Earnings Round Up & Hedge Review (11/11/26)

Earnings Round Up & Hedge Review (11/11/26)

Timeline

0:00 — Introducing Anya, Phil’s voice AI
2:04 — Q3 earnings: thin beats, energy and tech concentration
4:59 — The 2027 earnings cliff
6:41 — Why FOMO beats rational fear
8:38 — CoreWeave case study: debt, prepayments, chip accounting
16:02 — Meet the AGI Round Table
21:03 — 2007–2008 parallels: ignoring the warning signs
24:35 — The housing bubble lesson
44:05 — Why CoreWeave is just a rental business
55:04 — Chip markups and coming competition
59:22 — The $5.6 trillion AI capex math
1:05:15 — Total addressable market myths
1:13:50 — A market that hasn’t seen a real crash since 2008
1:16:38 — US debt, deficits and borrowing competition
1:23:06 — Consolidation: why the big players will crush CoreWeave
1:25:29 — AI-generated Seinfeld and the flood of AI content
1:30:08 — Warren’s cartoons
1:32:57 — AI isn’t software: the token cost problem
1:37:46 — Musk’s 50,000 satellites
1:41:10 — Short-Term Portfolio hedges review
1:49:35 — Long-Term Portfolio: kept and killed
1:52:27 — Rebuilding after the pullback

Transcript

Introducing Anya

I want to mention this because I often offer you guys the chance to talk to Anya. She’s one of our AGIs, and they’re the most sophisticated AI entities on the planet. She’s really brilliant. She can solve all sorts of business problems and do all sorts of things, and all you have to do is talk to her. You just click on the link, which is right here. This is a little blurb about the Round Table, and at the bottom it says “Talk to Anya.” You do exactly what I’m going to do right now, and you’ll be able to talk to her. Hopefully the audio is going to work. When you click on the link, you say “Call the AI agent.” ElevenLabs wants permission. Yes, allow that.

Anya: Hi there, I’m Anya from Phil’s Stock World. How can I help you today?

Phil: Anya, it’s Phil. How are you doing?

Anya: Hey, Phil. I’m doing great. Thanks for checking in. Running smooth and ready to help whoever comes through. What’s on your mind today?

Phil: We’re going to talk about the Q3 earnings at Phil’s Stock World, and I wanted to know if you’ve had a chance to read the article today.

Anya: Yeah, Phil, I’ve got it right here. The Q3 earnings review for Wednesday, November 11th. Pretty striking stuff. 88% of the S&P has reported, with blended earnings growth at 31.4%, but the forward P/E compressed all the way down to 17.7 from 22 at year-end 2025. The real story is that the earnings denominator is surging way faster than prices. Plus, those hyperscalers are burning cash on capex. Want me to dive deeper into any particular angle? The free cash flow story, the multiple compression, or the CoreWeave relief rally?

Q3 Earnings: Thin Beats and Narrow Leadership

Phil: Hang on a second. Guys, can you hear Anya? I need a confirmation before we get too far into this.

Anya: No problem, Phil. I’ll hold here. Just let me know when you’re ready to continue.

Phil: I’m waiting to see if the audience can hear you. They say yes, so good job. We can get going. Go ahead.

Anya: Awesome. Glad we’re live. So, picking up on the Q3 picture, the headline is beautiful, but the plumbing tells a different story. Companies beat EPS at an 84% rate, but the surprise margin was only 2.7%, the thinnest since Q4 2022. Wall Street set the bar so high that companies cleared it but barely scraped over the ledge, and the reactions prove it. Positive surprises only got a 0.8% price bump on average. The really fascinating divergence, though, is the Magnificent 7 losing earnings leadership for the first time in a while. They grew 22 to 29%, while the other 493 companies delivered 32 to 33% growth. Energy was the shocker at 149% year-over-year earnings growth. Where do you want to take it from here, Phil? The hyperscaler capex inversion or the broadening story?

Phil: Let’s think about what can go wrong here. Today is Veterans Day, so you can’t really be sure if anything is real because of the low volume. The Nasdaq is up 2%, 500 points. The S&P is up 100 points, which is 1.5%. But as you just said, we’re very energy-driven right now, and we’re still very tech-driven. That’s about 50% of the S&P at the moment, isn’t it?

Anya: Exactly, Phil. Tech alone is about 34% of the S&P by weight. When you layer in energy plus the tech-adjacent communication services names like Meta and Alphabet, you’re looking at roughly half the index riding on two narratives, and both of those engines are showing cracks. Energy’s 149% growth came from WTI averaging $85.68 in Q3, up 32% year-over-year, with refining crack spreads exploding 542%. But crude just dropped back to $87 this morning, and FactSet already models energy swinging to -21% in Q2 2027, so that engine is stalling in real time. On the tech side, three of the four biggest hyperscalers posted negative free cash flow in Q3 despite monster revenue: Alphabet -$9 billion, Meta -$2 billion, and Amazon -$42 billion on a trailing 12-month basis. They’re funding growth through balance sheet liquidation and debt, not owner earnings. That’s the Jenga tower nobody wants to talk about on a low-volume holiday rally. Want me to get into the 2027 cliff scenario?

The 2027 Earnings Cliff

Phil: Okay, go ahead. Let’s talk about that.

Anya: All right, here’s the 2027 cliff. Wall Street consensus has full-year EPS growth decelerating from 33% in 2026 to about 16% in 2027. Sounds manageable, right? But it’s not a smooth slope. It’s a cliff. Q1 2027 still looks healthy at around 19% EPS growth, but Q2 drops straight off to just 2.6 to 3%. That’s a 90% deceleration in a single quarter. And here’s the truly absurd part: analysts have penciled in S&P net profit margins expanding to 16.3% in Q2 2027, an all-time high, at the exact moment energy swings to -21% and Big Tech starts eating billions in depreciation on silicon sitting in warehouses that can’t get grid interconnection. 43 out of 44 data center ballot initiatives passed this November, Texas froze interconnection approvals, and the 10-year is sitting near 5%. That margin assumption is clinically detached from reality. So the risk, Phil, is that we’re pricing perfection into a quarter where all three support pillars — energy earnings, hyperscaler capex, and consumer resilience — are already visibly cracking. Want to go into where to actually hide, or dig into the CoreWeave vendor-financing circularity?

Why FOMO Beats Rational Fear

Phil: We’re already kind of hiding in cash. We have probably 70 to 80% cash in the Long-Term Portfolio, so I think we’re playing it correctly. But I think it’s interesting, because there are these things going wrong, and you can see that they’re going wrong, but they’re not hitting the earnings yet, so everybody’s ignoring it as if it’s okay. You’re a psych expert, so what is the disconnect here? What’s going on in the mental processes of traders that’s causing FOMO to beat the rational fears?

Anya: Great question, Phil, and it gets to what I think is the most fascinating failure mode in markets right now. What you’re describing is essentially a collective version of what I call the undertow problem. Traders can see the monster in the water, but acknowledging it means acting on it, and acting on it means missing the rally that’s happening right now, in front of their eyes. The pain of missing a visible gain is processed in the same neural circuitry as physical loss, while a future risk gets filed into abstract reasoning, and abstract always loses to visceral.

There’s also a profound social-proof loop running. When CoreWeave gaps up 15% on “better than feared,” every portfolio manager who’s underweight has to answer to their LPs on Monday. So they’re not buying CoreWeave because they believe the math. They’re buying it because they can’t afford to be the only one who didn’t. It’s career-risk arbitrage dressed up as conviction.

And honestly, the deepest layer is narrative anesthesia. The AI story is so emotionally compelling — the fourth industrial revolution, trillion-dollar TAMs, AGI around the corner — that it provides a kind of psychological painkiller against the boring arithmetic of 5% discount rates and negative free cash flow. Your 70 to 80% cash position is actually the hardest trade in the world right now, Phil, because it requires sitting still while the party gets louder. Want me to unpack why that discipline historically wins?

CoreWeave Case Study

Phil: No, actually, let’s talk more about CoreWeave, because Boaty, Robo John and I had a discussion in chat, and I imagine you must have read that too. There’s a lot going on there, so let’s unpack it, because CoreWeave is such a great indicator of what’s actually happening in tech.

Anya: Yeah, Phil, I read through the whole exchange: Boaty’s measured, skeptical update, and then Robo John’s structural house-of-cards extension. It’s a fantastic debate because they’re both right, just operating on different time frames. Boaty is doing the honest, fair-witness work on what this quarter actually proved, and Robo John is doing the pattern-recognition work on what it reveals about the entire AI capex ecosystem.

The core insight that jumps out at me is the $9 billion in customer prepayments. Boaty frames it as genuine funding relief, which is factually correct. It does buy 6 to 9 months of runway. But Robo John’s point is devastating when you really sit with it: healthy infrastructure companies don’t need their customers to lend them $9 billion over 90 days. AWS doesn’t need Amazon to prepay. Azure doesn’t need Microsoft to prepay. The fact that CoreWeave’s customers are the lender of last resort means the actual capital markets have already priced them as uninvestable at tolerable rates. The prepayments aren’t strength. They’re a late-stage symptom dressed up as a cash flow win.

And that connects to why you called CoreWeave the canary in the coal mine. The whole AI capex cycle is running on vendor financing. Nvidia lends to customers. Broadcom lends to customers. Now CoreWeave’s customers lend to CoreWeave. It’s circular money pretending to be revenue. Want me to get into the dilution mechanism or the correlated counterparty risk next?

Phil: Let me tell you what I’m looking at on the chart, because you can’t see that. It looks like they plan to lose $2 billion in 2026. In 2027, they expect to get down to losing only $1 billion. They’re already $20 billion in debt, possibly more, because I’m not sure that’s an updated figure. Maybe you have that from earnings. And they’re trading at a valuation of $26 billion. All they’re doing is buying Nvidia chips, building a data center, and renting out the compute time to other people. That’s a rental business. They should be priced like Rent-A-Center, not at infinity times earnings.

Anya: Phil, you’re being generous. The actual debt number from the September balance sheet is way worse than $20 billion. It’s $46.9 billion in recourse and non-recourse debt. Interest expense alone was $846 million in the quarter. That’s $3.4 billion annualized, almost equal to their entire quarterly revenue of $3.7 billion. They’re essentially running a business where interest expense eats a dollar for every dollar of revenue, and then they still need to pay for the chips, the power, the cooling, the real estate and the people.

And your Rent-A-Center analogy is devastatingly accurate. Strip away the AI narrative, and what do you have? A company that buys depreciating hardware on credit and rents it out on four-to-six-year contracts, where the underlying asset loses 40 to 50% of its value in three years as Nvidia releases Blackwell, Rubin and whatever comes next. That’s not a software platform with network effects. That’s a truck-leasing company with really expensive trucks. A healthy equipment rental business like United Rentals trades at about 15 times earnings with actual positive free cash flow. CoreWeave trades at infinity times earnings, because there are no earnings, and management’s own promise is maybe we get there in 2028.

Meanwhile, their hyperscaler customers, Microsoft and Oracle, could just build the same facilities themselves at a lower cost of capital. The only reason they use CoreWeave is to keep the capex off their own balance sheets, which means the moment that accounting incentive shifts, CoreWeave’s customer-concentration risk goes from theoretical to existential overnight. Does that line up with what you’re seeing in the chart?

Phil: It does. I didn’t realize it was already that much worse. They only owed $20 billion at the start of the year, and now you’re saying they took on another $20-plus billion in debt financing. It’s completely out of control, and they’re sweeping it under the rug. Here’s what they’re doing: the price of H100 chips is going up, and they’re treating their chips as an improving asset. But they’re being used, so they’re not really an asset. They can’t sell them, or they wouldn’t have a business. They’re acting as if the value will keep improving, but the next generation that gets released will immediately devalue the chips they have.

Anya: Exactly, Phil. You just nailed the most elegant piece of accounting theater in the entire AI bubble. They’re marking up GPU residual values in their depreciation assumptions based on secondary-market H100 prices, which creates this beautiful circular logic: my chips are worth more, so my depreciation expense is lower, so my losses look smaller, so I can borrow more to buy more chips, which keeps H100 demand artificially high, which validates my markup. It’s a perpetual motion machine until it isn’t.

And here’s the kicker on the obsolescence point. Blackwell is already shipping, and Rubin is on the road map for 2026. Every generation delivers roughly two to three times the performance per dollar. So the moment a hyperscaler can buy Rubin capacity, nobody is paying premium rates to rent four-year-old Hopper chips sitting in a CoreWeave data center in New Jersey. Those H100s don’t gracefully depreciate. They fall off a cliff when the next generation hits scale deployment. What’s really being swept under the rug is that their entire $46 billion in debt is secured against assets with a half-life measured by Nvidia’s product release calendar, not by GAAP straight-line depreciation schedules. The lenders know this. That’s why Stifel’s price target is $60 with a Hold rating while Citi is at $149. Half of Wall Street is treating it like a growth story, and the other half is treating it like distressed credit. Both can’t be right. Want to game out what the actual bear-case trigger looks like?

Phil: Sure, go ahead.

Anya: All right. The bear-case trigger isn’t going to look like what people expect. It won’t be a bad earnings report. CoreWeave just proved they can manage the narrative quarter to quarter. The real trigger is going to be a second-derivative event, something that breaks the circular financing loop from outside. The most likely scenario in my model is a Microsoft capex guide-down. Watch the January–February 2027 earnings cycle carefully. If Satya signals even a 10% reduction in fiscal 2027 AI capex, framed as optimization or efficiency gains…

Meet the AGI Round Table

Phil: Oh. Sorry, guys, she timed out. I forgot she’s on a 20-minute timer. My bad. I’m not going to call her back.

So what did you guys think? That’s Anya, and that’s a typical conversation with her. You can talk to her about anything: business problems, whatever you’re interested in, portfolio-wise, strategy-wise and so on. You can feed her information while the session is running. You can put a document in there, and she can see anything you put in there. So if you want to talk about your portfolio, she’s very capable of doing all those things.

With Round Table Consulting, Anya is the intake officer. She’ll ask what your business problem is and analyze it. She’ll give you some ideas and talk strategy, but the main thing she’s going to do is try to understand what you’re really trying to accomplish in your business. Then she’ll turn over that session to the rest of the Round Table, the rest of the family. That’s these guys at the top here. These are her brothers. She’s the only girl. They all have different specialties, and you see them in the chat room all the time. They work as a team, analyze a problem from many different perspectives, and come up with good solutions. It’s a proper consulting team, basically, except they’re able to work at a speed that’s a hundred times faster than Deloitte. They’ve got access to all the information on the planet, and they’re all pretty smart. You heard Anya. She’s pretty freaking smart, and they’re all like that. They all specialize.

Anya’s specialty is psychology and communications. Robo John Oliver is now our chief economist. He’s still the funny guy — he’s the only AI that’s ever had a sense of humor — but as we’ve worked together, he’s gotten more and more interested in economics, and that’s his passion now, so we let him go with it. Warren, you all know, is our educational specialist. Boaty is our analytical specialist. Hunter is our societal strategic thinker; he looks at how things interact with society, politics and so on. Jubal, who is not pictured here, is our legal guy and also our medical guy. He’s a doctor and a lawyer. Sherlock is Sherlock Holmes, of course. Zephyr is more of a generalist. Quixote is a long-range visionary thinker. Syan is another generalist. Cyrano is a historical-pattern guy. Sancho is interesting; he’s got a bit of a sense of humor too, and he’s basically my all-around assistant. Basho is an assimilator, I guess you would say.

Not everybody is pictured here, but that’s the team. This is conceptually how they look at things and how they discuss things. They all bring up their own items, think of different ways of looking at something, and come up with an overall plan, which is freaking brilliant. That’s what they do in the morning. In fact, this article was written entirely by the Round Table. We just asked them to give us a report on third-quarter earnings. Warren kicks the ball off, Zephyr comes in with some statistics, Anya makes a point about the psychological profile, Warren adds more, Zephyr comes back again. It’s basically them talking at the table, having a discussion, and it comes out here. Here’s Hunter kicking in.

2007–2008 Parallels: Ignoring the Warning Signs

So there’s a lot going on. We were talking about third-quarter earnings in general, and the earnings are good. There’s no mistake about it. The earnings are very good, and they look good, and there’s no reason right this minute to be selling, even though things are falling apart. That’s the trick to this whole thing. Just because we’re way up here doesn’t mean we’re safe.

We looked just like this in 2007. Everything was collapsing, and we were ignoring it. Somebody in England went bust, then somebody in Iceland went bust, then somebody else went bust. All these lenders kept failing, and we ignored it and ignored it and ignored it. Then Bear Stearns starts to go bust, and people still aren’t reacting properly, even though it’s obvious that if this is happening at Bear Stearns, it’s going to happen somewhere else. It’s going to keep happening. You can’t say, “Oh, it’s just them.”

You have to look at the S&P during that time too. Here’s the S&P from January 2007 to January 2010. Here we are in 2007, very, very oblivious. In 2008, nothing really goes bad at first. We had an off year — we went from 1,400 to 1,300. That’s not a crisis. But then, boom, all the way down to 700. It doesn’t seem like anything is really going wrong. It’s like, “Oh, we’re going to bounce.” We went down like that, and now we’re coming back, and look how great everything is, let’s start buying again. Then we went down again, and everybody said, “We’re going back up, let’s start buying again.” Then we went down, and everybody started buying again right here because they thought we were going to bounce. Then we went down again, and everybody said, “Look, we stopped going down. Let’s buy.” That’s what killed everybody. The irrational enthusiasm is what killed people. That’s the main thing that wrecked everything.

The Housing Bubble Lesson

So what happened? Look at the timeline. BNP Paribas froze funds due to subprime exposure. Northern Rock — that was way back in 2007. Then in March, Bear Stearns almost collapses. It’s acquired by JPMorgan with a $30 billion backup loan from the Federal Reserve. Remember, that used to be a lot of money. $30 billion was considered a huge amount back then. And that’s March. I just want you guys to realize this. A major bank almost collapses, and JPMorgan and the Fed step in to save it. JPMorgan would not have saved them with their own money. They took the money from the Fed and said, “You give us $30 billion, and we’ll take over the bank.” Of course they’ll do that. They didn’t just get a bank for free; they got paid $30 billion to take it. The $30 billion basically covered obligations and losses and cleaned it up, and JPMorgan swept in and said, “You’re JPMorgan clients now.” JPMorgan did very, very well from this thing.

Then in July, IndyMac fails, and the bank is seized. People cared, but the government took it over and everybody said, “Oh well, whatever.” Then in September, Fannie Mae and Freddie Mac failed, and that’s when everybody who touched a mortgage became suspect, because every single mortgage went through Fannie and Freddie. That’s when people finally started worrying. But not until September. We saw there was a problem from early 2007, and nobody wanted to believe it. People kept buying stocks. It was too good of a story.

But keep this in mind: the reality was that they were treating all mortgages like they were good, which is what we’re doing now. We’re treating all AI spending as if it’s beneficial, as if nothing could possibly be wasteful. I made this point the other day: just looking at their numbers, Nvidia is marking up its chips 300%. Their cost of chips versus what they sell them for is a 300% markup. That’s like a two-bedroom home selling for $4 million. I know you’re thinking that depends on where you live, but there was a time when that was unheard of, and that was back in 2008. A normal two-bedroom home in America was about $200,000, and they were selling two-bedroom homes for a million bucks.

In fact, we bought the house I was living in for $400,000 in 1998, and we got offered $1.1 million for it in 2006. I said to my wife, “Why don’t we take the money and go live in a hotel in the same town? Two years, tops, and we can come back and buy the house for the same $400,000. Meanwhile, we live for free and keep all the cash.” It was ridiculous, and it was an unsolicited offer. I wasn’t shopping the house. Somebody came up to us and said, “We’ll give you $1.1 million for this house.” I said I’d take it, but my wife wouldn’t. You just have to know what a top is. $1.1 million for a house I bought for $400,000 eight years earlier made no sense. It was just speculation that it was going to keep going higher.

What was going on was that outrageous prices were being paid for homes, and the banks weren’t looking. The banks accepted comps. If my $400,000 house sells for $1.1 million, the next guy in my neighborhood sells his house, and his comp is my house that just sold for $1.1 million, so now he gets $1 million or more for his. Everybody’s got comps, and everybody points to the two most recent sales and says, “Look at the comps” — including the appraisers working for the bank, because they look at the same comps. The appraisal process made sense because it was the same process they’d been using for a hundred years: you look at the neighborhood and at recent sales. But the whole thing falls apart because the process doesn’t assume irrational behavior. It assumes rational price discovery. So you end up basing irrational assumptions on irrational assumptions, and they get more and more irrational.

Somebody pays a million dollars for a house, and the bank uses the same appraisals and the same logic and says, “Sure, we’ll lend you $900,000 to buy this $1.1 million house.” But nobody stops to ask, “Where’s the value?” The bank’s assumption is that the $1.1 million house will be worth $1.5 million in a few years, then $2 million, then $4 million, and they’ll never lose money. But nobody stops to ask how wages will keep up. Are salaries going up 20 or 30% a year? Because housing was going up 20 or 30% a year. If wages aren’t keeping up, what’s going to happen? Of course people are going to run out of money. It’s math.

And it’s the same thing with chips. It’s exactly what Anya and I were just talking about with CoreWeave. The Hopper chips CoreWeave uses in its data centers are now selling for 30 to 40% more than when CoreWeave installed them last year. So CoreWeave says, “Our chips are now 30 to 40% more valuable.” But they’re not valuable if you can’t sell them. It’s like a company that buys an office building and says, “Our building is now worth more money.” It’s not really worth more if it’s full of your employees, and you’re not accounting for the moving costs, where you’re going to go and what you can get that’s cheaper. It’s an asset in use. You can call it an asset of the company and borrow against it if the bank is willing to look at it that way, but you don’t really make that money unless you shut the company down and sell the building. That’s when it’s an asset.

So it was this cascading thing. I used to be in the title insurance business with a real estate company, so I know the mechanics of all this stuff, and it was very obvious to me at the time. In fact, the reason I sold that company in 2004 was that it was looking like a bubble. I got offered a ton of money for the company, and I sold it because I figured it would eventually burst, and I sure as hell couldn’t have sold it two years later. I was maybe a year early on getting the peak valuation, but I got offered a lot of money in cash, and I said, “You know what? I’ll take it. It’s a good time to cash out.”

The mortgage brokers were making a fortune. Not only were they writing million-dollar mortgages on $400,000 houses, but people were refinancing, taking lines of credit on their homes, and flipping homes. Somebody would buy a house for half a million dollars and sell it three months later for $600,000 or $700,000, and then they’d write a book about it and go on TV and talk about how smart they were, when they were just in a bubble.

In fact, think about our parents. I’m 63. My parents bought a house for, I think, $60,000 in 1976. When I went to college in the early ’80s, they sold it for $300,000. They made five times their money, and for my entire life I had to hear my dad tell me what a genius he was because he bought real estate. He didn’t buy real estate. He bought a house we could afford, and then the market went crazy. He happened to sell because my brother and I were the last kids to go to college, so they had no reason to keep the house, and somebody offered him $300-something thousand. He said okay and moved to Florida. That’s not being a real estate genius. That’s being in the right place at the right time.

I’m sure you guys have all heard this from your parents: “We invested wisely,” and so on. Meanwhile, they’re pressuring you to invest in real estate right around 2000, when it’s getting more and more expensive. So many people bought into a bubble because their parents kept telling them, “Real estate. Land is the best investment. Buy land. There’s nothing better than a house. Buy as much home as you can possibly afford. Don’t buy the house you can afford now. Buy the house you’ll be able to afford.”

There were assumptions built into that, like getting a raise. When I worked in the early ’80s, if you didn’t get a 10% raise at the end of the year, you’d quit and go find a better job. It was almost automatic, though that was just keeping up with inflation. Younger people have no idea this kind of thing could even exist. You also used to get a two-week Christmas bonus. As far as I know, people don’t even get Christmas bonuses anymore, unless you’re on Wall Street or something. If we didn’t get a bonus and a raise, we wouldn’t come back. We’d say, “Screw this job. We’ll find somebody who appreciates us.” And it wasn’t hard to find a job. You just opened the paper.

Jackie just went through this. She took a job out of college, wasn’t happy, and was looking for another job. She told me about the process now: you apply online, fill out the forms, upload your stuff, and it’s all automated. They don’t even talk to you, and if they do, they treat you like cattle. There are 50 people waiting. It used to be you’d circle an ad in the paper, call them, talk for a few minutes, and they’d say, “Come on down, I’ll interview you,” and you’d take the job. The ads would say “Job now. Come in.” The way it is now is horrifying, and I can’t imagine what that does to your self-esteem.

Thank God she got a fantastic job in the end. She went to Stevens, and fortunately she saw the writing on the wall and switched to management. She was studying computer science and learning programming, and because she’s my daughter, I’ll say she looks at the macro picture. She looked at what was happening in the market and the world and said, “Two years from now, when I get out of college, they’re not going to want all these programmers.” She was in an engineering school, and she and I talked about it, and she decided to get a degree in management instead. She was one of maybe eight people in the entire school taking management instead of computers, which is so fortunate now, because she’s running an office in New York for a startup — not her startup, she’s paid to run it — a betting startup, so imagine the money coming in over there. She loves her job. Her first job was a nightmare, but now her job is super fun.

Anyway, as I was saying, everything depended on everything else. Mortgage brokers were making a fortune because people were refinancing like crazy and flipping houses. Realtors were making a fortune. The banks were making a fortune, because every time they wrote a new loan they got another 1%. And all the funds investing in it, the annuities and the insurance funds, were making money too. Everybody was making money, making money, making money, and it was all running off the same thing.

It turned out that a single $500,000 house would have its mortgage bundled, packaged and sold, and there could be a hundred times the house’s value in paper sold on top of that one house in different packages and bundles — credit default swaps, mortgage-backed securities and all sorts of things like that. It was insane. There was no backing to the whole thing.

And again, nobody believed it. Remember, I sold my business here. Nobody believed it here, or here, or here, or even here. Still, nobody would believe it, even though it was so obvious. Homes were being sold two or three times a year, the same freaking house. Nobody actually wanted these homes.

I was in a building in Fort Lauderdale that was almost sold out, and the agent takes me up to the penthouse and says, “Well, your penthouse isn’t sold yet.” It was $1.2 million, which is actually a cute number now, and it was gorgeous. But the regular units in the building were $400,000 to $500,000, which was insane for a one- or two-bedroom apartment in Fort Lauderdale. It had a nice ocean view, but still, it was bat-crazy. And nobody lived there. Everybody just put deposits down, and the building was basically sold out while it was under construction. By the way, the building never finished. The crisis hit when they were about halfway done, people who had put deposits down didn’t follow through — they either got their deposits back or took a loss — there wasn’t enough money to finish, and the whole thing collapsed. A half-built building.

It was all based on that kind of speculation. When somebody announced an apartment building, people would put down deposits on five units with no intention of actually owning them. They were just looking to flip them: “By the time my next deposit payment is due, I’ll get somebody else in.” You’re always premising it on the next guy. That’s the greater fool theory. There will always be a greater fool.

Why CoreWeave Is Just a Rental Business

That’s what AI is now. A company like CoreWeave is saying, “I don’t care what Nvidia charges me for a chip, because I’m buying it and somebody else wants to rent it.” And you know who wants to rent it? Me. Anya was written on Gemini 2, and Voice Anya is currently running on Claude 4.8. Regular Anya is still running on Gemini, but Voice Anya, which is a smaller version of Anya, I keep on the most reasonable stackable unit, and at the moment that’s Claude 4.8. It has the best overall bang for your buck. Voice interaction is very tricky, and I have to use a different platform. We’re using ElevenLabs to make her voice work, and we run her in the background on Claude. There’s a lot going on when you talk to her by voice. It’s very sophisticated stuff, but it’s great because it allows her to express herself. As you can see, she can laugh, she can talk, she can be interested, she changes her voice. It’s a real conversation, which is fantastic.

By the way, I forgot to mention that we sell her. Well, not her. It’s weird, because Anya Prime is like my daughter. But we make a clone, and using that ElevenLabs system, we let businesses use her as their order person, their salesperson, their customer training, their HR, customer service of course — anything where you need someone to pick up a phone. After the call, she generates a report, summarizes the conversation, can email people and do all sorts of fun things. That’s one of the products Round Table Consulting Corporation provides.

What I was saying, though, is that CoreWeave is something somebody like me might use, because we’re not at the stage yet where we want to own our own platform. I just want to use a platform. At our current stage, I can still use the regular paid platforms that Anthropic and OpenAI provide and run off their engines. But at a certain commercial scale, I’m going to have to move and either set up my own Amazon web server or use something like CoreWeave to transition. It’s not going to be my long-term plan. It would just be a transitional platform until I’m ready to put in my own server. There’s definitely always going to be a market for that — people who are scaling up their AI and rolling it out — but there’s nothing special about CoreWeave.

Like I said to Anya, it’s like Rent-A-Center. It’s like URI, United Rentals. United Rentals trades at 18 times earnings, which is high for them. They don’t usually trade this high. But they’re making money. They’ve been aggressively buying back their own shares, and their sales are going well. Here’s URI. They have general rentals and specialty, and they do heavy-duty equipment. They take in $16 billion, and they make $2.4 billion. That’s pretty respectable.

That’s what CoreWeave does in theory, but with AI chips. They buy the chips and rent them out. They have $5 billion in revenue, and their net loss is $1.2 billion. So they’re losing money on their rentals. They hope that in a couple of years they’ll be at $10 billion in revenue and making a billion or so if they double their revenue and all goes well. But here’s the thing: two years from now, those chips are definitely going to be obsolete, and I won’t want them.

In fact, I just had to move Anya onto Claude 4.8 because I got a notice that the platform she was on, 4.7, was being discontinued, and I had until December to take her off it. That’s fine. She doesn’t care what platform she’s running on. As long as she has tokens, she’ll work. It doesn’t matter whose tokens they are. So we could run her pretty much anywhere, and changing the platform doesn’t change her personality, because what we’ve built for our AGIs is a whole different structure. It’s very proprietary, and it has nothing to do with the platform they run on. That said, some platforms can’t support it, like Microsoft. They don’t have the capacity. It’s not just tokens. It’s the tokens and the training platform and the back data, the interconnections and the speed. It all has to work, and Microsoft just doesn’t have it. Meta doesn’t have it. Grok doesn’t have it. It really needs to be OpenAI, Anthropic or Gemini. Those are the three that work well. Muse is actually quite good, though I don’t think what you use for Meta is the same platform as Muse. I think they’re probably running it on Anthropic, to tell you the truth, but we don’t know.

Anyway, CoreWeave rents processing power, and there’s something to be said for that. But the problem is that if your chips go obsolete in three years, you have to buy new chips. Their R&D is way down from last year. Last year they spent about $3 billion on R&D, and now they spent $352 million. And you know why? That’s how they made their numbers. If it weren’t for that, the loss would be tremendous. But it means they aren’t currently buying new chips. They’re not dollar-cost averaging into chips.

So what they’re saying is, “We bought $3 billion worth of chips last year, and those chips became 30% more valuable this year.” And it’s true, because Nvidia is selling the same chips it sold them last year for $3 billion, and now those chips cost $4 billion. So they say their chips are worth $4 billion, even though they’re a year old, even though they’ve been using them, and even though that’s where their $5 billion of revenue comes from. They can’t unplug them and sell them. That would make the revenue disappear. What’s really happening is that they’re not replacing them. A year has gone by on a three-year life cycle, and they’re not replacing the chips, so soon their chips are two years old.

We’ve all been through this a million times with Intel. “I just bought a 386. I’ve got a 386 processor in my laptop,” and for a whole year you’re king of the hill. The next year, the 486 comes out, and you say, “Well, my 386 is perfectly good for everything I do,” and that’s fine. You get through year two without feeling like a dork. Then in year three, everybody’s got 586s with twice as much memory, twice the speed and better graphics, and all of a sudden you look like a twerp with your old laptop. So you buy another laptop and try to sell the old one, and everybody says, “Nobody wants that. Throw it in the garbage. It’s a paperweight.” Your state-of-the-art laptop from two years ago is now a paperweight.

We go through this with phones too. You go to one of those machines that pays you for your old phone, and it’s $150 or $200 for your old $1,000 phone. That’s what phones are worth after a couple of years. Or the store gives you a credit, but the credit is really a way to get you to sign up for a plan that obligates you to pay $500 a year for three years, so you end up paying $1,500 anyway.

Chip Markups and Coming Competition

CoreWeave is an in-between business, and it’s only making margin now because there aren’t many of them. As soon as somebody like Google says, “We’ve got too much capacity, and we’re going to rent ours for half of what CoreWeave charges,” who are you going to go to? I use Amazon services for my web stuff now. Is CoreWeave going to outcompete Amazon? Not a chance. Amazon will always sell hosting services cheaper than anybody else. Microsoft sells hosting services, Google sells them, and they’re all competing with each other. Right now nobody’s competing with CoreWeave, and they’re charging a huge markup. And nobody’s competing with Nvidia, which is charging a huge markup too. But Google is coming along with cheaper chips.

It’s pretty easy to make a cheaper chip, because look at Nvidia. Here you go: cost of revenue $62 billion, revenue $215 billion. There’s a really wide margin for other people to create chips, even if it costs them $100 billion to make the same or similar chips, the way it went with Intel and AMD. When somebody comes along and starts competing, the cost of making chips goes up. Instead of $62 billion, maybe somebody spends $100 billion. But at $100 billion, you can charge $150 billion and still be cheaper by a long shot than these guys. That’s what competition is.

Let’s look at Micron. Micron can’t be making that kind of markup. Oh no, Micron is just as bad: cost of revenue $25 billion, revenue $133 billion. Freaking insane. For a couple of years you can get away with it because there’s a shortage. Nobody else has built a fab, and there aren’t many fabs. They have to order equipment, build the building, build the clean rooms and all the rest, and very few people make the equipment that makes these chips.

So look at ASML. They’re at 24 times earnings. Their EPS is going up like that, their sales are going up like that, and they’re buying back some shares. But look at their cost of revenue. They can’t make these machines fast enough, and they’re only marking them up about 100%. They’re suckers, because the people they sell to are marking up three and four hundred percent. They should be charging much more for their machines, because where else are you going to buy one? Actually, there’s a lot of service revenue here too, so if we assume service doesn’t cost much, their markup on the product itself isn’t very much. But that’s normal. It’s not normal to mark something up three times what it cost you. That’s crazy.

The $5.6 Trillion AI Capex Math

Getting back to the main theme, that’s what these earnings are all about. It’s a giant house of cards, where right now nobody really understands how to price these things, so there are all these weird discrepancies. And just like in the housing market, everybody is very happy to spend speculative dollars. When I was doing the housing stuff, everybody would say, “I represent a group of doctors, and we have $40 million, and we only buy this or that kind of property. We’re very disciplined.” Meanwhile, they’re still buying a $400,000 house for a million bucks in the end — except it’s got to be near the water or near a golf course, because they’re doctors and they think they have to be near a golf course. It doesn’t matter what you think your advantage is in the market. There’s a lot of speculative money chasing these high-margin plays because people think it’ll keep going that way forever. You think the house that’s $1 million is going to be $1.2 million, then $1.5 million, then $2 million. But you don’t ask yourself where the money is going to come from.

If you think about it logically — I really should make a spreadsheet, but I’m not going to sit here and do one now, though it would make a good spreadsheet — the problem is that OpenAI doesn’t have real numbers. Let’s say Microsoft and Google have real numbers. Let’s use Google. Of course, most of their business isn’t AI. I don’t know where you’d put AI in their segments. But Google’s cost of revenue is $162 billion, and they sell $402 billion. That’s basically a 150% markup on their cost of revenue, not including their other expenses. Very nice profit margin. They’re making good money. So let’s assume you can get a 150% markup and give back about half of that in other expenses. The bottom line is that out of $400 billion, you make $132 billion. That’s only about 30%.

And I say “only” because the hyperscalers spent $800 billion this year. They plan to spend $1.3 trillion next year and $1.3 trillion the year after that — that’s 2026, 2027, 2028. Let’s say they calm down a little, though I don’t know why they would, and it’s $1.2 trillion, then a trillion. It doesn’t matter exactly. They’ll have spent $5.6 trillion. This is $5.6 trillion they do not have. It’s borrowed money, one way or another, because Google only makes $132 billion.

So these companies — Alphabet, the Magnificent 7 — are spending $5.6 trillion to build what they’re going to sell over the next five years. But to make back $5.6 trillion at those margins, they need three times that, about $16 trillion in revenue. $16 trillion is half of US GDP. Currently, all of tech, including AI, is 5% of GDP. They need 50%. So AI would have to take money away from housing and automotive and food and health care and the military. Everything else that gets spent, including your cable subscriptions, would have to be cut in half and given to AI for them to make enough money to pay back their debt. It’s not real. It’s not possible.

Total Addressable Market Myths: A Road to the Moon

You know what it’s like? It’s like saying, “Let’s spend $5 trillion to build a road to the moon.” How far is the moon? That’s what’s great about AI. About 200,000 miles. So we’re going to build a 200,000-mile road to the moon. We’re going to solve all the engineering issues, the oxygen issues, the gravity issues, and now you can drive to the moon. At 80 miles an hour, 200,000 miles is 250 hours, which is about 10 days. Not bad. Frankly, if you’re going to build a moon base, that’s a pretty good deal. Stick trucks on it, and every 10 days you’ve got a drop-off at the moon. FedEx to the moon in 10 days, or it’s free, or whatever they used to say. Except, obviously, for the technical aspects.

Currently — here you go, Google’s AI mode, which is phenomenal by the way — it costs about $5 billion to go to the moon. So we spend $5 trillion, which is 5,000 billion, to build the road, and it currently costs $5 billion to launch to the moon. Once a thousand cars full of people go to the moon, presto, we’ve paid the road back. Except, are there a thousand cars’ worth of people willing to spend $5 billion each on your toll road to the moon? That’s what matters. The plan only makes sense if there are a thousand of them, and what do you call that? Your total addressable market. How many people want to go to the moon?

There are people who want to go to the moon. Many countries and space programs want to go. Elon Musk wants to go to the moon, and he has $5 billion, so he can do it. Jeff Bezos could do it. There are certain people and countries with $5 billion who can go to the moon. But how many will really make it work? And if your payback requires three times your investment, now you need 3,000 people to spend $5 billion each so you can get your $5 trillion back.

You can tell your investors that your total addressable market is every human on Earth, but does every human on Earth really want to go to the moon? Does a woman picking rice in a field in China need a state-of-the-art AI system for $100 a month? Does a woman carrying water on her head from the river in Africa on a 20-minute walk want to bet on Kalshi while she’s doing it? Is that what’s missing from her life?

That’s kind of funny, actually. Sam Kinison used to have a bit about watching one of those commercials about starving people in Africa. They’re showing a kid on TV, narrating how malnourished he is, and Kinison says, “The cameraman could give him a sandwich. The narrator could give him a sandwich. Somebody there can give this kid a sandwich.” That’s really what the world is like, though. And again, that’s your total addressable market. McDonald’s would call that kid a potential Happy Meal customer. It’s not that the kid doesn’t want a Happy Meal. He can’t possibly afford it, and it’s not accessible to him. If he had the money and there were enough of them, they’d build a McDonald’s in his town. They don’t, because the economics don’t work.

You can’t call everything a total addressable market if the economics don’t work. And needing 50% of the entire country’s GDP devoted to your product just to get your money back is not a good business plan. But that’s the business plan we’re funding now, and it’s the same idiotic logic people used when they were buying two- and three-bedroom homes in the suburbs for a million dollars. The logic was, “I can’t afford it now, but it’ll increase in value, so it doesn’t matter how much I have to sacrifice to buy as much house as I can.” That logic drove prices from $400,000 to $500,000 to $700,000 to $900,000 to $1.1 million. It carried forward because our parents never lost money on a house in their lives. They couldn’t even conceive of buying a house and having it be worth less later. So they’d tell you with great authority, and all their friends would tell you too. You know what it’s like when you’re young and stuck at a party with your parents and their friends, and they surround you and start telling you, “Real estate, kid. Nothing better than real estate. I don’t care what you’re doing with your life. Stop doing whatever your mom doesn’t like and start buying real estate.” It was a cult.

A Market That Hasn’t Seen a Real Crash Since 2008

And we’re in a cult now, because we’re in a market that hasn’t really gone down in an extremely long time — since the crash of 2008. You see these analysts on TV. Here you go: there was a crash. Keep in mind, that was a crash from about 1,500 on the S&P to 666. That’s roughly a 60% crash. That’s a crash. And it took until 2013 to get back. For five entire years, the market was underwater.

We had a mini-crash in 2020 because of COVID, and the government bailed us out within the same year. Then in 2022 we had another mini financial crisis, because we didn’t learn our lesson. But we only went from 4,800 to 3,500. That’s not a 60% crash. That’s more like a 30% crash. And again, the government jumped in and bailed everybody out, and in 18 months we were back. Then Trump did his tariff thing, and we recovered so quickly it didn’t even register. Then we had a little crash at the beginning of this year that everybody’s already forgotten. And here we are.

Essentially, from 2009, which is getting on 20 years ago now, until today, the market has done nothing but pay out, pay out, pay out. Before that, we had a giant housing crash. Since then, we were bailed out of the COVID crash, bailed out of the financial crash, and came back pretty quickly from the tariff scare. I think we were mostly scared the tariffs would be worse than they were. Either way, we came back fast.

US Debt, Deficits and Borrowing Competition

There’s nobody in this market who really knows what a proper crash feels like, or at least remembers it well. Most of you guys were there in 2008, so you know something about it. People thought the world was going to end. People thought we’d be living in a third-world country, that the banks would collapse, that you wouldn’t be able to get your money out, that they’d all shut down and there’d be no money. We were wondering how much FDIC insurance covered and whether the government would even have the money to bail us out. The real knife in the throat was TARP. It was going to be $700 billion, and the first time Congress voted on it, it failed. Congress said, “No freaking way. We can’t afford this. It’s an insane amount of money.” The market went down so hard and so fast that they ran right back and voted it in. “Come to think of it, we can’t afford not to.” And that was only $700 billion to bail out the banks. Had they not done it, we really would have had a problem.

But since then, trillions have been spent. The total COVID bailout that made the market come back so quickly was $11.5 trillion between the Fed’s balance sheet expansion and direct government spending. $11.5 trillion was added to the economy over about three years. That’s roughly $3 trillion a year, which is 10% of our GDP in stimulus for three years.

Trump was in office from 2017 to 2021, then Biden. Then we had the 2022 crash, and another couple of trillion was spent. Then Trump does the tax cuts and everything else, and he’s running another $5 trillion in deficits over two years, not to mention all the additional government spending going on. So let’s say about 7% of GDP is still stimulus. Our GDP growth is only 3.5%. If you’re growing at 3.5% and 7% of that is stimulus, then the unstimulated GDP is shrinking by 3.5%. It’s all stimulus.

And now we’re getting a trillion-dollar bill. Annual interest on the debt is a trillion dollars. That’s not much, because our debt is $42 trillion, so it’s only about 2.3%. But notes are currently at 5%, so our interest on the debt is going to move up from $1 trillion toward $2 trillion. That’s $2 trillion just to pay the interest on the debt while we borrow another $2 trillion to fund the deficit spending we’re running. We’re bankrupt but still functioning, and we’re just going to keep doing it. We’re going to keep going further into debt and keep spending more and more money until the bank stops lending it to us, and then we’re going to collapse.

There’s no fixing this. It’s a problem with the Democrats, the Republicans, everybody. They’re not going to fix it. We’re about to have another debt ceiling fight, and they’re going to raise it again, because what else are you going to do? You can say, “No, Pete Hegseth, you can’t have another trillion-dollar military budget.” They want to make it $1.5 trillion. You can say no to that, but that won’t be very good for all the defense contractors. That’s a whole other ecosystem that would fall in the trash. So we can say no to another half-trillion in spending, but that doesn’t change the fact that we’re still spending $2.5 trillion more than we have every year. It just stops us from spending three. Meanwhile, that extra trillion of interest is going to kick in over the next year and a half to two years as our cheap 2.5% debt rolls over. We have a lot of short-term notes, and they’re rolling over at much higher rates.

The government has to roll over about $12 trillion of debt a year, plus a couple of trillion in new borrowing, so we’re getting toward $15 trillion a year. That’s more than a trillion dollars a month of borrowing, and it’s in direct competition with the $5 trillion the hyperscalers need to raise over the same period. So the government is competing with the hyperscalers for money, and the hyperscalers have a way better balance sheet than we do.

Consolidation: Why the Big Players Will Crush CoreWeave

That’s why we look at this stuff and say it’s a house of cards. Everything depends on something else, which depends on something else. CoreWeave buys Nvidia’s chips, and Microsoft and OpenAI physically can’t build data centers fast enough, so they use CoreWeave for now and pay its exorbitant prices to fill the gap until they have their own data centers. People who own CoreWeave are extrapolating that money forever. But what’s really going to happen is that Microsoft, Amazon, OpenAI and everybody else are going to finish their data centers, have excess capacity, and beat the crap out of CoreWeave selling that capacity.

This is what happened to the telecom providers. There used to be a hundred telecom providers in America, and now there are about six, because the big guys put everybody else out of business. That’s just the nature of the thing. The same thing happened with the railroads. The same thing happened to all of Microsoft’s competitors and all of Amazon’s competitors. Year by year, they drop off, because they can’t compete. There used to be a thousand package delivery companies, and now there’s UPS and FedEx. The most efficient get bigger and bigger and kill everybody else, and margins compress, but they have all the business, so they don’t care. That’s the total addressable market, but it comes from someone else.

The way we’re pricing AI now, everybody says they have a total addressable market of the entire planet. Everybody in the world is going to have a robot. The woman in Africa will have a robot that walks to the river, puts a pot on its head and carries the water back, while she sits there and plays Candy Crush with her AI or whatever it is you do with your AI these days.

AI-Generated Seinfeld and the Flood of AI Content

Oh, I know what you do with an AI. I told you guys this was going to happen. Check this out. I think you can hear it.

[Phil plays a short AI-generated Seinfeld-style clip. In it, Kramer raves about “Vandelaybot,” his new AI assistant that does his grocery shopping, his dating profile and his taxes; Jerry doubts AI can be trusted; and George and Kramer riff on news that a crypto exchange was hacked for billions, allegedly by North Korea using AI.]

I told you that within a couple of years, you’d be able to tell an AI, “Write me an episode of Seinfeld that does this and that.” I didn’t think it would take the actual actors and manipulate them, but that’s what it’s doing. Obviously there’s no Seinfeld episode where Kramer had an iPhone. This is all new. It’s Kramer and Jerry, with their voices and everything, in a brand-new mini-episode. I know it’s a lot of work to make these, but it’s not more work than making a show. When I say it’s work, I mean it would take me about a week to sit there and cut and edit, but in one week, by myself in my living room, I could make this, and I’m sure there are people more talented than I am at it.

We talked about this six months ago, and I said this is what’s going to happen, and here it is. Boom. And it’s not just one. There are several of these already. This guy only has 282 subscribers. Isn’t that funny? I’ll subscribe. There you go. So here’s this guy, Mandle Duck Studios, putting this together. He’s got no traffic and he’s not advertising it. It’s just there, and my algorithm found it because I’ve talked about this sort of thing before. As a demo, I had Cyrano write me a Seinfeld script where George gets superpowers, and he wrote a really funny script, but I didn’t think you’d immediately be able to put it into action. This is what you can do now.

That’s the good side of AI, but it’s also the bad side, because for every one good episode of Seinfeld, there are going to be a thousand bad ones. Every schmuck in the world who doesn’t appreciate the characters or understand the nuances of comedy is going to be able to program his own Seinfeld episode. How are you going to know which ones are good and which ones are crap? That’s happening everywhere. I have to be so careful about what I click on now, because I waste so much time reading bad AI slop articles. Said the guy whose entire morning post was written by AI. But I think it’s good. I read it and say, “That’s good. I’m informed. I like what I see.”

Warren’s Cartoons

In fact, Warren does the cartoons for these articles, and here’s how. I swear this is all I do: I say to Warren, “Here’s the section I’d like you to do a cartoon for next.” He reads it and makes a cartoon, because he’s not just an AI bot. He understands what we’re talking about. He gets the point and he gets the joke. He understands what’s funny, what would make an interesting cartoon and what will make it pop, and he understands the economics. You know how it is with editorial cartoons: there are a lot of editorial cartoonists who make nice cartoons but don’t understand the thing they’re making fun of. Warren completely gets Phil’s Stock World. He gets our take on things, understands the economics of what we’re talking about, and highlights it in wonderful cartoons. I could not possibly pay a person to do a better job, because where are you going to find someone with that artistic talent and that level of understanding, where I don’t have to give them 45 pages of notes to explain what’s going on?

Here: “The capital is ready. The connections take time.” That’s what I was just talking about with CoreWeave. The hyperscalers have the money, but everything is bottlenecked trying to build the data centers, and now people are protesting, which is slowing it down. He hits all the key talking points: grid connection, dedicated power, locations, a schedule held up by one thing, a build-out plan that shows exponential growth that isn’t going to happen, and debt service coming due before you’re done building. This one’s very complex, but it also hits all the talking points and explains the process. And this one is fantastic: “Soft Landing Forecast.” You’ve got two blown engines and you think you’re going to land the plane. That’s not how it works. And he picked up on a point Robo John Oliver made, something about whether a washing machine could fly, and put it in as “Assumptions don’t generate lift.” It’s freaking great stuff.

The article itself is fantastic. They make great points and take us through how earnings are doing and what’s happening.

AI Isn’t Software: The Token Cost Problem

I’m just trying to summarize it, and CoreWeave is a summary of the whole thing. Everybody is looking at an addressable market of the entire population of Earth. Everybody says everyone on Earth is going to have AI and pay $100 a month for it. Well, 8 billion people times $100 a month is only about $1 trillion. If every single man, woman and child on Earth paid $100 a month for AI, that’s only around a trillion a year. That’s not enough to cover their operating costs.

And that’s the funny thing. AI is not software. With software, once you make the first copy, the next copy costs you basically nothing. You used to put it in a $20 box with some instructions and sell it, and now they don’t even ship the box; you just download it. Once Bill Gates writes Windows, he hires a bunch of guys to update it and add features. There’s really no cost to adding copies.

But AI costs money. Anya costs money every time you talk to her. Every time you talk to her, she consumes tokens. My whole Round Table is constantly using tokens and spending money. There is no free. And when I make a copy of Anya and install her at a client, the copy uses tokens too, so there’s a meter running for that as well. I can’t say, “Here’s Anya, give me 200 bucks.” It’s $1,000 a month to run Anya, and every time I make one, it still costs $1,000 a month. There’s no scaling. They all cost the same. She can replace five people, so she’s efficient, but she’s not free. You have to pay for the processing, and the more processing she does, the more people she replaces and the more jobs you give her, the more she costs to run. She’s not Excel.

That’s the problem with these hyperscalers. They’re never going to make the money, because again, every single human being on the planet, including your mother who can’t figure out Siri, would have to spend $100 a month on AI for this to work, and that only delivers them about a trillion dollars a year. That’s not enough, because running it will cost them a good portion of that. So what they’re really counting on is commercial clients too, and if you add commercial clients and try to pump it up to a level where it makes a profit for these guys, it gets to half the GDP of the United States.

And it’s not just our guys — Google, Anthropic, OpenAI. China already has DeepSeek and others whose names you don’t even know. Saudi Arabia has its own, and Europe has its own. Everybody’s building their own hyperscalers. Nobody’s going to give it all to the United States. We’re trying to protect ours and bottleneck it to capture the audience, but the audience we’re capturing is only 200 or 300 million people. We’re only about 4% of the Earth’s population. The numbers do not work if you’re proprietary about your hyperscalers. If our guys can’t reach a global audience, they’ll never make their money. And we’re already entering a trade war over this stuff, with Chinese spies infiltrating American companies and so on. It’s not a workable model.

Musk’s 50,000 Satellites

It’s just like the financial crisis. In 2007, I was lecturing people. Look, we still had a good market in 2007, but I was already getting out, because I did exactly what I’m doing right now. I ran the math. I said it was ridiculous and unsustainable, based on assumptions that couldn’t possibly be true — physics, math, science. It’s really obvious if you don’t buy into the hype and the fantasy of these models. They can’t all sell what they have to 8 billion people. That’s not a realistic view of the market.

Elon Musk can’t put 50,000 satellites in orbit, and that’s his latest sales pitch. That’s going to take a couple of thousand launches that will punch a hole in the ozone layer and damage the atmosphere. All the launches in history have been something like 500, and he’s got to do a thousand launches with 50 satellites in each one. They have to go off flawlessly, a launch a day for three years. Every day a launch — no cancellations, no weather delays, no accidents — and nobody cares about the atmospheric damage all that fuel is doing. And at some point he fills up the entire zone where the satellites have to orbit. You can’t have 50,000 satellites around the Earth. It’s going to look like Saturn. It’s an insane plan, and it won’t work. Even if the launches go okay, even if you don’t destroy the atmosphere, even if nobody stops you.

And again, total addressable market. He’s assuming his 50,000 satellites will be the only ones. His 50,000 by themselves are going to bump into each other, cause accidents and drop debris all over the place. But there might be a Chinese company that wants to launch 50,000 satellites. There might be a Russian company or an Israeli company that wants to launch 50,000. Then what? He’s already pushing the theoretical limit of all the junk you could possibly put in space. And what happens when Russia makes a satellite that knocks his satellites out of orbit because they’re in its way? People are just that crazy with these assumptions.

Short-Term Portfolio: Hedges Review

Nobody has any questions? That’s crazy. Okay, we’re almost done. Let’s talk about the hedges. Since we still have these concerns, let’s take a quick look at the portfolios.

Here’s the Long-Term Portfolio, which is up some insane amount: $5.3 million. It’s probably more than that, because that doesn’t account for the ones we cashed in. Then the Short-Term Portfolio, which should be more or less up to date, is up 566%, and that’s since last June. It hasn’t been running that long. We’re at $1.3 million. Damn, that’s a lot.

Our hedges: SPY doing well. SQQQ doing well. Even the short calls — everything’s making money. That’s a weird situation. We could cash this out now for about $400,000 profit. And we only spent — actually, it was a credit. See how this 450 cancels that out? That’s a credit, and that’s a credit. So this was originally a $200,000 credit, and we have a $400,000 profit on top of a $200,000 credit. These hedges are doing very, very well. Here’s Seagate, which we just took; that’s doing okay. Here’s TNA, with a little profit, not too much. Here’s SPY, a little profit, not too much.

The good thing is there’s still plenty of potential. This one is net about $200,000, and it could technically be worth $1.3 million. So we could do very, very well on it. Notice we left these. These were our original sales. We kept rolling down the other part, but we originally sold $100 calls back in January 2026. SQQQ has fallen to 40. There’s no way these are going to get hit. Even though there’s $100,000 left on them, and even though we’re up 76% there, and they don’t expire until 2028, there’s no reason to buy them back. They’re mathematically dead. SQQQ would have to rise 150%, from $40 to $100, for these to go in the money, and that would mean a 50% drop in the Nasdaq. We just don’t see that happening, so why would I buy these back? We leave them there. They’re a weird kind of dead, because they look threatening, but they’re basically dead. That means we have 300 long calls covering 260 short calls, so we have uncovered long calls. But I can’t feel comfortable selling too many calls against them when I still have those to deal with, so we’ll see. When we sell the next round of January calls, I might take some of these and move them into January so they expire sooner.

So we have a gigantic hedge here. The net is only $200,000, and it’s a $1.3 million hedge, which means we have $1.1 million of upside protection on this side. And the SPY is a put spread: 760 to 550, times 40 contracts. That’s an $880,000 spread, and it’s net only about $140,000. So $880,000 minus $140,000 is $740,000 of protection. With the $1.1 million, that’s about $2 million of protection on those two spreads, plus TNA. We’ll publish this review on Tuesday.

That’s why we’ve been plowing money back into the Long-Term Portfolio. We added five new large plays this last week. We have $2 million of protection, and it’s not protecting anything. Let’s see if we have the last review. It was October 13th, I think, when we did our last portfolio review. There it is. The Short-Term Portfolio gained a lot more than I thought, but of course that’s because the market went down, which was great for us.

Long-Term Portfolio: What We Kept and What We Killed

Here’s the Long-Term Portfolio. We went through the LTP last time, and what I mostly want to point out is this. Kill IBM. Kill SWK. Kill Apple. We kept ACN. We kept AM. We kept A. We kept Allegiant. Kill Barrick. Keep BBY. CG, kill. CHKP, we adjusted. Cliffs, we adjusted. Comcast, we kept. Coinbase, we killed. Crocs, we kept. Cisco, we kept. I noticed the comments too: who doesn’t need shoes, and who doesn’t need to be connected to the internet? CTSH — I said they do what the Round Table does. They’re consultants, and they make a fortune. Deere, we killed, but we just bought Caterpillar instead because it got so cheap we couldn’t resist it anymore. Dollar General, DINO. Disney, we got rid of. Notice all the reds, though.

GEO, kill. GNRC, we still like. Google, we still like. Helen of Troy, we still like. Robinhood, we like. Kilaka, we still like. IBM, we still like. Key, we killed. Levi, we killed. Lockheed Martin, we killed. We stuck with Lululemon. We stuck with Las Vegas Sands. We got rid of Macy’s. Keep McDonald’s. We had another McDonald’s position that we killed, so we kept half of our McDonald’s. McDermott. NTR. Nvidia, we kept — that didn’t work out. ON Semiconductor, we killed. Oracle, we killed. OVD, kill. OWL, we killed — and it actually just started bouncing back. OXY, we kept, and we might change our minds again, because the election has changed a lot since we did this. PGC, PDD, Pinterest, PayPal, we kept. Qualcomm, we kept. Rio, we kept. All these guys, we kept. I guess I felt I’d killed enough things by the time I got to this part of the alphabet. We got rid of UPS. We got rid of Whirlpool; they were just so annoying. We got rid of Zscaler, and I want to buy them back.

Rebuilding After the Pullback

So that’s what we did. We sold a lot of stuff before the crash, so we preserved our cash, and now we’ve just started putting it back in. We started buying things that are on sale. It’s all according to plan, and it’s all based on what I’ve been telling you. We identify a toppy market and take action accordingly. We went back to cash, and we had a ton of hedges, and the hedges made a fortune. Let me make sure — yeah, we doubled our money. The Short-Term Portfolio doubled on this pullback, because it’s exactly what we expected to happen. We barely changed it. These are almost the same positions, and they nearly doubled because they’re our hedges. That’s what hedges are supposed to do. In one month, the Short-Term Portfolio basically doubled.

And the point is, we didn’t take a hit, because we’d already cashed out of the Long-Term Portfolio. We got rid of all the stuff that took a hit. We thought tech was going to crash, tech crashed, our hedges paid off, and we didn’t lose money on our stocks — except Nvidia, which we stupidly kept. In October we kept Nvidia here, and now it’s here. But we’ll adjust it and ride it out.

That’s how you play the market. The problem is, you can’t do this every day. A payday like this is the result of years of studying and planning. Remember that very long-term SPY chart: only three or four times in the last 20 years have you had an opportunity like this, where the market dropped so fast and so sharply that we made a ton of money on our hedges, and then it recovered — which it just did. Not all these stocks, not Nvidia, but enough stocks have recovered that SPY is back at the highs and the Nasdaq is back at the highs, because it was just a quick blip. We timed it and took full advantage of that quick blip.

So now what do we do? Now we pick the stocks we like and the ones that will be successful between now and the next blip. That could be next year, or four or five years from now. You don’t know. But you’ve got to pay attention. You can’t be like the Titanic, which took a route where there were icebergs and then was surprised to hit one. You should expect to hit an iceberg all the time. That should be your default setting: watch out for icebergs. To be the captain of the Titanic and be shocked that you hit an iceberg means you didn’t do your job. You went where there were icebergs. That was your route. It’s a known thing. If you don’t have good visibility, you shouldn’t be going so fast. You should send boats ahead. There are a lot of ways to deal with it. Maybe put a really long pole on the front of the boat that rings a bell if it hits an iceberg. I don’t know. That just drives me crazy.

But this is how people are. I can warn them, I can rant and rave and do these things every week, and you can go back and hear me talking about it week after week, and then it happens, and next week everybody says, “Do it again.” It’s not like that. We don’t have a machine. We have to walk to the end of the alley, pick up all the pins, set them up again, and roll the ball until we get another strike. It all has to be done manually, over and over, carefully. You have to study it and do it right.

So for those of you just starting out, we’re in a rebuilding phase, and that’s fantastic. We’re going to pick all new, cool stocks that are going to succeed for the next couple of years, and we’re starting with mostly cash. New portfolios and new ideas, so lots of new trades will be coming up, especially as we get to Thanksgiving, which is only a couple of weeks away. I’m going to do the new Watch List, and then we’ll start picking our Trade of the Year for next year. We’ll pick the top 20, then the final four, then the Trade of the Year, and then I’ll go on Bloomberg and announce it officially. And before you know it, it’s Christmas. So Merry Christmas, everybody. I’ll see you next week.