We've got an incredibly important conversation today. It's going to be a debate about AI, the future of this investment cycle. Is it a bubble? Is it a tremendous opportunity? Have wonderful assets been sold at far too cheap a price, with the drawdown in semiconductors from June getting quite significant.
We've got a bull and we've got a bear. We've got Gene Munster, managing partner at Deepwater Asset Management, and George Noble of the Noble Update Substack. Gentlemen, great to have you here. I think that not only is this an incredibly important conversation, there's two people who have got very opposing views. So, even though of course this will be respectful, you know, don't hold back. Let's have this be full of lots of sparks. And of course this is a very important time with the semiconductor stocks and the AI infrastructure stocks more broadly having had a very rough day after the blue tsunami of election day yesterday. Gene, let's start with you. So let's go opening statements. Gene, what is your secular view of AI and the associated stocks with it? Where are we in the cycle and your views on the current drawdown? And then George, you'll get your opening statement.
Thank you, Jack. At the highest level, we are nowhere near the top of a bubble. I think we are in the third inning of wealth creation. You probably want to own the trade from the third to the seventh inning. You don't want to stay to the end. I think in terms of the actual adoption, we are at the first pitch of AI. So, yes, we've had some headwinds here more recently, but I think ultimately this is the most important part. Everyone would agree that AI is going to be impactful. The question is, will the substance exceed the hype? And I think that will in fact happen. George, what's your opening statement? Where are we right here?
In agreement with much of what Gene said, I'm actually quite bullish on AI. AI, the technology. But as my former boss Peter Lynch used to say, the product's not the stock and the stock is not the product. You can have a wonderful product and have it be a lousy stock, vice versa. You know, I was discussing this today with my producer and I think this is a graphic example why I think it's a bad business. We're using Anthropic. I think we're spending 200 bucks a month on tokens or something like that. And if I'm not mistaken, we're getting in the order of something like $10,000 worth of stuff for it. So I don't know if I'm single-handedly bankrupting Anthropic or not. So the problem really is there, it's the product versus the business model, and in my view, in my opinion, it's a commodity. The Chinese are doing it for 95% less. I think the open models are going to dominate the industry, and just as an example, you know, the internet changed the world, it's certain the bulls were correct, but if you owned Global Crossing or Lucent or Nortel, you lost everything.
Gene, what do you make of that? And in particular, the criticism not of AI as a technology or that it's going to change the world, which you know very likely is, but that of criticism of the AI business model, the business model of serving tokens.
So at the core, the most basic, the fundamental principle around AI is taking energy and converting it into intelligence. And I understand that we're drawing an important distinction here between the use of AI and how impactful it will be, and separately the investment case. But when it comes to the investment case, my view is that taking that energy and moving it to intelligence, we have an opportunity to solve some very big problems. It could be for a more successful grid. It could be for curing disease. It could be for potentially solving for world peace. And at the most basic level, if you have a belief like I do that it will help solve some of the biggest challenges, how do you put an economic value on that? That's the question. George is accurate in terms of whatever he's doing with Anthropic. Anthropic is losing money on that trade right now. But ultimately, I believe that if in fact these models can do, and we can talk about the intelligence and where they're at and where they're going, but if they do help solve the world's biggest problems and a lot of other problems along the way, I think that the natural hand of the market will create value here.
Okay. I also want to introduce Gene. I thought you were going to say that Anthropic is profitable. So there's tons of people, and I was one of them, and George it sounds like is one of them, who are probably Anthropic has a negative gross margin on, but overall the gross margins are quite high and the operating margins are quite high as well in Q2 and Q3. I was giving George the benefit of the doubt here that there is a question about just in terms of how fast they're depreciating all that infrastructure and the impact on profitability. So you can I think look at this and objectively say that maybe it's not as profitable as they're saying it is, but again at the end of the day I think their cash position...
I would agree with that, Jack. It's much higher than what most people think. I think they've retained almost 70%, 80% of the cash that they've raised, they still have, which is remarkable given the kind of growth. And so, you know, I just want to maybe throw another kind of point. I don't want to mix concepts here too, but when we think about transformation, think about what has happened. Typically when you see big things happening, there are crazy things that happen in anticipation of that. And we can go and think about how fast we've gone for the last few years, which is awesome, and George would agree with all that. But ultimately the question is, is that momentum sustainable? And I think I've got a case that in fact it is. In fact, that despite everything we've seen today with these moratoriums on data centers, I still believe that the brain of AI over the next few years is going to be meaningfully two, three times bigger than most investors expect.
So, George, of course, you can respond to anything that was said before, but I also, you've called AI the greatest capital misallocation in history, and Gene says it's the third inning. So, clearly you have different expectations about what 2027, 2028, 2029, and 2030 are going to look like. So, George, could you kind of paint a picture? You know, obviously perhaps a bullish one about like adoption and stuff, but in terms of the economics, the pricing, how you kind of see this movie playing out? You know, have you seen this movie before? And then I want to get Gene's reaction to it.
Thanks. Very good. A lot to unpack there. So again, I think AI is a wonderful thing. I just don't think it's a good investment. I'd like to, by way of example, if you look to go back to 1999, the internet, and listen, this is not to say history rhymes, it doesn't repeat itself. There are ways in which it's very different, there are ways in which it's similar, and we can go through all of that. But if one was bullish on the internet in 1999, the bulls were actually right. The bulls were actually right. Peter Bernstein, a bank credit analyst, points out that I think internet usage compounded at 43% a year for the next 25 years, went up by like, I don't know, 500,000% or 500 million, some crazy, crazy number. So hold that thought. But at the same time, you know, many if not most of the companies that were involved in the internet in '99, Global Crossing, Lucent, Nortel, Cisco, AOL, go down the list. Many of them went bankrupt. Many of them saw their share prices fall 90%. And this is the way that investments usually work. Go back to the railroad boom. History is replete with examples. And so, you know, railroads wound up to be a very important technology. Energy wound up to be a very important technology. I'm sure AI will probably wind up being a very important technology, doesn't mean the stocks are attractive. And you know, if you put some numbers to it, and Gene knows the numbers better than I do, but you talk about capex running on, I don't know, a trillion dollars a year going to a trillion and a half next year, whatever. You sum it all up. How much money is being spent on AI over a five-year period? And I think people come up with numbers like approaching $10 trillion. Whatever the numbers are, the point of it is I'm interested in accuracy not precision here. You work backwards. How much gross profits have to be generated in order to justify this trillions of dollars of expense? And I can't get from here to there. And so there's a disconnect between the product, the technology, and the economics of it. And so, as an example, you know, if you look at individuals, it's either free for them or they're paying 20 bucks a month. You pay up, you can pay 200 bucks a month. I'm told we're paying 200 bucks a month. We're still getting a gazillion dollars worth more value out of it. And so, you know, the TAM, and Gene knows this better than I, he's been very correct on a lot of the, you know, Netflix and all the other guys. The TAM at 20 bucks a month is much different than the TAM at 200 bucks a month. And then they say, well, you know, how about all the corporate enterprise market? And we all read the same stuff. There's been information out recently, it came in the S-1, you know, 1% of the customers account for 80% of the business, whatever the numbers are. And then there's a question of, well, how many of those are actually making money? How many are loss-making? All this other stuff. Here's a very important point. I'm on a ramble here, but you really triggered me, Jack, with the question. If you look in the corporate arena, the enterprise level, almost universally, the big corporate clients, the supposed savior for AI, they're not taking this up. The adoption is not occurring the way the bulls think. In fact, you're seeing a lot of throttling back. But more importantly, what gets lost on a lot of folks, talk to companies themselves, they cannot or will not hand over their data to these large language models because of concerns over cybersecurity and compliance. And so you look at Boeing, Caterpillar, go down the list. So the way I look at it, yeah, there's a huge market for it if it's free, but then if it's 20 bucks, 200 bucks, I think it's much, much smaller.
George, let me ask you this question. You talked about I have a hard time getting from here to there in relation to the amount that's being spent on capex and then some sort of a multiple on what the earnings is going to be a few years from now. Basically, there's just not going to be enough revenue generated from that.
And what are you basing the 'I can't get from here to there' on? Is it that the $200 a month that you just mentioned? It's too expensive to really have wide-scale adoption? Like, help me understand what that gap is. Where that belief is that we're not going to get from here to there.
Yes. Thanks for the question. I'll just give you a framework on how to answer the question. It's obviously garbage in, garbage out. You know, you could put your own assumptions to come up with what you want. But if you start making calculations like how many customers does Netflix have, and you know, 20 bucks, and how big would the market be at 200 bucks, and start putting pencil to paper, and then plug some gross margin on that, and then look at the enterprise level and go through all that, you can't. And then you think about, in similar fashion, just to put this in perspective, the hyperscalers, you look out a few years, their depreciation's going to be running upwards of $450, $500 billion a year, which exceeds all their earnings. And so I just, no matter how I try, I cannot make the numbers add up. And this is all, the system is inherently unprofitable. This gets to a very important point I want to make. Jay Martin put out a really good piece a couple months ago on X drawing the analogy to the housing bubble. And the point is, you know, the system itself, the whole AI edifice, is very cash flow negative. It needs external financing. As we all know, most roads lead through either Anthropic or OpenAI. And if you let me ramble here for a second, I will land this plane. The key thing that Jay Martin likes to point to, the key moment in the 2008 housing debacle was not 2008. It was actually 2006, because that's when the rate of house price appreciation slowed. So all the Rube Goldberg perpetual money machine, where guys were buying things that they couldn't afford, but don't worry, you know, it'll go up 30% and I'll sell it, that started to freeze up. It didn't work. And the read across here to AI is that it's a very cash flow negative enterprise. The rate of spending is accelerating, not diminishing, and therefore it depends on ever larger amounts of external capital. And thus, if you look at the financings that have been done, you know, OpenAI and Anthropic keep doing deals up 50%, up 75%. Not because people are expecting a cash return, but it's because of prices ever going up. If the rate of price appreciation slows, as is now happening, you got a problem.
So, I still don't understand. I want to maybe I can answer the question that I asked you about how do we get from here to there? Because that's a fair question of like ultimately where's the revenue? Where's it going to come from?
And your argument is valid from what I would consider an academic basis, a historical basis. And I've recently received a t-shirt, I should have worn it tonight, that says 'it's going to be different this time.' Somebody was nice enough to send me that t-shirt. Of course, I'm careful when I say that this is different, but this is different. What's going on right now in AI, the significance of it is hard, I think, for humans to wrap our minds around. Now, this conversation, my comments may seem kind of fluffy, but the substance is what does it mean right now? The average intelligence, pop quiz. Jack, can I put you on the spot? Pop quiz. Average intelligence of a human is what IQ?
100. Thank you. That is impressive. I've asked a lot of people that, most people say 70. I don't know why. I digress. The average intelligence of a frontier model right now is 130. We're at general intelligence. There is 150 billion or 300 billion between OpenAI, from the model labs, OpenAI and Anthropic, 300 billion going to be spent in the next 12 months from enterprise customers. That's real spending, but that still doesn't answer the question: how do you get from 300 billion to trillions of spending? And the answer is that if we look at the adoption, you talk about like the Netflix and start going through anything that's knowledge work-related, and think about the disruption around knowledge work, and we can talk about pricing of tokens and premium tokens, but I think that essentially the addressable market, and this comes back to the SpaceX, Anthropic, the S-1s about these massive 20, 30, 40 trillion dollar TAMs, I think it's hard to argue that these aren't tens of trillions of dollars of TAM, any knowledge worker today. And then once we start to get to physical AI, and this is going to take longer, then we talk about the whole global economy. And I'd say one other piece, part of my enthusiasm about what these models can do. Not only do I kind of research and study the markets, but at Deepwater, we also have an affiliate company that's been around for three years called Intelligent Alpha. They have a ticker, it's an ETF called GPT. It's a global growth company. It's not comped against the S&P. GPT. Great ticker.
But we've learned a lot about models, a lot about infrastructure by being practitioners. And we are blown away at how powerful this is getting. And this sounds again like well-traveled, but let's look at the numbers. Gets back to the question about how do we get from here to there. If we look at the numbers, the big buzz right now is around the personalized agents, Muse, dots, Grokbot, all that's kind of storming, that was the big message around this last. If we look at the numbers, the usage, it is 20 million dailies, and there's five and a half billion dailies on the internet. Like, how we get there is two billion people spend a hundred bucks a month. Like, it sounds crazy. Like, how is that, how in what circumstance can it happen? Another way we can get there is right now the average spend, this is data from this company called Ramp, which basically is like a payment company, the average spend on AI per employee is 25 bucks a month right now. The average spend on software is 700. So this has so much room to grow. As again, the value cases, yes, there's going to be competition, yes, there's going to be some pricing that's going to go on. But ultimately, if you start to look at every industry that's knowledge today and physical in the future can be impacted by this, I think it's hard to build a case that we're at the top. One last piece, I should have mentioned this at the top. I don't think that the AI trade is going to go up into perpetuity, but I do think the next three to five years are going to surprise people, not just surprise, shock people at how resilient the AI trade is.
Gene, thank you for that. And I'm going to put a little pressure on you because I know you can handle it, because you know perhaps George's argument about yes, AI is going to be big. That doesn't mean that it's a buy now. It's, you know, you got to look at the stocks, not the product. You know, you totally could be right about robotics and all this stuff, but if it happens in 2032 and not 2027 and 2028, we might have a very serious problem. Tell us, you know, just to use George's language of getting from here to there, the spectacular growth of AI revenues over the past year, that has been very high, and I just want to acknowledge that fact. But in terms of AI revenues from the model companies, the clouds, where do we have to go to justify that ROI in 2027 and 2028? Because again, you know, in 2036 there could be robots that do incredible things. You know, robots could move mountains and whatever, powered by AI, but that doesn't matter if, you know, there isn't a very large growth in 2027 as there was in 2026.
So, I think that you mentioned the ROI question, just to be clear, talking about the ROI on like the NeoClouds or hyperscalers or just the broader AI ROI?
So the way I think about it is there is the part of the market is about earnings obviously, and part of it is about expectations. And when you invest, ultimately investing is a form of a religion. You have to believe something is going to happen that hasn't happened yet, and whether it's just continuation going on or whether it's a breakout or a breakdown, like there is some faith component that you have to have on this, and that's where the market multiple comes from, is a sense of what is the future opportunity of this company. I don't want to get too abstract here, but I used to cover as an analyst, I covered this little company called Sonic Solutions, SNIC. It was like comically small, like a hundred million dollar market cap. They missed, it was something like 32 consecutive quarters. The stock never went down. Not because it was trading below cash, but because for four, five years, five years, five, six years, they had this vision about high-definition DVD. As silly as that sounds, but the market always just kept saying that golden goose is out there and someday it's going to happen. When it comes to the ROI question, it's academic. The question is, we need to see movement towards that improvement, and we're seeing that with some of the Anthropic news. We're hearing it from SpaceX in terms of what they're seeing in terms of the return on capital. They say it's about a one-year return on capital, which is phenomenal, assuming that they're being honest, which I think they are. Ultimately, if the excitement is out there about what robotics potentially could do, all we need to see is small improvements. So small improvement example number one would be today, this year, about 1% of total ride sharing is going to be autonomous. And for anybody who's been in a Waymo or Cybercab robo taxi, it's awesome. It's so much better. And all that needs to do is go from 1% to 4%. And then people start filling in the blanks. All you need to do is see going from a very small percentage of surgeries that are done with physical AI robotics. And Jensen talked about this in Berlin last week to a little bit more, and the market I think will reward it. So my view is that the use cases of AI is going to go through the roof over the next two, three years, and that is going to basically fuel investors' optimism that this can go on for 10 years, and whether it goes on 10 years or not is beside the point. But if we are as close to a breakout, I think like FSD for example and Tesla, as silly as that sounds, like going from, you know, whatever it is, five billion miles a quarter to, you know, hundreds of billions. The hope of that I think is going to inject enthusiasm across the market that all that infrastructure needs to happen. So my view is it's less about, I think there is a path. We can talk about subscriptions. We can talk about enterprise spending. I think there's a path there. But what's more important is the optimism in the market. Right now it's dark. Understandable. I think we're going to see some blowout things coming from AI that it's going to be really exciting and really scary too.
Gene raises a lot of really good points for which I think I have a lot of really good answers. First I'd like to go back to read about Warren Buffett, and one of his first big investments was actually in the textile industry in New England back in the 50s and 60s. And I forget the name of the contraption, but there was this whiz-bang new technology, some type of fancy loom that was going to go a million times faster than all the other stuff. The idea was it was going to modernize the textile industry and efficiency would go through the roof and productivity improves and everyone's going to make a fortune, and so it would revive the American textile industry. Well, what happened was that got commoditized really fast, and yes it brought cost down immeasurably, but then pricing followed down, so it was a commodity. It quickly became a commodity, and I believe that AI is a commodity. And it remains to be seen what the pricing power is.
George, you've been really kind in giving me a lot of room here, but in the interest of really emphasizing the points here, I'd be curious in your, is it Jevons Paradox, the guy from the British Economist, William Stanley Jevons, like what's your take on that?
Yeah, demand increases enormously, but costs go down equally. That's the problem.
But Jevons Paradox is that the net...
Yes, costs went down for coal and they thought the coal producers were in big trouble, but the reduction in costs with the corresponding, of course, increase in demand was more than beneficial to the producers. And listen, Gene, you know, at the end of the day, it doesn't really matter what I think or what you think. What matters is what the market thinks. And let me get on to a very crucial point here. Whether you're right or I'm right, it's what the market thinks. Okay? And you know, when I was younger, I was more optimistic. Faith is a wonderful thing. I'm not sure when I invest I do things on faith so much anymore. But in all seriousness, keep in mind that yes, I agree when you see change at the margin which brings some of these things closer to reality, that provides grounds for hope. But also at the same time, keep in mind we're going through a huge upheaval in capital markets. We've come through, I started my career in 1981 working at Fidelity, and from 81 to 21 you basically had, you know, interest rates, inflation, a one-way trade all the way down, and everything went up. Stocks, bonds, real estate, commodities, everything went up. All right? And whether you're financing this stuff at, you know, bond yields are 60 basis points or 5% or whatever, and whatever PE ratios are, that's a huge difference. So, yes, there'll be more innovation. I have no argument, I'm sure you're right about that. But the market, there's this race between, you know, expectations and what the market demands. And as you know, it's funny. This is a great story, Gene. Two great stories. So, I remember, I'm old enough to remember when Disney made plans, they announced they're going to open up a theme park outside Paris, Euro Disney. So, everyone had these great things, you know, oh, they're going to have X number of customers and blah blah blah, and how many tourists go to Paris and blah blah blah. And you can make up all the blue sky calculations you wanted. It was a concept stock with the emphasis on the syllable con. They made one crucial error. They opened the park.
Disney. The hope was no longer. You had to have substance behind it.
And the same thing, by the way, same thing. Eurotunnel. I was running a hedge fund back in the day. Eurotunnel, same thing. You know how many cars are going to the Channel? Blah blah blah blah blah. You can make up anything you want, any scenarios you want. But then once the rubber hits the road, no pun intended, you got a problem. And so...
But by the way, with AI, we're going to have many years of people being able to dream up where this goes.
Well, yes, but the party can only continue as long as the market keeps serving up the capital because they're cash flow negative. And the market's telling you right now, witness, you know, Oracle, CDS, credit default swaps, etc., etc., canceled IPOs. The market's not willing to take so much on faith these days. And the crucial part is if people can take away one thing from what I'm saying tonight, because you make a lot of really good points, Gene, and that is the reflexive nature of all this. You know, it's not that, oh, you have the companies over here with revenues and earnings, then the stock prices over there and somehow they're not related. No, this is reflexivity on steroids. In other words, the share prices are the fundamentals because the share prices access to capital is, and if the companies are asking the market for more money and the market says no, or the market says we're going to give it to you only at a much higher price, that impacts the growth rate and in turn that impacts the valuations.
So that's why I come back. I totally agree with you. I mean, I'd really be keen to learn more about some of the work that you've done because I will confess I'm not an AI nerd. I'm saying that affectionately, okay? I haven't gone on the web to do that.
Affectionate term, geek. I prefer geek, but...
geek. AI geek. Okay. But, you know, as I, I'm not saying you're a liar. Please don't take it the wrong way. But you've heard the phrase before, you know, liars figure, figures don't lie. Look at the numbers.
And so, going back to your question you threw at me when I used the phrase can't get from here to there, I start there and I end there. You know, internet consumption has gone up by a million percent in the last 25 years. It didn't mean it meant that still almost...
What did we, here's a question.
Because we use historical examples too. You mentioned earlier about the 40%, that's a correct number too. I'll agree with that number about that compounding growth of the internet today. Now more than half of it is bot traffic, right? But let me ask the question: is like, what can we learn anything about what's happening with the internet traffic? I think we can, as potentially what could happen with inference.
Meaning what? Meaning what?
Basically, if we thought like internet traffic was, I thought in 2005 it was going to cap out, or 2002, and grow at 8% a year, and based on my numbers at 44%, but look at the last five years, it's more like low 20% compounding annually. And my point is like these transformations, all the underlying driver or the use case is just much bigger and much more sustainable when you have these paradigm shifts. This is a paradigm shift. This is not 3D TV. This is not Vision Pro from Apple. When you have a paradigm shift, the size and duration of it is much bigger than you think. And I think that's what we learn from internet traffic. The analogy, of course, is how much inference is going on today. And if we roll forward, let's talk. I think I fully interrupted your thought there, George. I want to be pointed but yet respectable. Do you want to finish your thought?
It's okay. There's so many things we can talk about. I don't even remember what I was saying.
Yeah. Yeah. Okay. I'll be the rude one now and I'll interrupt you, Gene, and I'll say, PE, I'll ask you the question, which is, you know, people don't like the number ARR because you take, you know, a favorable period of 28 days, multiply it by 12, I understand, but like let's just say a quarter, quarterly revenue times four, so annualized revenue of Q4 2027, Anthropic plus OpenAI plus Grok, at what level of monetization in terms of annualized revenue of those three, which your estimate of course now will be better than mine, but I'll say is, you know, probably maybe like 130 to 150 billion, at what number is it in 2027 that you can give, you know, your new friend George Noble a call and say, actually George, you know, look, you had a legendary bear call from, you know, June to November, we had quite a panic in the semiconductor, the AI trade, but actually I think the ROI has kind of so far yet been proven. What's kind of your threshold? And then I want to hear if George agrees with that.
So, the numbers that I look at from Anthropic and OpenAI is the 2027 run rate of 300 billion. So, we're looking at slightly different. Is that okay if we kind of start there?
Sure. Sure. Sure. Yeah. Yeah.
And so, the question is like, what's the number need to be to feel like we're moving in the right direction? And my sense is that they're going to grow in 2028 probably 100%. I think they'll probably do their guidance and the analysts will come out probably something like 70, 80% kind of growth, a much lower number, but I think it's going to be much higher than people believe. So the question that I have is 600 billion for two companies, that's two companies, that's pretty impressive, but when you look at all the trillions that are being spent on infrastructure, does that justify it? And I think the answer is yes. But what's more important, gets back to this internet usage kind of analogy that I was making relative to inference, is that if the view is at the end of 2027 that 2029 is going to grow at 80%, then I think this holds together. And ultimately I think that is going to happen because I think these tools, like just try them. I mean, I beg you, go Tesla, go and do, you can take a two-day test drive. See what FSD is like. If you haven't, I'm sure you've heard, you've been annoyed by people who have tried it and said it's amazing. Try it. You will be amazed. Try Muse. Try Grokbot. You'll be amazed. It's changed like our entire world at how we do things at Deepwater is totally changed based on using these bots, these personalized agents since mid-September. I mean, it's crazy. So, my point is if that group that's saying, 'Oh my, this is crazy like crazy good,' if that group is really small today, which it is, like fractions of a fraction of a percent, and it's going to be bigger in two years, I think these companies can grow much faster. So, back to that, those are two great ones to watch. What's going to happen with OpenAI and Anthropic, we're investors in OpenAI and feel very good about holding this in through the IPO.
And so I got Gene's vision of 300 billion in revenues, annualized revenues by the end of 2027, and then a double of that from 2028. So 600. Is that good enough for you or do you think even with that we're going to have some problems?
I think it's very difficult to have much faith in those numbers looking out. You know, in bull markets time horizons get extended.
But why, but George, why? Because history has said that, that's why it's hard to, I mean the law of large numbers say it's impossible. Is that, I just want to understand why you think that's...
No, but again, you know, we're talking about the top line, but also the bottom line. So for instance, now we'll go from the frying pan to the fire. It's supposed to be about AI. Well, SpaceX is an AI company really when you get down to it. Three quarters is AI. Okay. But I'm going to do the read across to the other Elon Musk company. I started my career as an automobile analyst in 1981. I was the guy carrying Peter Lynch's bags visiting Lee Iacocca.
I was in the room. Okay. So autos I know something about. That is... I will freely admit that I lost money in the past shorting Tesla, but I also happily report that this year has worked out pretty well so far. And the reason I'm mentioning this is, you know, one can make big predictions about revenues. It's hard to know what's going to happen six months from now, about, you know, three years from now. If you look at what, not only the Tesla revenues been way off, but even worse the margins associated with those revenues. It's been a complete and utter disaster. All right. And so, you know, history is replete with examples of where, you know, as Yogi Berra once famously said, you know, predictions are difficult, especially about the future. Okay, you want to try the Big Dig on the side in Boston, which was supposed to be three billion, it wound up coming in at 19 billion. I mean, I go on and on and on. And so I get, in a bull market where people are extending time horizons and cost of capital is diminutive, that might have, you know, dreams have more appeal. But in a world where we have a rising cost of capital, I think before we start debating whether or not '28 or '29 numbers, what that might look like, if the markets are not going to give you enough, aren't going to give you the capital just to get through to next year because they're starting to say, 'Whoa, wait a second.' And keep in mind, Gene, we know they always say the bond guys are smarter than the equity guys. All right, bond guys are like, they got, they're like, you got some explaining to do. All right, the equity guys are the dreamers. And so, you know, let's first get past this hurdle. Then we can worry about whether the '29 estimates are correct. You may well be right, but I'm just saying right now where I sit, I don't think you're right. And more importantly, it doesn't matter what I think. The market is not buying what you're selling.
So, George, I, you know, when Gene said, you know, he articulated a vision of robots and stuff, and I said, 'Okay, that could be 2032, but it doesn't matter.' I thought, George, on your behalf, I got Gene to make a commitment or a prediction about the next 24 months, which is the here and now, because of course, you're right that in bull markets times get extended. So, I think that is, you know, relatively near-term. And so, you're basically saying that you don't think that it's going to happen.
I don't, I listen, I will put it this way. I think there's an article of faith that one has to believe in to go with that. There are signs. I mean, I go back a few months ago. Kind of funny. I had Gary Marcus and Ed Zitron on a podcast, and it was the very week that the token thing was peaking, and Gary made the comment, 'Just watch.' And he was 100% right. And then I find it curious that, you know, somehow folks like Gavin Baker and Brad Gerstner, you know, forgot to mention what the third quarter numbers look like, or Anthropic forgot to mention what the third quarter revenues looked like when they came out with the S-1. Of course they knew. And so there's tangible signs that things are already slowing. And therefore, I may be wrong, but I'm just saying there's a lot of doubt here. And therefore to take Gene's forecast, which may be right in the end, but to just take them as an article of faith, that's a bridge too far for me.
Well, I want to say, George, on the Anthropic S-1, you're right, the Q3 revenue didn't come out, but I'm actually pretty sure before it came out, revenue was released, and then in the updated one, we have the revenue.
Yeah, but there was a lot they withheld. No, there was a lot they withheld, which would, and I won't say anything illegal was done. It's just how should I say, information was released. The truth was they were rather economical with the disclosures, and it was done in a way which kind of cast the company in the best light. But that's not really the point. The point is there are signs that things are slowing already.
But let me... signs of things slowing already. Yeah, Gene, get your thoughts on that, that the revenue slowed from a ridiculous quarter-over-quarter number, well in excess of 100%, to a number in the 40% quarter-over-quarter, which is still very high. But again, it's this deceleration.
We're still so early in this adoption curve that we can easily have a massive reacceleration. I would put it this way. Two years ago, companies were talking about doing AI. Over the past year, they've started to dabble. And the dabble, and I would say like, give me some hard use cases of like where AI is having a big impact. I don't think we could come out with a ton of them today. But yet you have OpenAI and Anthropic there. What's the run rate? Probably exiting this year, like you said, 150 billion, going from basically zero to 150 billion over three years. Something is happening. When you see that kind of growth and that low of adoption, that is a sign that something bigger is going on.
I don't know, my, so this idea, I think if, as I'm kind of, from a high level, it seems like we're kind of talking about two different things, like you're talking about cost of capital and history of railroads and a very academic approach to this, back to that theme. But I think if you look at what's actually going on with this technology and use it, I think you'll be blown away and start to believe things are going to change. We're going to get some high unemployment stuff. There's going to be some very hard things that are going to happen.
So, Gene, much of what you say may in fact come to pass. However, I want to go back, it's not so academic. I want to go back on Tesla as an example. Okay? You look at the forecasts that were made...
But Tesla stock has gone up.
No. Gene, Gene, this is where I beg to differ with you, please.
Up until 2021, Tesla stock was a monster.
I never owned it. I was short from time to time. I wish I'd owned it. Shame on me. We all make mistakes.
By the way, I want to put it from 70 billion to a trillion.
Right. Right. So, Gene, please let me make my point. Tesla stock over the last, but the bloom is off the rose right now. Tesla's stock is flat over the last five years against a market which is up 50 or 60% if not more. All right. More importantly, there's a stock price and the fundamentals. Those two aren't always the same. Tesla is going to sell about 1.7 million vehicles this year. Okay. The estimate five years ago for Tesla was 6 million cars to be sold this year. Okay. Tesla, I'm out of date now, but already last year they were supposed to make, I think over $8 a share. You look at the estimate going three years prior, it was, sorry, they made about $160 a share from memory. The estimate three years prior was for $8 a share. So I'm not saying that AI is necessarily Tesla. I'm just trying to illustrate how again, predictions are difficult, especially about the future.
But if we agree with those numbers, I agree with those numbers. Six million cars supposed to be 2027, they'll do two million, something like that.
But the stock is at the same place. That to me says the market is looking beyond all the downward revisions and saying what else is to come, back to this idea about behind door number two, and behind door number two is obviously FSD, robo taxi.
Right, and I would say a couple things. So, A, it's always what's the next thing, but you know, at a certain point, and I would beg to...
I strenuously differ with your characterization of what's going on with the Cybercabs and FSD and that whole thing. The fact is they don't have Level 4 performance. FSD, they're selling a bunch, but the regulators aren't going to allow the robo-taxis because they've not been able to prove Level 4.
By the way, we could—Jack, can we take a bet on that? Not a bet, but just make a prediction on that.
Sure. But then I want to move away from Tesla because I feel like you both are so strongly held in your views.
No, I'll be brief. I'm going to land a tough blow, but a brief one. Elon Musk is the only person—only serious person in the world—I mean, I take that back. He even gave up a few months ago saying they weren't going to—that's not the point. Tesla, you know, if I come to you, Gene, if I come to you for 12 consecutive years and I say we're going to have self-driving cars, Level 4, Level 4, Level 4, at about the third or fourth year you might say George is full of it. Some might say George is committing securities fraud, all right? It's just not credible. I mean, they don't have Level 4. It's going to take Level 4 to be able to get to where they need to, to get all the Cybercabs. It's just not happening. And Tesla's admitted with their own technology they can't get there. And they admitted in court that they don't have Level 4, but then they say to the investing community something totally different. But let's get back to AI.
So I guess what the bigger point really here is, Gene, you could be totally right. I'm just saying, you know, before we get to the promised land that you're suggesting, you know, the—and you know this story, you get involved in a stock, Peter Lynch always used to say, can you upgrade the story, downgrade the story? What do you learn incrementally? You know, is the TAM getting bigger, is it smaller, this, that, and whatever. And I guess what I'm saying is, you know, right here, right now, clearly is evidenced by it's—and it's not just the stock price is going down, there's tangible signs that things are slowing. But then on top of that, the reflexive nature of this—we're clearly in a way, in a very fundamental—the stock prices are the fundamentals because of the reflexive nature of it, that these companies have to earn the credibility of the market, the trust of the market. Gene, I'd be happy to come back six months from now with Jack and say, you know what, you were totally right, I was totally wrong, okay? But this is not—it's a show-me situation, it's not a trust-me situation.
And you know, in a bull market people extend horizons, they're willing to believe anything. In a bear market, it's like, well, wait a second, hold on. You said what? And that's kind of where I think we are. And so the real problem right now is it'd be one thing if they weren't so dependent on external capital, but they are. And you know, when OpenAI last week, they go to raise some money and they raise much less at a lower price than expected. Then you see IPOs canceled all over the place, and who knows what's going to happen with Anthropic. I mean, their cost of capital, their stock price—that's the cost of capital. Forget about the bond yields. It's going up. Going way up. And the question—the easiest way to make this problem go away: show me the money. I hope, listen, for humanity's sake, I hope you're right. I just—I need to be convinced. That's all.
Yeah. So, Gene, George's argument is in the long term you could totally be right, but financial conditions are tightening sufficiently now that it may not happen. What do you make of that? And do you think that, like, if, you know, we have a 2008 or 2020 moment where just the credit markets freeze and shut down and Microsoft has to borrow at 9%, like, then what happens to the 600? And that is by definition kind of a very negative case, but I'm throwing it at you and seeing how you react.
My answer is today is different than yesterday. I think that that is a real conversation about the near-term, and I feel that. But when you talk to—we do venture and public investing—when people like me say something is 10 years away, they don't know when it's going to be. Five years, that means they think it's going to happen, but they're not sure when. If they say it's a year or two away, there's pretty high confidence in it. I think in the next one or two years we're going to have breakout. The next six months—this isn't like me trying to get out of a corner in case the market flips upside down. But the reality is what happened with the blue wave is negative for the AI trade, and we saw that getting priced in today. The question is, is it more negative? And ultimately the question is—and essentially to play this through to the cost of capital—is that if it gets difficult to get these data centers up and running, if it's that—moving them from Texas to Ohio or to some other place can impact the cost of capital, that can get investors nervous. That's a problem. But at the end of the day, the—basically intelligence at scale—well, I think it's an eventuality that it will impact almost everything. And I think that means that eventually these data centers are going to get built. Or based on Elon's tweet today about—I can't believe SpaceX wasn't up on this data center news. I understand that it's a negative trade against Elon because of the change in the parties, but this is like—they're going to get the data center thing fixed, whether it's in Texas or whether it's some other state. It will get fixed. And so I do believe this is a riskier part. We're entering a riskier part, especially going into—thank goodness that the earnings just happened, maybe. But when that December earnings call is at the end of January, that's going to be the first time when people are really going to sharpen up in terms of what those moratoriums mean to the pace of the buildout, because that's important to keep this whole thing going.
If I'm not mistaken, Gene or Jack, did I see something? 76—I think they did some polls. 76% of those who voted were opposed to data centers.
