Anthropic finally dropped its S1 and there were some very important data points. I think a clear picture emerges and it is what I have feared. A whole bunch of companies reported across multiple industries. But with respect to tech, this week's results raised troubling questions. Google, no way to sugarcoat this one. The market did not like these results at all. Intel and Amazon reported and the results weren't great and the stocks got rocked, and the market is now going to question whether the growth in AI capex is over. There are just too many variables to predict whether Anthropic can deliver profits to investors. Let's let that sink in as we dive into what Anthropic released and what they are perhaps hiding. So, let's get started.
Hi, this is Steve Eisman and this is another episode of the Real Eisman Playbook. This is for the week ending Friday, October 30, but recorded Thursday night, October 29th. For Substack Premium this past Wednesday, October 28th, we posted an interview with the Substack newsletter writer and podcaster, Doomberg. The global energy markets are in a chronic period of undersupply driven primarily by the hot wars around the world that are often deeply intertwined with energy production. You know, we believe World War II started in 2014 for reasons we can talk about. Focusing on oil, the war in the economy. This coming Wednesday, November 4, we will release an interview with Substack newsletter writer Herb Greenberg, someone I've known for a very long time. We had a wide-ranging conversation about the market and faulty narratives and we hope you tune in. If you like the work we are doing, please visit us at steveisman.com and subscribe to Substack and YouTube.
On this week's wrap, we will discuss some more thoughts on France. Anthropic dropped its S1. Finally, a very long list of companies reported and their implications. Some more thoughts on Core from a viewer and thoughts from some viewers about the problems getting power to AI data centers and also about the telepocalypse. Let's get started. I've spoken quite a bit about France over the past few weeks and the problems with French sovereign debt. No need to repeat everything again. I just want to add one point. Size. US sovereign debt is 40 trillion. French sovereign debt is only 4 trillion. Since both countries have a debt to GDP of 120%, simple math dictates that US GDP is 10 times the GDP of France. On that basis alone, France just does not count all that much in the world of sovereign debt. It's one more reason why no one needs to own French sovereign debt. And that's partially why rumblings from some French politicians that France does not have to pay back the debt is ridiculous. The size of the capital markets has exploded over the last several decades. Today, for example, BlackRock manages $15 trillion. BlackRock is bigger than France. France is just too small to threaten anyone in the fixed income markets. France needs BlackRock's liquidity way more than BlackRock needs to own French debt.
Let's let that sink in as we dive into what Anthropic released and what they are perhaps hiding. Anthropic finally dropped its S1 and there were some very important data points. For the first six months of 2026, 25% of revenue was from the top five customers, which is not bad. Anthropic has 6.22 billion in debt, which is also not terrible. The adjusted operating margin in Q2 26 was a positive 19%, which is quite good, but—and this is a key point—that was at the height of token maxing, about which the prospectus is silent. Now I have two major takeaways from the prospectus. First, over the next several years Anthropic has made compute and equipment commitments with the following, and I'm going to list them: Microsoft 45 billion, Google 115 billion, TPU providers including Google 150 billion, Amazon 110 billion, SpaceX 121 billion, AMD 20 billion, NScale 45 billion, Lambda 35 billion, Broadcom 161 billion. It all adds up to 800 billion in future commitments and that is enormous. There is no margin for error here. These numbers suggest that starting in 2027, Anthropic's compute costs could explode.
But even more important are the trends in revenue and margins. And I'm putting up a table that shows Anthropic's quarterly revenue from the third quarter of 2025 through the third quarter of 2026. Now the prospectus does not show third quarter numbers, but they are already in the public domain. All the commentary in the traditional media was about how great the numbers look through the second quarter, which they do, but it's the third quarter that I think really counts. I also show the company's compute and infrastructure expense during the same periods and I then calculate Anthropic's efficiency ratio, which I define as its revenue divided by the compute expense. And I also show the sequential percentage change in quarterly revenue. Now for those of you who are only listening, I think a clear picture emerges and it is what I have feared. During the first half of this year, I thought that Anthropic and OpenAI were benefiting a lot from token maxing. That was where enterprises told their employees to use AI and to hell with the cost. But that ended this summer when costs got out of control and companies now had the option to switch to cheaper open-weight models. So I assumed that once token maxing ended, growth would slow. Still there would be growth but slower. And I think that is what these numbers seem to show.
If we go back to the fourth quarter of last year, just before token maxing, Anthropic's quarterly revenue grew sequentially by 43%. Its revenue divided by compute expense was 0.69 times. Then token maxing took off and so did Anthropic. Q1 26 and Q2 26 sequential revenue growth jumped to 135% and then 147% respectively. By Q2 26 the revenue compute expense ratio had jumped over two times from 0.69 times in the fourth quarter last year. So this all looks great. But now we look at the third quarter. Token maxing is over. Anthropic is still growing but slower. Q3 26 sequential revenue growth was 45% versus 147% in the second quarter. Maybe even more alarming, the revenue divided by compute expense ratio in Q3 declined to 1.70 times versus 2.08 times in the second quarter. Growth has slowed.
It's also important to point out that Anthropic cut its token pricing by as much as 50% in October so as to spark demand. According to OpenRouter, Anthropic's token demand has increased, but I doubt enough to offset the price cuts. One school of thought is that Anthropic's strategy is to get their product deeply embedded into enterprises and then jack up the price. Also, if data center openings are delayed and demand stays high, prices will be artificially inflated until compute supply is fully available. There are just too many variables to predict whether Anthropic can deliver profits to investors. What does seem clear is that the IPO will go through and the narrative will gain some steam again. OpenAI is sure to follow soon and Mercury will be out of retrograde on November 23rd, which some viewers assure me matters the most for us investors. We shall see.
Last week, Meta reported and announced that it had over 800 billion in AI-related commitments. This week, to fund those commitments, Meta announced it was raising 95 billion in equity. 35 billion right now and the rest later. Nvidia announced that it was buying 3 billion of that amount as a sign of support. And now for this week's earnings reports. A whole bunch of companies reported across multiple industries. But with respect to tech, this week's results raised troubling questions. They raised questions as to whether the AI capex cycle is slowing. They raise questions as to whether the SaaS apocalypse is back. The stocks reacted terribly. So, let's get started.
First up, Boeing. Boeing reported, and this has been an incredibly long turnaround story. Boeing reported a loss of 34 versus a loss of $747, which sounds great until you realize that they missed expectations of a loss of 21. Revenue climbed a pitiful 4%. The star performer for the company, unsurprisingly, was its defense unit, which saw revenue jump 15% to 8 billion. But the all-important commercial aircraft business saw revenue fall 3% to 10.7 billion. This company is still in turnaround. It is still struggling and the stock was down 10% on Tuesday because of the miss and because management provided disappointing cash flow guidance on the call. The stock is now down over 20% this year.
Moving on, Visa reported. Now, we all know that the bottom of the K is struggling and the top of the K is benefiting from a long bull market. Visa's results show that the top of the K is still carrying overall consumer spending. Visa reported earnings per share of 348, 17% growth and a slight beat. More importantly, payment volume was up a strong 9% and it's the payment volume data point that tells you about the health of the consumer. Mastercard also had good numbers, although not quite as good as Visa. Dollar payment volume increased by 7%.
S&P also reported. Now S&P and Moody's wield a duopoly, but a duopoly that is not immune to interest rates. S&P's results weren't bad. Revenue was up 7% and earnings per share was up 9%. The stock opened up 5% on Tuesday but closed down. Why? Well, Tuesday was one of those days where investors displayed fears about the impact of AI on software companies. Those fears have subsided since the summer, but they are now reappearing. Since both S&P and Moody's have software businesses, both were down. Moving on.
GE Vernova is a big part of the AI data center story. It builds gas turbines that create electricity. There are only a handful of companies that do this. And so GEV has been doing well. The company reported earnings per share of $4.81, up, get this, 193% versus last year. Orders of 29 billion were up 96%. These results are very powerful and will last as long as the AI story lasts. The stock was down on earnings day, though. This is probably an example of a stock selling off after performing so well.
I track politics very closely and I'm always looking for an edge. With the midterm elections coming up, the key question right now is which party is going to control Congress and what that means for the economy. I find it useful to look beyond the polls and see where people are actually putting their money. So I go on Kalshi to see where the traders are setting the odds. Right now on Kalshi, the market for whether the Democrats sweep both houses has yes trading at 65% probability. Kalshi also shows how those odds have changed over time, which I find to be an incredibly important data point. Now, personally, I think that 65% prediction is not unreasonable, but a bit high because the polls generally undercount Republican voters. I can't wait to see who is right. Kalshi is a federally regulated prediction market where you can trade yes or no contracts on real-world events from elections and government policy to interest rates and economic developments. And if you want to check it out for yourself, Kalshi is offering viewers of the Real Eisman Playbook a special promotion. Trade $25 and receive $55 using my exclusive link in the description or pinned comment. Take a look at the markets and see whether you agree with what traders are predicting.
And now we come to Wednesday night. Wednesday night witnessed a torrent of important earnings reports that moved the market very negatively on Thursday. So let's get started and examine it. First up is Microsoft. Now the company's earnings themselves were quite good, but there were negative implications for others. Earnings per share of 494 was up 16% and a beat. Total revenue of 92 billion was up 19% and also a beat. Investors love to focus on Microsoft's cloud division called Azure because that is the company's hyperscaler division that is growing because of AI. And here too numbers were strong. Azure revenue of 33 billion grew, get this, 48% and was also a beat on these numbers. Microsoft was up a few percentage points after hours Wednesday night but closed flat Thursday because of a correction in many AI stocks. So what was the problem? Microsoft reported 45 billion of capex and finance leases during the quarter which was up about 30% but well below the consensus of 51 billion. On that news the entire Neocloud complex, CoreWeave, Nebius and INWT were down 5 to 9%. The Neocloud group was already suffering from much higher debt costs. This was just one more insult.
Google, no way to sugarcoat this one. The market did not like these results at all and the stock was down 12% on Thursday. There were several positive results to Google's numbers. On the positive side, revenue grew 25%. It's impressive. On the negative side, earnings per share was $267, which was lower than last year's $2.87. More on the negative side, free cash flow was a negative 9.1 billion which was worse than expectations. But even that was not the most important news. Google's search revenue was up only 12%, which was also a disappointment. By way of comparison, in 1Q26, search revenue grew 19%. Now, why is that important? The fear surrounding Google for some time has been that AI search would replace Google search. Now that sentence doesn't sound so bad until you think it through. The Google search business is a capital-light, high-moat business. It's an amazing business. It is probably being replaced by a highly capital-intensive, low-moat business. The capital intensity is the spending on AI data centers and the spending to create LLM and agentic AI models. I call it a low-moat business because switching costs between LLMs looks inexpensive. The Google search business is just a much better business. And the results last night indicate that Google search is being replaced. Yes, slowly but replaced.
ServiceNow. ServiceNow reported and the results were not good and the stock was down a whopping 27% on Thursday. Now, in the summer of 2025, the SaaS apocalypse narrative seized the software sector and caused it to correct sharply, NOW declined from 210 in July 2025 to below $100 in early summer 2026. Now the problem with the SaaS apocalypse narrative was that it's too blunt. It will be hard for AI to destroy every software company as many are embedded deeply in enterprises. So starting this summer, the software group rallied hard with NOW, for example, climbing to 159 from below 100. The software rally was so strong that some commentators argued that the SaaS apocalypse was completely wrong and the word should be retired. Then on Wednesday night of this week, ServiceNow reported third quarter numbers. Earnings per share was a dollar, up 4% but a slight miss versus expectations. Revenue was 4.1 billion up 21%. The problem was that ServiceNow provided muted fourth quarter guidance for current remaining performance obligations. Think backlog. The guidance was for growth of 18.5% versus consensus of 19.7% and versus 21% in the third quarter. This points to a sharp deceleration in future revenue. Those weak backlog projections brought back the SaaS apocalypse fears.
T-Mobile reported and while the results were fine, they were also, in my view, largely irrelevant. Earnings per share of $322 was up a strong 34%. T-Mobile added 274,000 accounts, which was better than expected, but lower than the 396,000 last year. Now, why do I say these results don't matter? Right now, there are three major mobile providers, T-Mo, AT&T, and Verizon. Elon Musk has announced that he wants to move Starlink from a satellite phone provider to becoming the fourth mobile phone provider in the United States. Now, this will be very expensive, but that has not seemed to have stopped Musk before. I call this the telepocalypse, and it is not going away anytime soon. That's why despite these good numbers from T-Mobile, the stock was flat.
On Thursday, Quanta reported. Quanta, among other things, builds new power plants for utilities. It is a major beneficiary of the AI capex story. Unsurprisingly, its results were very strong. Earnings per share of 606 was up, get this, 82% versus last year. Revenue of 12 billion was up 37% organically. Backlog reached a record 64 billion and the company also raised guidance and the stock was up 5% on this news. It's nice to discuss a company that has little to do with AI but which is doing well. Eli Lilly has clearly won the weight loss drug wars. Third quarter earnings per share was $748, up 7% and a beat. Revenue was 24 billion, up an amazing 39% versus last year and Lilly was up 7% on Thursday.
Caterpillar. Now Caterpillar has been partially an AI story. It provides generators and turbines and its heavy machinery is involved in AI construction. That is why the stock has done so well over the last few years until Thursday. Earnings results only modestly beat expectations. So for example, total revenue of 20.68 billion was up 17% versus last year, but was barely above expectations. Earnings per share of 720 was up a nice 45%, but also only slightly higher than expectations. Given investor jitters about AI infrastructure construction, barely beating was not enough to calm fears. And as a result, CAT was down 9% on Thursday, which is kind of shocking.
And if Wednesday night was not bad enough, Thursday night, Intel and Amazon reported and the results weren't great and the stocks got rocked. First, Intel. Now, Intel has been riding the AI wave. Prior to Thursday night, the stock was up 150% this year. However, this report was weak. The stock was down 18% after hours Thursday night. Why? Well, there was nothing wrong with the results. Earnings per share was 52 cents, which was up 126% versus last year and a beat. And revenue beat by a touch, too. So, what's the problem? The problem was the forecast. Intel projects that the PC slump will persist through at least the first half of 2027 and decline 10% in all of 2027 following a 12% decline in 2026. Also, Intel's numbers were hurt by supply chain problems with its data center component business. The company projected revenue of 15 to 16 billion for the fourth quarter which is below the consensus of 17 billion and is projecting fourth quarter earnings per share of only 25 versus the consensus of 43.
Amazon also had not such great results. Earnings per share were $1.89 versus $1.95 last year and versus $1.96 estimated. So, a miss and down year-over-year. Now, revenue of 202 billion was up 12%. But that also missed. Here's the key. It kept the 2026 capex budget unchanged at 220 billion. Now, prior to this quarter, every hyperscaler increased its capex budget every quarter. Now, Amazon is keeping its capex budget flat and the market is now going to question whether the growth in AI capex is over. The ramifications of that are very broad-based and extend into every part of the hardware tech world from servers to chips to every other kind of equipment put into AI data centers. By the way, Amazon was down around 6% after hours.
And now for the mailbag. There is no question that the problems at CoreWeave have caused a huge correction in the entire AI complex over the past few weeks. Simply put, the problem is that the cost of CoreWeave's debt has risen from 9% to 15%. Now, one viewer commented on that change and stated, quote, 'Correct me if I'm wrong, but the CoreWeave coupon is 9%.' True. The price of the bond on the secondary market is lower because of the default risk, so it's an effective 15% yield. Also true. CoreWeave is still only paying out 9% when the bonds mature. True. It's not like the bonds have a floating rate. That's true, too. Now, CoreWeave has 36 billion in total debt, but when you add its operating lease liabilities and other financial obligations, that amount climbs to slightly over 50 billion. The point is that CoreWeave's business model means that it will be building more AI data centers and to build them they will have to raise new debt and at current prices that debt will cost the company 15%. I'm not sure building an AI data center with such expensive debt makes sense.
One more thing to think about when analyzing Core. When I take a step back and think about CoreWeave's business model, I think it is most similar to specialty finance lending companies. What does Core do? Borrows from the fixed income markets, builds out data centers, and rents out the compute. What does a specialty finance company do? It borrows from the fixed income markets and makes loans. The issue is that the fixed income markets can be fickle and when they lose confidence, companies fail. I've seen this happen many times to specialty finance companies. CoreWeave is competing with hyperscalers that have much, much larger balance sheets and other businesses to support this buildout. It is playing a dangerous game. And if the fixed income markets don't regain confidence, it's a game that CoreWeave could lose.
Not a day goes by without some aspect of AI appearing on every front page. What's nice about what we do is that we receive comments on YouTube and Substack from viewers who are deep in the trenches in varied industries. I learn a lot from these comments and I received a comment from a Substack subscriber named Joseph who is an electrical contractor building a data center. From his direct experience, he believes that the binding constraint in building AI data centers and getting them online is power. Joseph says that 2027 plus will be all about who has electricity and who doesn't. GPUs, high bandwidth memory, and money, he says, are not the problems. He says that the data center builders are paying his guys $150 extra per day just to show up and that's on top of the $65 an hour wage. He believes very strongly that power is the issue. The newer transmission lines are 24 plus months out and the substations take at least one to two years. He agrees that the hyperscalers are truly building the data centers, but the electrical connections are a long way out. Quote, 'Announced capex sounds great in the media and politically, but stranded capex is what we see because so many still don't have power.' Finally, he says that Amazon and Meta will feel the squeeze more than Microsoft and Google. CoreWeave, Oracle, forget it. They are way back on the 'gonna get power next' line. He sees this whole thing slowing down not because of money, not because of GPUs or memory, not because he doesn't have a hundred transformers in his laydown yard, but because there are no power poles at all anywhere near the job site. He also thinks that Microsoft and Google are best positioned because their sites seem better planned and better managed. I thought I'd share these comments because they provide real insight into what is happening on the ground. If this subscriber is right, AI data center construction will slow down sometime next year.
Finally, as you all know, I recently coined the term telepocalypse. Starlink is attempting to create a fourth terrestrial mobile service in the US. And I spoke about this a little bit before when I was discussing T-Mobile. Now, Elon Musk has stated that that is his goal. As a result, the three telecom providers, AT&T, Verizon, and T-Mobile, have seen their stocks under assault. Wayne, a Substack subscriber, posted the following in my Substack chat. Quote, 'Telepocalypse thoughts. I have a family member that works with telecommunications satellites. He said, "Device to satellite communication requires a large number of very low orbit satellites that do not last very long in orbit. The prospect of supporting such a network is very expensive. The terrestrial cell towers will likely be much cheaper than a satellite network. The benefit would be for remote areas, but it is a small number of consumers willing to pay a premium for that. I will track this risk, but right now I think the cellular companies are safe.' Now, I have heard similar arguments and it looks like Musk may try to rent space on towers all over the country, but that too could be prohibitively expensive. However, given his history, being expensive has never seemed to stop him before. For now, I anticipate Musk will announce more plans, and that means that at least for now, the telepocalypse narrative is just going to gain steam.
This week on October 26 on our free service, we released an interview with Ben Steel, senior fellow and director of international economics at the Council on Foreign Relations in New York. We discussed inflation, the Fed, the deficit, and what is happening in France. Ben is also an author of several books which we also discussed. The book we focused the most on was his book on Henry Wallace, who was vice president of the United States under FDR and almost became president, and how that would have dramatically changed US history because Henry Wallace was a Soviet fellow traveler. Shocking. The best way to support the Real Eisman Playbook is to subscribe to Substack through steveisman.com and to YouTube. And we greatly appreciate your support. And that's the wrap.
This podcast is for informational purposes only and does not constitute investment advice. The host and guests may hold positions in stocks discussed. Opinions expressed are their own and not recommendations. Please do your own due diligence and consult a licensed financial adviser before making any investment decisions.
