Do I think we're in a bubble? 100%. Do I think we're going to get a 30 to 50% drawdown at some point next year? Yes. But do I think you can make a lot of money between now and then? I think the answer to that is also yes. Corporations, if they don't find somewhere else to cut their budget, their stocks are going to get absolutely killed because right now the market's desperately trying to figure out, are you helped by AI or are you roadkill? And the fastest way to show you're roadkill is to say, 'I'm using AI, but my numbers are going down.'
All right, welcome to the Risk Reversal Podcast. I am Dan Nathan. That is Dan Niles. He is the founder and portfolio manager at Niles Investment Management. Dan, welcome back to the podcast.
Always a pleasure to be on, Dan.
Well, I really appreciate it. I think our listeners, our viewers know you are easily one of the most transparent people when it comes to your views about markets. You put out an extensive tweet thread on Sunday nights at Daniel T. Niles and you update some of those views. I know you go on CNBC a bunch and we always appreciate you coming here. Not just because of the analysis that you are doing on a real-time basis. I know we just got through a plethora of tech earnings over the last few weeks and your real-time speaking about it on CNBC, which I always find absolutely amazing. But also your sense of history. I mean, you and I met in the late 90s. You were at Robertson Stevens, which was at the time, I think, one of the most epic boutique investment banks covering technology. And you know, to think about where we are in the IPO cycle and this and that, whatever, you know, it was always Goldman and Morgan lead left, but you guys were doing a lot of deals and your analysis and companies wanting you to cover them was a big part of that. And obviously you had this amazing career at Lehman covering hardware and semis and you made some epic, epic calls there. And then you've been an investor for over 20 years. Neuberger, you had your own fund and then you launched Niles Investment Management. All of that together I just find, you know, fascinating. You were an engineer too before that. So a lot going on there. What I thought we'd do, and I always love getting your macro take starting there, and then let's review Q1 and Q2 guidance. And of course, you've been kind of vocal on what's going on in the markets. A lot of people when they say bubble, they think it's about to burst. That's not your view whatsoever. You've been talking about that for a while and I think your history being around these tech cycles is very informative to that whole notion of bubbles and everything like that. All right, I'm going to shut up. Macro, you always take a top-down view, I think, to help frame a lot of your different themes and then you do a bottoms-up sort of thing with individual names and you meet in the middle and you develop your investment strategy. Can we talk about the macro first? Where you are, how you think things are evolving? It's been a volatile, let's call it year and a half on the geopolitical and macro front, but that's also presented opportunities to you.
Yeah. So, if you kind of look at a post I put out on March 31st, it was entitled, 'History may not repeat itself, but it often does rhyme,' which is a famous quote attributed to Mark Twain. And what I was trying to say there is I was comparing the period of time we were in to what happened in 1997 and 1998. Remember Netscape Navigator came out at the end of '94, that kind of kicked off the internet buildout, much like ChatGPT did at the end of 2022. But in 1997, you had a currency scare out of Asia and the S&P went down about 11% within that year, peak to trough. And then in 1998, you had the Russian bond default and the failure of Long-Term Capital Management and the S&P went down 19% intra-year. It finished the year up though 31% in '97 because underneath all of that was this big buildout of the internet infrastructure and that was in year three and four. So the comparison I was trying to make is, okay, if you look at where we are on March 31st, I said it's kind of similar, right? S&P's drawn down about 9% peak to trough, but this is another macro scare caused by oil prices surging on the back of Iran. The good news is Iran is easy to walk back from to some degree relative to the things that led up to a currency crisis or bond default, which take multiple years and are more structural in nature. So my feeling was this would be a good buying opportunity because much like you had the internet infrastructure buildout underneath '97-'98, you had the AI buildout underneath what was going on this year. And so if you go back and look, that was March 30th was actually the bottom. So that post was one day removed from the bottom. Since then, the S&P through yesterday was up 19%. The Magnificent Seven through yesterday was up 26% and the semiconductor index was up a whopping 80% just from March 30th. So you've had this big surge. Now the question is, what? And to your point earlier, you can be in a bubble, which we are, and have that go on for a very long period of time. And bubbles only form when you have generational investment opportunities. So think of building out the canals in the late 1700s, the railroads in the early 1800s, or most recently the internet infrastructure buildout in the 1990s, and those can go on for a long period of time. But if you call the bubble too early, you're just dead wrong, right? You're going to give up a lot of money or get carried out on a stretcher trying to short these things if you're too early. And the one thing you have going on right now that you didn't have during the internet infrastructure buildout is in January, on January 30th, you had the formalization of this thing called OpenClaw, which really kicked off the whole agentic move by corporations. So Dan Nathan, instead of, you know, prior to that saying, 'I'm going to ask ChatGPT a question, it's going to give me an answer,' might go to ChatGPT and instead of asking, 'Well, you know, how did internet stocks perform in the late 90s,' you might go and ask ChatGPT, 'Hey, go to Bloomberg, pull down data for all of these 20 internet companies, then go to the SEC website and go look at the 10-Ks and see if there were any write-downs, and then go construct, see what the news stories were at the time at the Wall Street Journal, and then create a spreadsheet where I'm putting all of this data.' When you do that, that takes 10 to 100 times more tokens. And you can see that in the tokens being generated. And so you have this step function change in compute demand caused by that that started on January 30th. And so I think you're going to see very strong demand for at least another year until you sort of anniversary that. And I think stocks continue to go higher because of that. Do I think we're in a bubble? 100%. Do I think we're going to get a 30 to 50% drawdown at some point next year? Yes. But do I think you can make a lot of money between now and then? I think the answer to that is also yes.
All right. I have so much to ask you about. I just want to finish up on the macro because you use the example in '97 and '98, Long-Term and, you know, the Asian debt crisis and all that sort of stuff, and there was a lot going on. And I think back to where I was on a trading desk and I was like a year or two into the business. It's amazing how little I knew about anything, by the way, but macro was a big, you know, it just wasn't something I considered. I didn't think about, you know, interest rates or, you know, so just give me your 411. We just had two huge sort of events over the last year and a half. The tariff rollout, a trade war, you know, reshoring. I mean, there was a lot of things, right, that, you know, it was definitely worth considering. The market freaked out about it, right, in April of 2025, but before you know it, we were back to, you know, off to the races, that sort of thing. Here we are now with this situation with Iran. And, you know, I think a lot of folks did not expect the sort of disruption that we've had with supply chains, specifically with oil and a whole host of other things, but, you know, the speed in which we came back, the markets or investors just immediately discounting basically both of those situations and we're much higher, right? So how are you thinking about the move higher in yields and what it means because it seems like, you know, we're going to be hanging around above 4% in the 10-year. You know, how are you thinking about a new Fed chair? How are you thinking about just, you know, some of the relationships on the trade front because Trump went over to China. Nothing was really accomplished. I mean, we come out of that sort of thing. So I'm just... And there's probably a couple other things to think about. A US consumer that seems to be sort of weak or being propped up by, you know, a higher-end consumer. How are you thinking about all of that excluding, let's say, the geopolitical stuff because it doesn't seem like that bothers you too much. If anything, it creates opportunities for you.
Well, I think you have to remember '97-'98 or look at 1999, the Fed was actually raising rates. And '97-'98, if I remember correctly, the 30-year, you know, rates were actually higher than they are today, but valuations were a lot higher. So I look at that together, I go, well, where are rates, but then where are valuations? And so the example I would give is Cisco was kind of the Nvidia of its day, right? It became the most valuable company in the world. But at its peak in 2000, the company was growing revenues at close to 60%. But the PE, the forward PE was 140 times. So 60% growth, 140 times PE. Nvidia this year is expected to grow revenues at 80%. And the PE is 25. It's hard to call that a bubble valuation. Now there you can definitely make an argument. Well, if Nvidia is trading at 25 times, should these other companies be trading at 30 to 40 times that aren't growing nearly as fast and something's got to give, right? But it's very hard to say, hey, this is a major problem because we've seen this before and if you have a massive demand driver underneath it, EPS this year is expected to grow 25%. '97-'98, EPS wasn't even growing double digits. So you have an earnings underpinning that's driving all of this. And so that's why to me, I know where, you know, inflation is seeping in obviously, but if oil, at the end of the day, I know one thing for certain, we have midterm elections in November. If oil is still up here, the incumbent party will get absolutely slaughtered in the elections. And so the current administration wants to have oil prices down. And that's one of the reasons I've looked at this and said, 'This is going to come to an end.' And so I'm looking at that and saying, 'All right, this is going to keep the market wanting to go higher.' Because the other part of this is, yes, inflation's going up, but the incoming Fed chairman, Kevin Walsh, he was picked because he wants to cut rates. Now, how can you justify that? Well, he's focused on trimmed mean PCE, which is at 2.4% versus core PCE, which is at 3.2, and headline, which is at 3.5. So he's looking at that going, hey, inflation's actually coming down. And then he's making the case that AI is deflationary. Now you can make up whatever you want to justify whatever policies you want, right? Let's not forget in 2021 the Fed said inflation was transitory and it turned out to be the worst inflation in 40 years. Our job though is to go, all right, what is somebody going to do? And then how's the market going to react to that? And so I look at Kevin Walsh's stance and saying he's going to push back hard against what the bond market is saying, which is the Fed's going to hike at some point by the end of this year. And so that's why I look at the two factors that have really driven the market since the end of 2022, which is AI, right? Capex is up 70% year-over-year this year driven by agentic. And then I look at easy money and I go, 'Well, Kevin Walsh wants to continue those policies.' And then I look at the bond market and I say, 'Yeah, Treasury yields, 30 years, are up at where they were prior to the global financial crisis, but still relative to the late 90s, they're not high.' I look at that together and I go, 'Yeah, I think the bubble keeps inflating for the rest of this year.'
Let me just hit that Nvidia for a sec because again, you know, 25 times growing revenue at 80%. It's just astounding, especially if you think about this three years on from that huge quarter where they, you know, kind of, you know, like beat massively and guided up massively and the stock just kind of took off, right? And then, you know, like six months earlier was a $200 billion market cap. And now it's, you know, five trillion. Okay. So like, you know, I don't really care what the valuation is because at some point it's not going to take much in my opinion when you see, you know, margins start going back towards 70% and probably on its way below that. It's accrued just, you know, just so much market cap that, you know, to me I just think it gets harder to move up, you know, dramatically from here and you could say, well, it just moved up 25% off those recent lows, you know, or something like that. But I want to look at like Intel and this is a company that you just had such a beat on back in the day. Intel, Intel, this is in the late '90s. And I think, you know, when you started turning, you know, on the bubble, meaning, you know what I mean, the infrastructure build, that sort of thing, it was really centered around those sort of two names, right? If I have that correctly, if I think back in my history. Right now though, there's no way to think about an Intel as anything other than an egregious bubble. I go back to peak earnings and sales in 2021, $5.50 in earnings, okay? You know, like I think it was $80 billion in sales, 57% gross margins, and here we are now, and that's when the stock was trading at 70 bucks. Now, here we are at $123, right? And the company's expected, let's just be so generous, right? Now, you know, $1.10 is consensus. Let's call it $1.75. I don't care, right? And they're expected to do maybe $60 billion in sales, call it 65 gross margins, maybe 40%. How do you justify that stock given where it was, let's call it five years ago, where it is right now, and what expectations are.
Well, this is all you need to know. And this is what I wrote in late March when I said, 'Look, I think Intel is really going to benefit from agentic AI,' which is in AI, you move from eight GPUs to one CPU. Some people think it gets flipped. I don't know that it gets to that. But if you go from 8:1 GPU to CPU to one, that's all you need to know because your numbers with semiconductors, I never have price targets. I never have estimates because on the way up, your estimates are always way too low and on the way down, they're always way too high. And valuations just don't matter until you get to that turning point. And so with Intel, you could look at it a different way and say, 'Okay, they're under earning, right? They guided double consensus for this next quarter that they guided to for June. So, you know, the numbers are wrong. You're just trying to figure out how wrong could they be. And if you're going from an 8:1 ratio to one, they're really wrong.' And so the other way though, if you look at valuations, you could say, well, they're under earning on a margin basis. So relative to AMD, which is their closest competitor, or TSMC on the foundry side, on an enterprise value to sales basis, they're 30 to 40% lower than those guys. Let's see what happens. Don't forget for me, I run hedge positions. So I can be long Intel and short something else that I go, I think this is massively overvalued. And all I am trying to figure out is on a relative basis what is more important with this new thing called agentic AI that showed up in January and in three months you have not figured out the full impact of that. There's no chance you have.
All right, but let me ask you this because, and whether you believe it or not, I mean, I think you have to take, you know, the word of the COO of Uber and what's going on at Microsoft. So, agentic AI, right? And a lot of these engineers working at some of these massive tech companies, it seems like that's where it's happening first using these tools, right? Creating agents, like Uber just said they just ripped through their entire 2026 budget in Q1, you know, for like token pricing has gone through the roof, right? And that's great for a whole host of players within the ecosystem except for the companies that are spending on it and trying to demonstrate the return on the investment. And, you know, Microsoft, which is obviously a huge investor in OpenAI, that's been a problem for them. They have not figured out how to sell, you know, seats on Copilot using OpenAI technology. They invested in Anthropic, right? They've integrated Claude into that, giving users choice, but they're still not growing at a pace that, you know, investors are willing to kind of underwrite right now. So, do you think this is just a blip? We're just hearing a few one-offs or is this going to start being a trend if some of these companies are not able to defend the cost on compute in the near term?
Oh, you're going to have a ton of companies that can't defend it. And that's what I wrote up, I actually wrote it up this last Sunday. But the question is who gets hurt? So one way to think about it is on an annualized revenue run rate basis at the beginning of 2025, OpenAI and Anthropic were running at $7 billion in revenues. At the beginning of this year they were at $29 billion and the most recent data we have is that they're at a run rate of $69 billion. So you went from $7 billion 15, what is it, 15, 16 months ago to $69 billion and to your point, Uber has got to figure out how to justify that. Now what are they going to do? One of the things they will do is figure out where can I cut because if I'm spending that much amount, well, what don't I need? So I think software, you're still in the early stages of seeing the disasters that are going to happen there. And, you know, these are round numbers, we think of software as a trillion dollars in global spend a year, but IT services is like a couple of trillion dollars in global spend a year. Do you need all those high-priced consultants if you've got Anthropic or ChatGPT able to do it? Right? The other part is also, well, do you need as many engineers? Because if you have a great engineer that's then using AI to spin up 10 agents underneath them, then maybe you can get rid of pick your number. And if you think of knowledge workers, you're talking about, yeah, the numbers vary, but let's say $30 trillion to $50 trillion spent on knowledge workers per year. So that's where you're going to have a humongous value destruction because to your point, when Microsoft invests $5 billion in Anthropic in November, gives their engineers access to Claude and then all of a sudden they're pretty much phasing that out by June because the bills are too high and they're forcing them to use Copilot, then yes, you're going to have some issues. And this whole token maxing thing was smart in the sense that, hey, let's make sure all of our engineers get to using this. But then the problem, which is obvious, is that, well, you tell somebody this is the metric I'm focused on and then they'll just run really inefficient code that does stuff just to produce a lot of tokens and so the boss doesn't think that you're slacking off versus writing efficient code. Right. It's the, what is the phrase, like as soon as you make a metric the touchstone, the metric stops being useful. That's kind of what happened with token generation, right? So you're going to have to find a middle ground on this, but you're going to see all of these AIs get a lot more useful. And to your point, corporations, if they don't find somewhere else to cut their budget, their stocks are going to get absolutely killed because right now the market's desperately trying to figure out, are you helped by AI or are you roadkill? And the fastest way to show you're roadkill is to say, I'm using AI, but my numbers are going down.
Dan, how are you, just as a little bit of an aside, how are you personally using this technology, right? Like so I, you know, I know you're an early adopter. I know you have a technical background. I know you've covered, you know, these companies that, you know, have come back from the dead for all intents and purposes, you know, like if you think about the CPU makers which you just, you know, just discussed, you know, AMD and Intel, they've had a second life here. We can talk about storage and memory in a second, but how are you using this technology because you cover a lot of ground, you know, as it relates to your space. I know you broadened out, you cover a lot of other sectors. We talked about the macro, that sort of thing. Are you getting a lot of benefit and what are you creating agents and using OpenClaw, that sort of thing?
Well, forget about me for a second. I'm fortunate enough to have my son Warren, who's 25, has been, you know, messing around with what was AI before it was AI. And so he's in the process of getting his startup launched. And so I can see, given he was living with me not too long ago, how his use has evolved of this thing. And most recently, like I went in, I looked in, you know, he was showing me, hey, you know, I've got this four-page prompt of stuff that I need this AI to do. And I'm seeing all of these agents that are writing different pieces of code. And as he joked to me, he goes, 'Well, my job's secure for at least another six more months because it's made a bunch of these errors and I got to go in and clean it.' But he goes, 'It keeps getting better and better and better every day.' And I'm sitting there, I'm watching this thing as it's, you know, and I'm seeing he's got like these triple monitors as it's spitting out all this code. And, you know, because he's looking at it saying, 'All right, what's going to happen to a software engineer going forward, right? But what's going to happen is that software engineer is going to be armed with Claude and he's going to be 10 times more productive. Now, the bad news is because these agents are running 24/7,' as he said, he's trying to sort out not getting up at 3:00 in the morning to check and see what, you know, it's written and see if there's issues because now all of a sudden it's working 24/7, right? And so when you see what's going on at the cutting edge, you know that's ultimately going to come back to you. And obviously my workflow has changed as well. But it's just been fascinating watching him because, you know, he started off using ChatGPT and then he went to, you know, Grok for a period, then he went to Gemini as the models keep changing. And as he said, he goes, like the advances are so fast. And he's gone and given lectures on how, to some tech CEOs, on how some of these new advancements in the algorithms, how that's going to change what the hardware demand is underneath it. And so it's really cool seeing that at its highest levels to some degree. And then of course I'm using it in my job, which is great because now it used to be I would, if I had a question, because I invest across all markets, right? So healthcare, biotech, utilities, whatever. And I used to pick up the phone and if there's a sector that I don't know quite as well, but I have an idea I'm working on, I'll pick up the phone, call an analyst, etc. And now what I'll do is I'll give, you know, a one-page prompt of this is what I want you to do. Go write me a report. And then if I don't understand it, then I'll say, 'Okay, explain it to me like I'm a, you know, a 20-year-old. Okay, now explain it to me like I'm a 10-year-old, a five-year-old.' You know, until I get to the point where I go, I understand it. I don't have to wait for some Wall Street analyst to call me back. I've got an analyst now that knows all the data through the history of the internet and doesn't forget a darn thing. And that can call me back at 2 a.m. if I wake up and I go, 'Oh my god, I just thought of something.' And then of course I've got stuff underneath that that says, okay, if the semiconductor index is up 20% and inflation is at this level and it's moved up this much, but M2 money supply is at this, but the Fed is potentially forced to raise, like I can feed it all these different scenarios. And so if I have a question...
All of a sudden it can go in and do all this analysis for me and get it in whatever format I want. And so the amount of productivity that I've gained just at my, you know, plebeian level compared to what my son's doing is huge. And you're going to see that across all of society. And the people that are going to get killed are the ones that just don't embrace this to supercharge their own productivity because that's really what it's doing. It's incredible.
No, that that that's awesome. Um, and a lot of folks who are listening to this, I mean, you know, you can actually go to the models or the LLMs and it can help you figure out what the prompt should be, right? Like it it really is like absolutely amazing. Let let me ask you this really quickly and we'll get back to um, you know, some of your thoughts on uh, what's going on and and I got to hear what you have to say about Micron and some of these other names because I know that Micron was an epic name for you back in the day. Um, what are you using? What what sort of publications do you read um just to kind of that you think are just really good um, you know, chronicling just what's going on in AI and just the advancements and you know all the other stuff that's going on? The friendemy stuff, I find The Information amazing on this and I'm just curious, FT is really great on their AI coverage and which is odd if you think about it, but what are what are some of the things that you're reading on a day-to-day basis?
No, I mean I follow a lot of different people quite honestly. Um, you know, there's a ton of podcasts that I that I'll look at, but really it's sort of one-offs in terms of what I'm looking at. Quite honestly, I like talking to my son because, you know, other people read, you know, watch TV before going to bed at night. He likes to read the latest AI papers when they come out. And so I'm always like, okay, like he'll call me up and say, hey, Dad, you know,
He's like an agent for you, too. I mean, you getting... Yeah.
I mean, it's great because he finds that stuff fascinating. You know, some people might like to, some 25-year-olds might like to watch Euphoria before going to bed. He likes reading the latest research paper coming out of whether it's DeepSeek or Stanford or whatever. And then you know I'll watch a bunch of podcasts as well that um I like. I have follows on people like Andrej Karpathy and uh you know you can there's a lot of different there's so much information that quite honestly the hard part is not being up 24/7 trying to to read all of this stuff and and trying to understand it. Um.
What what do you make of what do you make of some of the doomers? Some of the naysayers like this guy Gary Marcus gets gets tossed around. You probably see him on Twitter and you know there's a lot of Substacks out there that are just calling kind of... They're looking at a lot of the circular, you know, financing that sort of thing. I mean do you read that stuff just to kind of know what what some of the, you know, the other kind of views are because there's obviously a ton of bullishness. You don't have to go far to see that.
I was talking about circular financing in the middle of last year. So like yeah obviously do I know what's going on? Absolutely. But in '97-'98 there was circular financing. Back then they called it vendor financing going on then too. This is not new news like this happens every...
But it took down companies like Sun. I mean there there was plenty of...
It. But you made a ton of money. I mean think about it this way, right? Alan Greenspan gave his famous speech on irrational exuberance in December of 1996. Was he right? Absolutely. It was just three and a half years too early because the S&P went up 90% from when he gave that speech to its peak. Um, NASDAQ was up like almost 300% and I think the semi index was up like 450%. I'm doing this out of memory, but so like did you want to give up that kind of gain? I don't. Now the good news is you're going to see signs that this is coming to an end. But you're not that's not now.
So because agentic is real, it's a massive pro... it's a step function in what we had from what we had prior to that. And so that's going to keep this going at least for another year. And you've seen it, right? Where these stocks, these companies are making way more money than you ever imagined possible. I mean, think about Nvidia, right? People have been talking about, oh, it's a bubble, it's a short, etc. Nvidia's revenues in July of last year were growing 56% year-over-year. They've accelerated for three quarters now. They're growing at 85 in the quarter they just reported, and they guided to 95. It's trading at 25 times. Apple's in the 30s. It's not growing at 80%. So my thing is that yes, you can have a bubble and still be a good time to invest because you make a ton of money in bubbles, but do I know that this is going to break just like canals, just like railroads, just like radio, just like, you know, electricity, just like fracking. We had a bubble there, too. Just like the internet. Absolutely. But I don't think it's ready to break yet because the productivity boost you're getting is still in the early stages from agentic. But is everything you talked about going on? 100%. But I just don't think it's timing. That's what it comes down to.
Two things here. You remember the quote um Scott McNealy, then CEO of Sun Microsystems. The stock, you know, topped out at 65 bucks in early 2000 and it bottomed out at five and I think that, you know, Oracle bought it for what, seven, you know, whatever it was, right? And McNealy, this famous quote was, 'At 10 times revenue to give you a 10-year payback, I have to pay you 100% of revenues for 10 straight years in dividends.' And this assumes with zero R&D for the next 10 years I can maintain the current revenue run rate. Do you realize how ridiculous this is based on those assumptions? And he was talking on Bloomberg at the time. You know, what what do you make of that now? Because that looks almost cute comparison to what we're seeing now on on some of these multiples and again valuation not a timing tool, history rhyming if not repeating, that's what you know we started this conversation off. So I'm just curious when you see that going back let's say 25 years, what does it mean to you and is any of that applicable to the way you know things are moving right now in the space?
Well, yeah, obviously. And I think the and I was talking about this since the middle of last year saying, 'Look, not everybody's going to win.' And OpenAI, I've said this multiple times, I think they're going to be at the epicenter of problems because right now it assumes that everybody wins. And I think you've got Google that's going to win in consumer. Why? Because they've got 13 different products with over a billion users each. They can train their AI with plus the cash flow to fund it. So they win in consumer and in corporate, Anthropic, that's what they focused on from the very beginning where OpenAI was trying to figure out how to get Dan Niles and Dan Nathan to pay for a subscription when we got trained on, hey, we get answers for free from Google, so why would I want to pay for a subscription? Which is why only 5% of their subscribers actually pay. So they've got an issue trying to then switch to now look a lot more like Anthropic and that's why they're focusing more on coding and then they've cut back on their ancillary projects and they're doing all of this stuff because they're in a huge catch-up mode going, oh yeah, corporations will pay for stuff. Consumers have been trained by Google and Meta and every other internet company that you don't need to pay. And so that's where the problems are going to start. And I think you're going to start to see that as you go through the rest of this year where much to your point, right? Microsoft has said, 'Hey, I'm going to force everybody to use our internal Copilot.' Um, obviously that's a detriment to Anthropic, but you're going to see more of that stuff like the Uber people saying, 'Hey, we blew through our entire token budget in four months.' Like that's not sustainable, right? You you have to figure this out. So there's going to be some come-to-Jesus moments in there. You've already started to see that shakeout this year, right? There's a reason why Microsoft stock is still down like 15% or so, I think, year to date. And then but you've got a Google stock which is up 25, right? So, and that's just going to get worse as you go forward and ultimately when this does break then it's going to get really ugly.
Help me with this um because I think this is probably going to get the ugliest. Um so let let's talk about Micron. Um so you know about a year ago um I don't know how it wasn't hard for you know investors to kind of extrapolate a bit because you know high bandwidth memory was something that was being used right in this entire sort of buildout right alongside Nvidia GPUs that sort of thing and you know this is a company that was not putting capacity online, they're scared shitless of doing that for every cycle, you know, what I to kind of worry about overbuilding that sort of thing. There's been fits and starts, you know, this is a cyclical business that sort of thing. So now if I look at Micron and I say well if everybody missed it including the company right, the stock was trading at $64 about a year, you know, I don't know, 13, 14 months ago. Now it's trading above $900. It's got a trillion dollar market cap. In 2023 they lost $4.50 in in calendar year. This year they're supposed to make $86, okay? Like literally think about that in earnings. Revenues have gone from 17 billion, okay, expected to be 150 billion this calendar year. Negative margins go to, let's say, 70s, something that looks like this is a company that's never ever had margin anything near that. So my question to you is, up nearly 900%. Okay, the stock, what is it discounting here? Right? Because you know we already have what the consensus is right and at some point and you could maybe make the argument well you know you've only seen Nvidia accelerating when a time where people are like they can't keep mid-70s margins, they can't keep growing this way off these massive numbers that sort of thing. How do you compare the two? Um and I know that it's apples to oranges but like I guess my question is right here, you know, UBS the other night put a they the stock was trading at I don't know. They they they tripled their, you know, um their they went to $1,600, you know what I mean? And a price and I know you don't give a about price targets, but that got the stock up 20% that alone. So like just help me think about Micron here.
So when I think about Micron, much like the discussion on CPU to GPU ratio, if you look at the Vera Rubin architecture that's going to be shipping very shortly in Q3 and you look at the specs for that, it uses like nine times as much high bandwidth memory as Blackwell. Nine times. I think near-term Micron's fine. Where I start laughing is when I hear people say, 'Well, it's different this time and that this is structurally a different company.' Because for those of for some of your viewers that that may remember this coming out of COVID, if you remember, a lot of the semiconductor companies had signed these long-term supply agreements because what happened? Well, COVID happened. Everybody had to get online. You had this big surge in demand. And then you had these semiconductor companies like Micron and pretty much every semiconductor company signing these long-term supply agreements because demand exploded on the internet and you couldn't get the supply. And so and I remember people saying, 'Oh, we're booked out.' You know, this is Micron talking. We're booked out for the next several years. That in 2023, Micron had negative to flat gross margins for three for four quarters in a row. Gross margins negative. So to think it's any different this time ultimately is ridiculous. The first time I by the way heard that the semiconductor industry was not cyclical anymore was in August of 1995. I was at the Semiconductor Industry Association forecast dinner. I had been a semiconductor analyst now doing the job as a research guy for a year and I forget who the CEO was that had to speak at the time. Um I but he was the CEO of one of the big companies out there. I don't want to name the name in case I get it wrong. I think I know who it was or remember who it was, but he said, 'Yeah, the semiconductor industry is not cyclical anymore because we have chips going into things like PCs, phones, automobiles, networking, etc.' This is August of '95. The semiconductor market peaked one month later and then it went down 55% in nine months and it was the worst down cycle we'd ever seen. Intel wrote off a billion dollars of DRAM inventory. Micron stock, you know, got absolutely destroyed. And so the point being is that every upcycle you have to have somebody come out and say, 'Well, it's structurally different.' Now, if you want to get to the architectural reason why I think you have a problem at some point, it's the following. When you use an AI, you want it to remember, but you're also trying to get your cost down. One big issue is the memory wall. In other words, as you're having it do these expansive things, you need more memory. But high bandwidth memory is like a hundred times more expensive than solid state memory. It's also a lot faster. I think what you're going to have beyond this at some point as you get into next year is you're going to have an optical backbone that then connects your CPU GPU to solid state memory which is a hundred times cheaper. And that optical interconnect is going to provide the speed that you want while simultaneously letting you get your cost down. And remember DRAM doesn't remember the power goes off. It forgets everything that Dan Nathan asked it. Solid state memory remembers everything, right? So there's some advantages to that as well. I think ultimately NAND is going to be much more interesting than high bandwidth DRAM is going to be. But, you know, this is a problem for getting this is getting way too wonky for...
No, no, but but but listen, but I can already see how this is going to end. At least how I think it's going to end.
And then we're everybody's going to go, 'Oh my god, how could we have seen this? It isn't really different this time.' Which, by the way, is what people were saying in 2023 right after COVID, right?
Which is what they were saying in 1996, in 2001, 2002.
