This Chart Signals the END of the AI Stock Bubble.

Quick reminder, coupon code expiration at meetke.com. Today segment we're going to cover the most bearish chart in artificial intelligence and it comes on the back of Anthropic wanting to increase its credit facilities massively. $15 billion expanded credit line before IPO. They're basically telling banks, hey, if you want an opportunity to be part of this IPO, you need to lend us more money. which they had a credit line of $2.5 billion. This is now 7x that size and 3x the size of the preipo credit line that SpaceX got. So, a lot more credit line capacity. Usually, you get a credit line as like a contractor when you've got a job to do for your customers, but you got to pay your employees in the meantime, and then you pay it off after you get paid by your customers. So, the hope here is that Anthropic is going to be able to collect more revenue to help pay some of those bills that they're doing, especially since they just signed a $45 billion deal with Lambda. That's a six-year contract that's going to cost them $7.5 billion a year for compute. They've got a compute contract going on with SpaceX. They've got one with Lambda. There's a lot of compute contracts. So, the Bills are definitely a big deal at Anthropic, but that's not potentially the most scary chart. The most scary chart actually has to do with what we've got here. This is a zero hedge article uh discussing what Goldman Sachs calls the most bearish piece. I'm going to jump to the most bearish piece right here for you and uh we'll get right into some of the core concepts of this. The very and and probably most important thing to know uh today is that we do have coupon code jhole expiring at meetke.com. Get in before that expires tonight because we're going to have a big price increase. In addition to that price increase, we are ending the reinvestomesai. That's at meetinvest.com lifetime access signup option. So, if you don't get into that today, you'll never be able to get lifetime access again. So, uh, consider checking this out at mekevin.com and meet mereinvest.com. Now, with that out of the way, look at this chart. This is a chart right here on LLM pricing. And this is, uh, usagebased. So, it's sort of, you know, weighted average usage based. How much money are people paying per 1 million output tokens? And really, what you've seen is we're flat from last year. last year is kind of right here where I've been covering it. This is really when this index started, this chart started. So, we don't really have it going back any further, but we had two lumps over here. The two lumps in my opinion came from bullishness on co-work and bullishness on uh coding. So, aentic uh work efforts for non-coders, but then also claude code and the big moments that we had earlier this year. So you could see the the peaks here in how much people were willing to pay for 1 million output tokens in January with another peak in May. Really, we collectively consider this the anthropic moment. The issue here is that this is now collapsing and there's a real implication that if that line is collapsing, what does it mean for the future of Frontier Compute and the spending that's happening? A lot of people believe that we'll just have the next level outcome for uh where enthusiasm for coding is going to come from uh or for AI. So maybe it's not coding anymore because we've already had the coding moment. Maybe it'll be healthcare or it'll be finance. But there's a little problem with that assumption and then we'll get back to the Goldman piece. See, the economist made a fantastic argument on why this coding moment for anthropic is potentially risky and not as durable as people think. And if we analyze it, they basically argue that coding is structurally different from many other parts of the economy. In coding with artificial intelligence, you have a lot of availability of data. Think GitHub. It's really easy to get a lot of data to train on. Think about how easy it is to validate coding data. It works without bugs or it doesn't. There's an exploit or there's not. It's a little more binary. Uh humans who use artificial intelligence in coding become blatantly and rapidly more efficient. So there's an incentive for humans in coding to code with uh artificial intelligence. Whereas some of the benefits in finance or healthcare could be more long-term and less immediate. And so you potentially have lower adoption in other sectors or less velocity of adoption. And then you also have this idea that engineers are the earliest adopters of tech anyway. That is only slightly different from this in that coders might benefit the most right away, but coders are also generally early adopters, software engineers. So the economist makes this argument that essentially says be careful thinking other categories are going to take over this AI explosion that we saw with anthropic where everybody was drooling over tokens and they were willing to spend $2 per million output tokens on average and now that has collapsed to under $1. Part of the reason this has collapsed to under $1 is likely because of the release of openweight models which are becoming more and more powerful. Now, that is actually bullish for parts of the economy and it's one of the reasons why we've seen Dell stock, which we've been covering on the channel. Make sure to subscribe. It's one of the reasons we've been seeing Dell stock skyrocket. And I think it potentially is going to help generate this hardware 2.0 rally before the big doom and gloom bubble pop. We'll explain that in just a moment. But focus on this for a moment. These this chart is not a use chart. This is a chart of how much companies are paying per million output tokens. When it falls, that is bullish usage. It's not a chart of usage. It is bullish usage. That is bullish for anybody running enterprise AI. Is bullish for anybody running local models. It's bullish for OpenAI uh purchasing Max Studios to run inference to get faster outputs for their customers. It's probably bullish for a company like Cerebrris issuing essentially ultra fast tokens via their hardware uh as well as uh Nvidia's uh release of the Grog model mostly because now you can differentiate a token. A token is no longer a token. There's a normal token and there's an ultraast token. Anyway, this is also bullish in my opinion value providers. uh and we've been saying this regularly about uh the stack that you know tokens are hard to monetize in the long term because they commoditize infrastructure commoditizes value providing is what wins. We saw this surge recently in software companies. We've been talking about this and that's why we had a software rally 2.0 thesis or sorry 1.0 thesis software rally 1.0 know is that in the third and fourth quarter we would end up seeing software stocks end up bottoming out and therefore an expanded software ETF like IGV which is giving back a little bit of gains today but nominal relative to this trend we can really see this trend and this bottoming process is over and now we're still on the trend up so the bottoming process is over the trend up is continuing for software these are the value providers these value providers in my opinion are also like what we're doing with our meat reinvest product you You don't have to buy it if you don't want. Just go look at it. Look at the product we have at Meat Reinvest and imagine you go wanting to buy a house and you look at a particular county here. I'll pull it up on screen just so you can see it and then you if you don't want to, you don't have to go over there. You go look at a particular county and say, "Hey, you know what? I want to go to the homes AI tab. Hey, I want to find a deal. Show me a deal. Okay, what do we got here? We got Ventura County. Okay, here's some deals. What's this? Here's a deal listed for 522. I know that deal. It happens to be a foreclosure. It's a good deal. Okay, the AI instantaneously tells me, Kevin, if you were looking for a deal in Ventura County today, look at that deal. Wouldn't be surprising for it to get multiple offers. And honestly, I actually think the app is conservative on it. I think the after repair value on this is probably closer to about 675, but you might have to spend a little bit more on it. So, you might have to spend closer to 60 or so. So, it's actually relatively close on its estimate, but this is a deal people can gain equity on. This is not a pitch. This is an example to you of how you can use, you know, we run these on GPUs that we own and CPUs that we own. We own the hardware. That is an example of how cheaper tokens that we can get an ROI with on our own machines can actually provide real world value to people, to agents, to brokers, to consumers through this app. And if we sell that app, we make money, the consumer, and the customer makes money. It's win-win. It's like what Palanteer pitches. So, cheaper tokens benefit us. I cheer this chart, but it does not mean that there is not a big problem. And that's what we're going to talk about here. Now, quick note, like I just want to be clear, you know, we're not fundraising or anything right now. People out, Kevin, this is crazy. Like, I thought you were your startup was a real estate company. It is. I just want to be clear about that. We're a real estate company. That's why we changed the name to reinvest because we just take profits or you know extra uh leftover money and reinvest it pretty much all into real estate. Uh and so yes, real estate company but secret AI prop tech firm. It's kind of cool. All right, so what do we have here? Here's the big problem. The AI trade just ran into a unit economics problem. And this is coming from Goldman Research arguing that there's a major concentration of revenue where Broadcom is pitching how big of a customer uh Anthropic is going to be, but how that could actually end up being problematic because Anthropic and OpenAI are material customers. And if there's a drating occurring in tokens, then you might end up seeing the price of work collapse faster than what the compute buildout has been uh expecting, so to speak. Now, I call this the V-shaped problem. So, let me first explain what I think the Vshape is, and then I'll make that comparison to Coinbase in just a moment. The V-shaped problem in my opinion is that right now a lab like Anthropic makes a lot of money off of selling Frontier tokens. I think the next growth curve for a company like Anthropic is really this scurve on selling you software. So that's the ARR on selling you software. The problem is this is going to fall and so you end up getting in my anticipation, right? We don't know guarantees, but my expectation is you're going to get an intersection where you actually get what's looks like a V. You kind of see that V right here, right? See that? The down and then the up. Okay. The quicker that V is, the better because software picks up the decline in token margin basically, right? Anthropic basically is pulling a Coinbase to some extent. When Coinbase IPOed, the whole pitch was, hey, how much money we make on the spread of people being able to buy Bitcoin on our platform and tokens on our platform, well, coins or tokens on our platform is going to decline. So there was an expectation that they were kind of IPOing at peak tier one revenue, if you will, and that they would make that up by making more money on tier 2 revenue, which was basically like financial services related to crypto. And we've seen how long it's taken for financial services for crypto to really take off. And then there are a lot of other structural risks now in that, such as Fed Now being another structural risk. In fairness, some of these companies like Coinbase recently are doing really well, but that's because Bitcoin popped up from like 60K to 80 and there's like a lot of enthusiasm around crypto again. But that doesn't change the fact that we were branded a V by Coinbase and what we got was a really long V that actually looked more like an L. Uh, and so that's what I wrote on the side here. That's the risk for Anthropic as well that you end up getting an L and it takes really long for them to build out that ARR business to offset that token growth because as they argue here more growth plus lower prices is not automatically bullish if pricing falls faster than consumption grows. This is the standard Ba base case. pricing for tokens collapses so fast and even though consumption is growing, you're either selling them for smaller margins because the Chinese models are compressing those margins or you are selling um just overall less profitable tokens. So I guess that's almost the same thing there. You're selling like the growth rate slows down for the tokens andor the margins collapse. Those are that's sort of the AB component there, right? So, think about the Google IO event for a moment. Google IO event, wow, token usage is up 50x from 2024 to 2025. And then if you look at 25 to 26, token usage is up 6.9 uh times. 6.9 times. So, you're up 50 times of 6.9 times. Well, that's still massive token growth. You're into like quadrillion tokens advertised at the Google IO event. But the second derivative is negative. The second derivative is the growth rate. The growth rate went from a massive positive number to the growth of growth rate which has turned negative. We went from 50x to 6.9x. That is a decline in the rate of growth. Second derivative. So the rate of growth can decline and you can get a decline in pricing because of openw weightight compression. And that is what we're already seeing in the chart right here. And that's why this is the most bearish chart in artificial intelligence. Yes, right next to that coupon code expiry maker Kevin.com. Many people say that could be a tax writeoff uh for your investments or your education or however you want to uh declare that. talk to your CPA about that. I'm not your CPA. They do say that there are some upside potentials. So, we don't want to be all bearish here. The upside potentials they argue is that a gentic demand could end up boosting token usage and margins. Maybe that maybe software companies can get their profit faster, right? The Vshape could be tighter. uh or we could get frontier models like Astra that actually lead people to be willing to pay more for better compute. I personally think that the hope that better models like Astra are going to lead this to go up is fading. Like you might see a very short up and down like this. So you get kind of a little lumpiness uh like what we've seen over here, but probably even less so on new models going forward. So really, you're hanging a lot of your hat expecting that this AI buildout is going to keep going on frankly margins staying wide enough for Anthropic and Open AI to keep paying their bills. That's the core base case. And the more open weight expands, the more risky this becomes. Now, does that mean things are over? Nope. In my opinion, if this is a circle, it's a pretty crappy circle. you know, it's it's such a crappy circle, but I want to be clear about my opinion because people sometimes get confused, and that's fine. We'll add some clarity here. I think we're probably here in the uh AI hardware cycle. Okay, so that means I do think we still have another little rally to go. I call this the hardware 2.0 rally. Uh but during that same rally period, I actually think we still continue going through the software uh rally. That doesn't mean all software companies are going to win. This morning in the course member live stream, I know a lot of uh course members uh ended up selling UiPath because we identified some red flags that were not immediately obvious in earnings. And so if you're a course member and you haven't seen that course member live stream yet, I highly encourage you watch it because rather than just looking at the last financials, we looked at some other documents that indicated, uhoh, there's a little oopsy dupsies and a red flag built into what's happening here. What is this going to mean in the long term? And so we broke down our opinions on that. So not all software uh is going to win, but I still think the software rally could stay durable while we have a little bit of a hardware rally 2.0 going. The only way hardware rally 2.0 keeps going is if Anthropic tells us despite this decline in token pricing, we can still pay our bills and we're doing good or that software revenue is coming in. Other than that, it becomes quite bearish for how close we are to the top of potentially the AI trade. If oh, Anthropic can keep paying its bills, then Broadcom looks really cheap. So, you kind of have a binary outcome here, right? anthropic uh bullish S1 on on their IPO documents, then that's going to make you bullish uh frankly Marll and Broadcom and and also AMD and Nvidia, you could throw those in as well. If that ends up being bearish on the S1 or sequential quarter overquarter earnings look bad at Anthropic, that would be bearish the same stack and would mean you're closer to the top. So yes, that means simultaneously I have the thesis that we could still have another good six months. You know, I think hardware and stocks overall can actually hit all-time highs maybe for the next 6 months, maybe 12 months. But we have some serious risks coming by the end of 27 and into 2028 as memory prices start peaking out on their growth rates and as much more supply for hardware comes online. Now this piece goes on to mention that and I think this is actually overly bearish but they argue that H100 pricing is also falling. I don't think so. I think this is normal volatility. You see relatively close to the top. So I actually see this as more bullish uh and and cutting towards my argument that we still have time uh not infinite time but we still have time right now. over here you they mentioned that uh JP Morgan's data uh has noted that volumes for tokens are up 47% but dollar spend was only up 7%. That same compression is bad for those frontier labs in other words and that even though token growth is exploding you've got that second derivative growth rate that's a problem. They argue that yes, maybe Astra can provide an additional shot in the arm. But let's be clear, cloud inference priced in millions per tokens only works while the model is too big to run locally. This is a critical issue and it's exactly why a company like Dell is doing so well right now. Dell is doing so well right now because the more you could expand local compute hardware and stand up a data center in your office right now, plug it into a 240 or 370 volt architecture. Boom. You get online because Dell comes and drops a rack, loads up the software for you. It's got the switch built in. It's got the UPS built in, not the company that's heavy in debt and pension liabilities and the battery backup. Uh, it's got the GPUs, the CPU, it's got the full stack in there. It's got the copper wiring, the fiber, uh, whatever, the Ethernet, the InfiniBand, it's already done. That's why Dell is winning because companies want this locally because there actually is an ROI. I get a lot of people say, "Oh, Kevin, how is there an ROI in owning your own compute?" There is. That's just the way it is. is we are saving massively more money than we are spending on our hardware in compute that we would be spending to run things like uh our homes AI software in what we call our terminal which of course we could change counties and we can go to different areas but what's also kind of cool is you could even see properties that would lose you money if you bought them and so we're expanding that we actually have a whole lot more in the pipeline but uh you know we're releasing this sort of data in time as that's still in beta and it gets better every day. It's very exciting. But uh understand when we look at this, we also have a problem coming from the fact that Mac is now blowing out these amazing 256 to 512 unified RAM products, which is bullish. Apple honestly could totally surge their growth rates from a Wall Street expected 9% to 20%. Which again, bullish Apple. I want there to be a discount on freaking Apple stock because I'm bullish Apple. I just think it's a little pricey right now. Like if I'm going to pay the Apple pricing right now, I feel like I'm better off buying cyber security plays at at these sort of values. Totally topic for a different video. Uh but this suggests more competition for those frontier tokens being sold. the more sizable these uh and and more capable these local computers get, the more you could run massive models that you just otherwise cannot run uh elsewhere. You know, I've got uh actually got a little note on this here. Where is it? Artificial intelligence. Let's get this little note here because I think it's very interesting. So, because we'll actually look at some of these. We already know this is old news. Okay, that's sort of my base case thesis. tokens and infrastructure commoditized. We already know that. But I want you to see this because this just you could take a screenshot of this. You know, this just compares a local computer of a 5090 to the Mac. And even though the 5090 could be 50% faster, you hit a lot of red X's really fast on that 5090 in terms of what kind of models you could run. I mean that Max Studio can run DeepSeek R1, Llama 4, Quen 3, 235 billion parameter model. The 5090 caps out at Quen 3 32 billion uh parameters. So you really hit uh some capping out. And this allows you to do a whole lot more in terms of local uh agentic reasoning, coding agents, longer research sessions, uh more filing review as a financial services company, better compliance, lower hallucinations, better compliance for legal work. You can do this locally. It's fantastic because now you control the data and you don't get the hallucinations that come from not having enough uh head space in your memory. So this is where they argue that these shops end up they they stop becoming a token customer and they start becoming a hardware customer but they only become a hardware customer once. That's the thing. It's like once you bought your local hardware you don't really have to go buy it again. That's the bare case by the way for Dell. The bare case for Dell is what's the terminal growth rate? Yes we're booming now but once people buy that hardware what's the renewal cycle look like and the growth therefore right? And this is where anthropic really boils down to. As they write here, durable commercial models start to look like seats, agents, and outcomes. Holy smokes. What have we said here? For the longest time, oh, the winners are the value enablers like software companies. Holy smokes. Thanks, Goldman Sachs. You've been subscribed to the Meet Kevin videos. Heck, they're probably subscribed hook, line, and sinker to the alpha report we send out every day. That's freaking been on fire lately. Let's go. [laughter] And that doesn't mean it's always a big W, but boy, I yesterday we called Tesla running up uh and software running like Palunteer. Today, give back on software run on hardware with specifically a call on a bounce on uh Marll and AMD. There's some really good calls lately. Again, I'm not saying they're always great. You don't want to get overconfident because mistakes happen. Uh but uh I'm very very happy with how our alpha report has been the last couple days. So, uh looking at the rest of this here, they actually then move on to suggesting that advertising could be the future. Holy smokes. This is why Kevin thinks advertisers are the sleeping giant investment opportunities right now. There are three and I have exposure to these three so I want to be very clear. Okay. Also got exposure to Marll. Want to be very clear. Very tempted by Broadcom but don't own it. But got exposure to uh Meta app and uh Netflix because I think advertising is the sleeping giant right now. I think you can get into these cheap right now while you know software is already taking off. We got exposure to Salesforce as well but all these companies are going to wait. Even like a Service Now, if you read the Service Now earnings calls, you'll see they're complaining about hyperscaler costs. Those go down with this information and what's actually happening. Keep in mind also, Salesforce just did a massive ad campaign and a million dollar giveaway. I don't know if Salesforce paid for that million dollars. They may have, but they did that with Mr. Beast. I think more of that is going to happen from these value providers because they're this is like a land grab right now. They mentioned that excess AI demand therefore is not automatically bullish because you end up squeezing critical material bull cases for artificial intelligence like anthropic and open AI and that's potentially why credit spreads widened. This is an inver inverted chart, so it's a little bit harder to see, but the blue line is stock prices. The red line is credit spreads. As the red line goes down, and the blue line stays stable, you're basically saying the spread here is an increase [music] or so plus risk, right? Risk is going up. That's the argument over here. And then of course, they summarize this by saying tokens being 30% cheaper, but volumes being up 20% means that inference revenue falls. And I write the note that obviously that crush is anthropic and then you run into this V-shaped risk factor. Maybe bots will save us. You know, Grockbot, but there's some major risks here. So, something to keep in mind for the AI. >> Why not advertise [music] these things that you told us here? I feel like nobody else knows about this. >> We'll we'll try a little advertising and see how it goes. >> Congratulations, man. You [music] have done so much. People love you. People look up to you. >> Kevin Praat there, financial analyst and YouTuber. Meet Kevin. Always great to get your take.