ... I've got enough people in my universe that would I believe would be candid and say, Chris, you're starting to slow down. You need to do some you need to think about not being having your finger on the trigger of capital.”

“And then last year, it comes out that employees are gonna start to be graded on how much they use AI. And Meta's leadership expects employees to use AI a lot. What we've seen, there's a lot of internal memos that have kind of come out over the last few weeks. And one of them, the CTO, Andrew Bosworth, says in the future, AI agents are actually primarily going to do the work and that the human's jobs will be to supervise them, direct them, and and and help them improve.”

“Tesla was able to launch its RoboTaxi in two new markets by April, where we see them starting in Dallas and Houston. And interestingly, they're launching immediately into the unsupervised RoboTaxi. So if you remember, when they started in Austin, there was a human safety monitor who would ride in the front passenger seat of the vehicle. But with the Dallas and Houston launches, they're going directly to no safety monitor, which tells me that the software is progressing well, testing is going well enough to enter new markets and immediately not need the safety monitor.”

“If you take margins down from today's 12.8% to 10, and you're gonna crucify a 26 multiple to earnings. So multiples come in when margins come in. Wall Street Investors in general don't like compressing profit margins and they punish the stocks with lower multiples, lower and lower multiples when margins come in. I think there's a heck of a lot of risk in margins for reasons related to the CapEx, but also reasons related to how the economy is structured and competitive forces.”

“The Invest America accounts, now known as the Trump accounts, right, are more capitalism. They make every child a capitalist from birth. A private owner gives them a thousand bucks in a four zero one k like account that they own and control, their family has on their phone. At the start today, there are 65,000,000 kids in the country qualified for an Invest America account. Every kid under the age of 18, and every kid under the age of two will automatically get a thousand bucks in their account. Remember, the accounts have to be funded and established by our 200 birthday, 07/04/2026.”

“With Tesla starting a heavy capital expenditure program, at least 20 billion, if not higher, to fund its transition to an AI and robotics company, I think the market was worried coming into 2026 of where would that cash come from and could Tesla still generate strong free cash flow or even some positive free cash flow while that program is ramping up. And so positive free cash flow in the first quarter puts them to a good start to the year.”

“As of March 31, US publicly held debt reached 31,265,000,000,000, while GDP over the preceding year was 31,216,000,000,000. For those of you math buffs out there, that puts the debt to GDP ratio at 100.2%. The US hasn't finished a fiscal year with debt above 100% of GDP since 1946. It briefly crossed the threshold during the pandemic in 2020 when GDP temporarily shrank and the government borrowed massively, but the ratio fell back to below that as the stimulus ended as growth resumed and as inflation boosted nominal GDP. This time, the drivers are structural.”

“Well, the Internet is optimized for humans. The Internet is not optimized for other chatbots to use. And so at a certain point, this is what people talk about. At a certain point, the Internet might not be for us anymore. Right? The Internet might be a place where agents go to talk to other AI agents. Right now, the Internet is a place for humans, and so it's sort of built for us. And so there's a lot of cases where, the AI has to act like a human to get to what it needs because it's not built for the AI.”

“Megan looked at data from a website called Blind, where people who work at tech companies can post anonymously about their employers. In 2024, roughly 20% of the posts about Meta were negative. This year, more than 80% of the posts about Meta are negative. Because they're being asked to train their digital replacements, essentially. That is exactly how someone put it to me was, am I automating away my own job?”

“Go back to the history of Micron's earnings. I will read them from 2020 to 2027. $5.14, $7.74, negative $5.34, 70¢, $7.59. So I think the question is not are earnings real today, but are they sustainably real moving forward, or are we just at a point in time where things are at the right position cyclically? The last peak in earnings for Micron was $8.35 in 2022. So analysis to earn over 10 times that next year. It's just not likely to be durable.”

“Currently, AI is too cheap. People are just out there consuming token, because it's heavily subsidized, and, they ask, you know, the most powerful model for cooking recipe. Part of the reason why people run up these huge bills using claw open claw is exactly because the the system has not been optimized for token efficiency. And part of the reason why is because price has not really been reflecting the genuine cost.”

“It appears to me that OpenAI, through all these partnerships and announcements, is trying to create escape velocity. You know? And that could be against the model providers. It could be against a hosting provider depending on how you think the market plays out. But it'll it it creates an interesting stress test for anyone else in the ecosystem to say, are you gonna lay chase? It just feels like they're they're daring people to to follow them, and and I suspect a bunch don't.”

“On February 5, OpenAI released GPT 5.3 codex. In the technical documentation, they included this. GPT 5.3 codex is our first model that was instrumental in creating itself. The codex team used early versions to debug its own training, manage its own deployment, and diagnose test results and evaluations. Listen to that again. The AI helped build itself.”

“The aggregate numbers we're looking at, I, my intuition is that it's almost certainly not reflecting AI productivity. We are seeing measured labor productivity, the which is a residual. I suspect a big chunk of it is actually a composition bias that we sort of, move towards you know, if if the labor intensive sectors are shrinking and then you have a lot more investments, you know, that is very capital intensive, and then your productivity is measured on dollar value, you know, generated. It's gonna look as though, right, that the productivity has gone up.”

“There's a great video that that Pink references that we can put in the show notes where Bezos has asked about the decision to leave D. E. Shaw and start Amazon, and he said he used a regret minimization framework. He said, oh, it's only a nerd could do that. But he said, that he imagined himself being 80, and would he care that he left D. E. Shaw maybe for win a bonus? Or would he care more that he didn't take this chance, this kinda instinctive chance that he felt like he had to take? And he immediately, after thinking about it in that way, wanted to go do it.”

“The vertical integration of their stack is what truly sets Hyperliquid apart from other protocols.”

“So what Mark Zuckerberg has said is he wants everyone to have their own personal super intelligence. We think people, as I reported last year, he thinks people have the capacity to have more friends than they do, and that AI can solve some of these problems. The average American, I think, has I think it's fewer than three friends. Three people that they consider friends. And the average person has demand for meaningfully more. I think it's like 15 friends or something.”

“A memo went out on Tuesday from a researcher who works on building the models. And they said, hey, guys. Our models need to get better at learning how to use computers. And so therefore, we are now going to be monitoring your keystrokes, your mouse movements, and your click locations, feed that data to our AI models to help them understand, basically, how to use a computer. That sounds kind of dystopian. A lot of employees were not happy about this. The top ranked comment on this post was, this makes me super uncomfortable. How can I opt out? Spoiler, there is no way to opt out.”

“You could say in the case of the shareholder, those repurchases supported the stocks and that's why we're trading at 26 times earnings today and that's probably the case. But those were dollars that didn't go into reinvestment in property, plant and equipment or acquisitions. That was money that was spent simply levitating a stock to make executives rich, driving up the stock price to higher and higher levels.”

“If you plot tax receipt, federal tax receipt is the share of GDP over the course of the US modern history. It's always about 18 to 20 percent of GDP. We've been through significant tax reform, right? We've cut personal income tax, we've cut corporate income tax, we've got a level curve, and we've raised tax and whatever. Regardless of what we have done, the total tax receipt is about 17 to 20 percent. Now, the problem with that is that then the spending has been rising secularly and because the debt load itself, which last peaked at over 20 to 100 percent of GDP during World War II, and it went down in the 70s and now came back up.”

“STRC is specifically designed to address the yield gap by focusing on price-stable return mechanisms.”

“On 400,000,000,000 of CapEx, if you're writing off the asset over ten years, which is too long, there's a debate over whether it should be three or four years or five or six years. It's not about the debate, it's you're putting depreciation on the income statement, depreciation expense, and it's a real expense, it's a real charge. If your depreciation schedule linearly on a straight line basis is ten years, on 400,000,000,000 of CapEx, that's $40,000,000,000 of depreciation expense.”

“Since no one can say definitively whether this is a bubble, I'd advise that no one should go all in without acknowledging that they face the risk of ruin if things go badly. But by the same token, no one should stay all out and risk missing out on one of the great technological steps forward. A moderate position applied with selectivity and prudence seems like the best approach.”

“Global energy markets are responding to the news that the Iran conflict creates extreme oil price volatility.”

“Together, I think these are basically the four companies that are committing about $700,000,000,000 in CapEx for this year, and almost all of it is in AI infrastructure. Dan Herbach check over at Ramsey Theory Group put out a note today saying that Wall Street's AI boom is heading into a, quote, ROI reckoning. Alphabet's q one EPS is projected to drop 6.4% year over year despite revenue growing to 106,900,000,000 because the 175 to 185,000,000,000 in AI CapEx line is grinding their margins.”

“I thought it was good. First time writing, he's not going to be as funny ever as Warren, But I think he hit on all of the things he needed to touch on. He paid a nice, brief early tribute to Warren as he should have done, really demonstrated that he's a Berkshire guy. I mean, he gets the culture, he gets the integrity, he gets the value system of the place.”

“We are entering the next chapter of the story, which we ought to call news fatigue, where the headlines become overwhelming, the news becomes confusing, and eventually we get bored of it, and we decide to stop caring altogether. This is what happened with Iraq, it's also what happened with Afghanistan, and it is now happening with Iran, and we are seeing that reflected in the markets.”

“As enterprise efficiency increases with automation, seat-based SaaS is facing a crisis in its core business model.”

“It's a one thirty thirty strategy, essentially, where you go one thirty long, 30% short. So the short positions effectively generate losses that can offset gains from selling your concentrated stock position. They tend to be pretty high fee, though, because they're high input. You could be paying, I don't know, one and a half percent on top of whatever your advisory fee is just to kind of manage that. You're also not really eliminating your taxes. You're just deferring them. Eventually, the tax man always gets paid.”

“Now you have AI calling itself, right, which, you know, changes the magnitude of of AI compute demand probably by hundredfold or or potentially more. It depends on just how crazy you let the AI agents go. And that obviously changes the picture completely.”

“I'm gonna read this out loud. As part of its aggressive economic and political campaign against Brazil is investigating PIX, accusing the payment system of unfairly utter undercutting US financial and technology companies like Visa and Apple. I mean, that's the most absurd thing I've ever heard. Undercutting Visa? Like, do do they realize they have Visa and Mastercard have, like, the top two operating incomes in the history of American business? Like like, there's there's no one that needs less protection than these guys.”

“The headline for all of this is that the 08/02/2026 enforcement deadline for high risk AI systems under the EU AI act, it's getting pushed back. What they're saying is that stand alone high risk systems classified under annex three now have a 12/02/2027 compliance date. And AI embedded in regular products under Annex one gets pushed all the way to the 08/02/2028. So we are pushing this years out. The industry basically just got a sixteen to twenty four month deferral on the rules that they were supposed to be ready for in three months.”

“The information actually reported that Friar stopped reporting directly to Sam Altman last August and now reports to Fidji Simo, who runs the applications business. And Sam Altman has reportedly been excluding Friar from certain financial conversations. I think that's definitely not very normal CEO, CFO. Friar and Altman put out a joint statement calling the report, quote, ridiculous, which in, you know, CFO speak means it's basically true.”

“Meta finds itself sort of racing to catch up. Right? And so they were giving out $100,000,000 offers to researchers trying to basically, like, rebuild this team and become competitive. Multi year deals worth $300,000,000. These are for scientists and engineers. This is wild. It's like NBA superstar money. With some of them receiving a $100,000,000 straight up in year one, a $100,000,000.”

“We are looking at a timeframe where the market bottom for Bitcoin is likely to materialize in late 2026.”

“Sony is also coming out swinging. So they have something called Sony AI, and they have a project called Ace. And they just published on this their their project was just published on the cover of Nature, one of the, you know, top journals for science this month. And it is the first known autonomous robot that beats elite and professional level human table tennis players in real matches. Ping pong, I mean, I am I'm no expert, but I do love ping pong. And that is a game that if you see the pros play, the ball is flying so fast, is moving so quickly, and the precision has to be insanely high.”

“We are entering the next chapter of the story, which we ought to call news fatigue, where the headlines become overwhelming, the news becomes confusing, and eventually we get bored of it, and we decide to stop caring altogether. This is what happened with Iraq, it's also what happened with Afghanistan, and it is now happening with Iran, and we are seeing that reflected in the markets.”

“I think anybody that's been a student of financial history has, you know, studied different types of activities that that historically, let's just say historically, have created red flags. And the reason that, you know, any AI you talk to would know what you mean if you said circular revenues is because someone has used it in the past in a way that that wasn't good. And I just described those things to Chai GBT and ask it for its analysis both as an accountant and as a financial investor. And the AI itself, you know, would would find its way toward company names like Enron and WorldCom and those kind of things merely by describing the type of transaction.”

“Business investment surged 10.4%, the strongest growth in nearly three years, driven almost entirely by AI spending on equipment and intellectual property. One common economist at Pantheon Macroeconomics estimated that AI investment accounted for roughly half of overall GDP growth in the first quarter. Think about that. One category of spending is responsible for an entire percentage point of economic growth. That is how dominant the AI build out has become as an economic engine.”

“A memo went out on Tuesday from a researcher who works on building the models. And they said, hey, guys. Our models need to get better at learning how to use computers. And so therefore, we are now going to be monitoring your keystrokes, your mouse movements, and your click locations, feed that data to our AI models to help them understand, basically, how to use a computer. That sounds kind of dystopian. A lot of employees were not happy about this. The top ranked comment on this post was, this makes me super uncomfortable. How can I opt out? Spoiler, there is no way to opt out.”

“I don't know if you saw the Dworkish podcast where Jensen was on there and he was saying that the biggest constraint they have right now is actually plumbers for their data center. That's where the bottom of that is, it's just like real life plumbers and that has a lot of real world impact if suddenly the average family does not have as easy access to a plumber.”

“With Tesla starting a heavy capital expenditure program, at least 20 billion, if not higher, to fund its transition to an AI and robotics company, I think the market was worried coming into 2026 of where would that cash come from and could Tesla still generate strong free cash flow or even some positive free cash flow while that program is ramping up. And so positive free cash flow in the first quarter puts them to a good start to the year.”

“I'm having my roof repaired. I've got this €100 house with beautiful Spanish tiles that are also quite, nettlesome when they break. And, I've had it's been a struggle for me since last fall. I've been looking for contractors, and it's quite difficult to find them. When they are busy, they they don't answer your calls. They've got terrible websites. When they show up, they don't speak a language. And and then it's just a few of them. So this type of thing, I just don't see how AI is actually directly impacting it right away.”

“Bloomberg reported that a small group of unknown, unapproved users have had access to the model Mythos for two weeks. They broke into the model the same day Anthropic launched Project Glasswing, a controlled rollout giving a select group of companies exclusive access to the model. Anthropic says it is investigating the security breach from these unidentified individuals but has found no evidence of malicious use so far.”

“The aggregate numbers we're looking at, my intuition is that it's almost certainly not reflecting AI productivity, right? We are seeing measured labor productivity, which is a residual, right? And you basically sort of take a look at how total GDP growth is subtracted out, the stuff that you can measure and you attribute the rest of it to labor productivity. I suspect a big chunk of it is actually composition bias that we sort of move towards. If the labor intensive sectors are shrinking and then you have a lot more investments, and that is very capital intensive, and then productivity is measured on dollar value generated, it's going to look as though that productivity has gone up.”

“Nothing has ever taken hold at the pace AI has. It's able to change the world at a speed that approaches instantaneous, outpacing the ability of most observers to anticipate or even comprehend. In the past, infrastructure was built for a new technology, and it often took years for that infrastructure to be fully utilized. In the case of AI inference, however, demand already exists and is growing rapidly, and I'm told AI is supply constrained.”

“When we look at the speed of execution, Hyperliquid's advantage is vertical integration across the platform.”

“I mean, look at the metaverse. Right? It didn't happen. And it's $70,000,000,000 down the drain, but they're gonna spend more on AI now. Do you think they'll change their name again like they did when they were going all in on the metaverse? What would you name it? Like, what would you change it to? I thought of a name that they could they could use. Okay. Let's hear it. Skynet. That means nothing to me.”

“There were three or four really key things that the media misses and most commentators miss on it. And so where operating earnings were $44,500,000,000 for the year, that was down by almost $3,000,000,000 year over year. So a number of things transpired. So one thing you've gotta do is adjust for currency movements. So in the footnote to that operating earnings release, they tell you about any gains or losses on the currency translation of Berkshire's denominated in foreign currencies.”

“I've had this conversation with countless people. They feel as though they are earning something when they get a dividend. This is passive income by getting a dividend, and that's just not the case. Settlers paribus, all things being equal, if a company pays out a cash dividend, its price should drop. Its market cap should drop by roughly the amount of the cash that leaves the portfolio. And so you are trading optionality for an income stream. If you want an income stream, invest in bonds.”

“In the long term, it means we're going to completely change how we manage software vulnerabilities, because up until this point, at least in theory, you find out about a vulnerability, you work really hard, you patch it, maybe it'll take a few weeks, maybe it'll take a few months, and you hope that the bad guys don't manage to figure out the same thing and exploit it first, right? So we've had that lag, and now I'm sure this is a complete emergency for all the major tech companies.”

“Meta also believes it can rethink the company's org chart. In March, Megan reported on an internal memo that laid out how Meta was creating a new team focused on AI development. They would have a very flat organizational structure. It would be 50 employees reporting to one manager, for instance. So, like, really getting rid of the middle layers of management.”

“Greg announced a couple of weeks ago after the stock declined post the earnings release that he had initiated share repurchases again in consult with Warren and made sense in that the valuation relative to intrinsic value at least per my calculation was back down to where it was in 2024 when they stopped buying shares back. So I've been great made the announcement the stock rose. We can go down a rabbit hole for a minute if you want, but even beyond my GAAP adjusted earnings, simply taking operating earnings, which Berkshire doesn't make has a supplemental release to the 10 ks's and the 10 Qs's and they'll strip out from GAAP earnings, the earnings from the stock portfolio and they'll break out earnings by subsidiary.”

“In 2024, roughly 20% of the posts about Meta were negative. This year, more than 80% of the posts about Meta are negative. Because they're being asked to train their digital replacements, essentially. That is exactly how someone put it to me was, am I automating away my own job?”

“Meta has 3,500,000,000 daily users around the world. That would be a lot of people that could use the chatbot. It's a lot of people that could use the chatbot. Exactly. And so, you know, investors and analysts argue that Meta has the distribution. That's not the problem. Right? Like, if you're Claude or you're ChatGPT, the companies behind those chatbots, you're trying to grow. You're trying to get people to use these things, but, like, you're starting from scratch.”

“So I come up with when you just do a simple average of my four methods, a progression of 9.3% year over year, which gets you to a little over 1,200,000,000,000, almost 1 and a quarter trillion. My market cap would be intrinsic value and on a per share basis that went from the b shares a year ago were $5.22. I've got them at $5.70 per share now and the a shares are up to 855,396.”

“If we implement 50 different state rules that these companies have to jump through and companies that are that are competitors that are competing in the broader world don't have any of them, there is zero chance that's not gonna create mud and slow down The US players. There's just zero chance. And I'm certain the people that are writing these laws don't understand that there might be some global, you know, consequence of what they're doing.”

“While I mentioned it in my December memo, I want to point out again that some AI revenue is currently circular in nature, derived from AI companies buying from each other. The chain of revenue has to ultimately rest on end users paying for real economic value. And while that's increasingly the case, the question of how much revenue is circular remains an open one.”

“In the long term, it means we're going to completely change how we manage software vulnerabilities, because up until this point, at least in theory, you find out about a vulnerability, you work really hard, you patch it, maybe it'll take a few weeks, maybe it'll take a few months, and you hope that the bad guys don't manage to figure out the same thing and exploit it first, right? So we've had that lag, and now I'm sure this is a complete emergency for all the major tech companies.”

“What I do think AI is going to impact is we're gonna have much more modeling, a lot more varied modeling by everybody, really, and we're gonna have richer models, and I think we're gonna have much better agentic simulations. So this idea that, currently, how do you evaluate the impact of a policy supposed that we there's this discussion of the third is going to cut massively reduce the size of this balance sheet and, and and in in tandem cut interest rates. How would different parts of the economy system, financial system respond to it? And that is actually something that agentic simulation, right, can actually do, you know, a lot of work on.”

“Google is now kind of doing the opposite. They signed the contract that Anthropic walked away from, and they essentially allowed Gemini to be used for, quote, any lawful government purpose. Gemini 3.1 is already on the GenAI MIL platform, which is about 3,000,000 Pentagon staff. So Google's own DeepMind people are basically watching the company, take the deal that Anthropic refused.”

“Well, I think one thing that you can't get away from is this the fact that despite oil being sustainably above $90 a barrel, US producers are not rushing to drill. The Dallas Fed did a survey which showed rig counts actually declined. Even after a month of 90 plus dollar oil, oil executives are saying that the price swings are a bit too wild, and they can't really stomach it moving based on tweets and paper market manipulation. It effectively makes it impossible to plan capital budgets.”

“Lululemon is a couple things. It is, first and foremost, a premium athletic apparel brand, so your yoga pants, your run to train apparel, your footwear, your accessories. They have 811 stores with 65% of the revenue from The Americas and growing business in China and international markets. But they're also a brand in crisis story. They have a leadership vacuum. They have a proxy fight with their founder, margin collapse, and they're seeing North American deterioration all happening simultaneously. And that's why since May, these companies fallen dramatically. I think that a lot of the things that have benefited Lulu of the early days of athleisure was the fact that in a lot of ways, they were the only game in town. A lot of that's changed. For that reason, for now at least, I would stay away.”

“A friend of my daughter-in-law heads the department that writes advertising copy for an ecommerce company. She told me AI could replace 80% of her staff. I can't imagine software companies will need as many people to instruct Claude to write software as have been writing software up until now. And I believe driving is one of the top jobs in America, taxis and limousines, buses and trucks. Waymo, driverless cars, already handle roughly one fifth of the taxi trips in San Francisco, and I see them all the time in LA.”

“And so where I see the government potentially being involved is maybe setting some standards for disclosure. If you have an AI that has certain capabilities, do we need to track that? Do we need to notify anybody? And how is this going to be coordinated?”

“Coinbase and Circle have done this deal where Coinbase will allow you to earn 4% on your stablecoin balance, which, you know, to to get that kind of return at another bank, even a neobank, you have to have your direct deposit go there. Here, whether it's $10 or or a million bucks, you know, you put it in stablecoin with Coinbase, and you start earning 4% daily. And on top of that, you can transact immediately out of that account. So you don't have to, like, move it from your savings to your checking to get it to do ACA. But you can send stablecoin immediately in microseconds, and it'll cost you a few pennies.”

“And so where I see the government potentially being involved is maybe setting some standards for disclosure. If you have an AI that has certain capabilities, do we need to track that? Do we need to notify anybody? And how is this going to be coordinated?”

“The training phase must not be thought of as loading the model with information, which I had done until now. It goes far beyond that. It consists of teaching the model how to think. By absorbing text, the model learns how to understand reasoning patterns and form them, how arguments are structured, how to generate new combinations of ideas, and how to apply learned reasoning patterns to novel situations.”

“Well, the Internet is optimized for humans. The Internet is not optimized for other chatbots to use. And so at a certain point, this is what people talk about, at a certain point, the Internet might not be for us anymore. Right? The Internet might be a place where agents go to talk to other AI agents. Right now, the Internet is a place for humans, and so it's sort of built for us. And so there's a lot of cases where, the AI has to act like a human to get to what it needs because it's not built for the AI.”

“And that's one big differentiation between, I think, the AI bubble and the Internet bubble is that the AI bubble is very much adopted and used by everybody almost right away, unlike the dot-com bubble where you actually had a lot of sort of unused capacity that was at the end built out.”

“So I'm quite skeptical of this argument that we should preemptively, you know, justify rate cuts on the anticipation, unproven anticipation that AI was going to have this miraculous and dramatic impact on inflation. If anything, I think, you know, the AI build up, which still happens, you know, all in all the ways, all the rage. We're talking about hundreds of billions of dollars of, you know, CapEx in The US economy that is not very good at manufacturing stuff, you know, that has a lot of frictions that has all sorts of electricity shortages, you know, dilapidated grid system.”

“It's the art of war. The generals tend to fight the last war. And while there's COVID and we did see a little inflation as of the last few years, really, the major crisis of most people's recent memory is o eight, is the financial crisis. And that was a big deflationary impulse. But the previous crisis is usually not the next crisis. The next crisis is usually something very, very different. And in this instance, it really is inflation, that the risk for most people's portfolios is that inflation continues to eat away purchasing power.”

“The traditional seat-based SaaS model is facing an existential crisis due to AI efficiency gains.”

“We did this survey and asked people if you could start your career over again, would you do things differently? And in that survey, 70% of people said yes. And we did it again with Wharton just to make sure we had, you know, a a true academic survey going on, and they did a lot more people. And that number was still six and ten. There's a great book called The Power of Regret by Daniel Pink. He said, one of the most robust findings in the academic research and on my own is that over time, we are much more likely to regret the chances we didn't take than the chances we did.”

“So what Mark Zuckerberg has said is he wants everyone to have their own personal super intelligence. We think people, as I reported last year, he thinks people have the capacity to have more friends than they do, and that AI can solve some of these problems. The average American, I think, has I think it's fewer than three friends. Three people that they consider friends. And and the average person has demand for meaningfully more. I think it's like 15 friends or so.”

“Overall, it was a down day driven by news around Open AI. The CFO, Sarah Friar, said the company may not be able to pay for competing contracts if their revenue growth does not accelerate from where they're at. They miss their own targets of new users, revenue, and raise concerns that they'll be able to continue to spend as much as they are on their data centers, which kinda hit the tech industry as a whole, and that's why you saw the Nasdaq down. About 1% on the day, S and P down about half a percent.”

“The bottom line, the nineteen seventy three oil shock planted the seed for the renewable energy industry. The twenty twenty two Russia Ukraine shock, we out watered it. The twenty twenty six Iran war may be the event that forces it to grow. On the next Invest Talk, we'll look into this story, China's calculated play, how Beijing is turning the Iran war into an economic advantage.”

“So I am quite skeptical of this argument that we should preemptively, you know, justify rate cuts on the anticipation, on proven anticipation that AI was going to have this miraculous and dramatic impact on inflation, if anything. I think, you know, the AI build-up which still happens, you know, in all the rage, we're talking about hundreds of billions of dollars of, you know, cutbacks in the US economy that is not very good at manufacturing stuff, you know, that has a lot of frictions, that has all sorts of electricity shortages, you know, dilapidated grid system, you know, you have shortage of hardware and that's literally being driven up in price by the demand of AI. That's causing its own inflation.”

“And I think one reason why a lot of people underestimated the most recent acceleration, and then you have pointed it out yourself, you know, the computers underestimate us because most people don't code. The vast majority of people don't code. And the fact that AI has enabled people who don't code to actually, you know, interact with, you know, code is actually a little bit misleading in that if you don't code, you wouldn't ask questions that you will ask if you did cold. It's a bit like watching tennis having never held a tennis racket yourself that, the moves you watch, you know, Roger Federer make looks a lot easier.”

“Bloomberg reported that a small group of unknown, unapproved users have had access to the model Mythos for two weeks. They broke into the model the same day Anthropic launched Project Glasswing, a controlled rollout giving a select group of companies exclusive access to the model. Anthropic says it is investigating the security breach from these unidentified individuals but has found no evidence of malicious use so far.”

“Mark Zuckerberg himself is working on building a CEO agent to help him do his job. I think the agent is helping him retrieve information faster. So before, he might have to go through multiple layers of people to find information. Now, he can just ask his agent to go find wherever it is and, you know, the emails or the drives or whatever and get the answer versus having to, you know, do that like, telephone game of, like, hey. Can you go ask this person this?”

“And there's something else. AI doesn't have skin in the game. It doesn't feel the weight of concentrated positions or the fear of capital loss. Its willingness to take risk might not be constrained by humans' normal risk aversion. The best investors sense potential risk intuitively, and this contributes greatly to their success.”

“If Meta gets their way, they're gonna revolutionize the Internet again. Think what they did the first time around with Facebook, what they tried to do the second time around with the metaverse, it didn't work. This is sort of like their attempt to do the metaverse thing again, but in a way that they think is gonna be more successful now. Yeah. I mean, look at the metaverse. Right? It didn't happen. And it's $70,000,000,000 down the drain, but they're gonna spend more on AI now.”

“Then if we look at factory expansion plans, the transition to take the factory that was making Model S and Model X vehicles and take that to produce the Optimus Humanoid Robots is still on track for to enter production by the end of this year. And so with Tesla's two big pillars in the future going to be self-driving cars and Humanoid Robots, both of those initiatives are on track with the positive free cash flow and that led the stock to rise after hours.”

“But it's it's a proxy for it's gonna be a tough hurdle for the aggregate of all of these guys in this arms race because the numbers are just frankly staggering. And I don't see how you've got enough revenue and then profit opportunity to make the whole thing generate a return on capital. Maybe they do, maybe they don't. I don't need to play in the game, but I think it's a tough hurdle.”