**Prepare** Critical Nvidia Stock Earnings

The Wall Street Journal just reported that Anthropic expects their potential revenue to be $30 trillion at some point in the future. That is like a total redefinition of total addressable market since that's basically 100% of the entire GDP of the United States in 2025, which sounds bullish maybe for Nvidia. And that's what we're going to talk about in this segment. Nvidia expectations for earnings. the new jalapeno chip from Open AI. We got to talk Nvidia robotics, Nvidia openweight, enterprise AI, GPU rental rates. What's the trend that's going on? What is Edgewater telling us about GPU and server demand? Where is server demand plummeting and where is it taking off? Let's break it all down and let's get started. So, first thing that we need to do is uh recognize that Nvidia earnings are tomorrow. Uh they are as usual after the bell. We do expect Nvidia to beat. Historically, they always end up beating. Uh current uh numbers just so you have them really quick. On adjusted EPS, you can write them down. So, when the the news comes out on the wire service, which you could run that through the uh meet Kevin Alpha app if you want, or I'll be live and we'll talk about it as well. So, uh write these down. Adjusted EPS expectations $29. Nvidia beats seven out of eight of the last times. Uh EPS is expected on the gap basis at $2.15. Nvidia beats eight out of eight times on this. Revenue is expected at 92.29 billion with a beat eight out of eight of the last times. Bank of America actually thinks they'll be closer to 94 billion. And that ends up being a really big deal for Nvidia because even though Nvidia can beat on revenue eight out of eight times, Nvidia in the last four earnings releases has gone down afterwards, which is an insult to this business that makes so much money and is so freaking cheap right now. It's crazy. Uh but anyway, uh $94 billion is the Bank of America expectation. you really have to beat the expectations uh that are on the higher end. So current expectation on average consensus is 92.29. Usually for the stock to go up, you have to beat the higher of the expectations. It's such a weird like psychology game. Anyway, net $50.98 billion expected, which if you do the math on that, the net divided by the expected revenue means they're taking about 55.2% of the bottom line. every dollar that they make, 55.2% is after tax profit. It's so good. Anyway, um the big focus on this earnings report is all going to have to do with margin, especially since we just saw a 15% price increase, which was actually lower than the 20 to 30% price increase that Edgewater Research was projecting for the third quarter. We had a 15% price increase to offset some of those memory prices. And we've seen some respecting and some rejiggering of some of these leading edge uh Nvidia chips where we're seeing them come in with lower headroom for uh VRAM. So basically lower uh exposure to that memory issue which then makes you wonder like why why are we making like creating shrinkflation in chips? Why are we reducing memory there? But it could also be that maybe we just don't need that much memory in these chips. We'll talk about that in a moment as we talk about enterprise AI, which I think is a really important focus and it makes me really excited about the future and investing in stocks because I feel like we're sort of as a startup at the bleeding edge of what enterprise companies are doing with artificial intelligence, which is totally different than in my opinion what the Frontier Labs are doing. Anyway, we'll talk about that in just a moment. So the Jalapeno inference chip was just announced by uh well OpenAI via Bloomberg. So Bloomberg released this article that says OpenAI claims its new chips can outperform Nvidia processors in tests. This chip is called the Jalapeno chip uh and it beats the GB300 in their benchmarking. Obviously this is biased and I'm bringing this up because it is something that people see as a risk factor. Now, this jalapeno chip is made in partnership with Broadcom. Now, Broadcom is a really interesting play because they are pretty heavily indebted. Their balance sheet kind of sucks, but their valuation is dirt freaking cheap right now. Uh they are trading for under a one peg. We just did a big analysis on them in the course member liveream. I always encourage if you haven't yet, consider joining the meet Kevin membership. uh and you can get this analysis on a regular basis. We do have a massive price increase at the end of the week and we're changing the pricing structure. So, existing members will be getting sort of a goodie bag that other people joining afterwards won't get. But, I mean, Broadcom according to our stock AI has a uh $95 price target. The issue is we have a red flag on the balance sheet for the company because they are so heavily um you know indebted. So, your your balance sheet is a red flag. valuation is just a clean green flag. They're down like 26% from their peak after their last earnings because of the timing of delivery of AS6. Okay. So, why what does this have to do with OpenAI? Well, they're doing the AS6 for OpenAI. That's they're also doing the Google TPUs. So, OpenAI now like, hey, we're getting ready to tape these out probably bullish Broadcom. It's kind of bullish the whole stack. But these ASIC producers, so think Marll and think Broadcom. Uh Broadcom has better margins than Marll, but uh Marll has a better balance sheet than Broadcom. Marll is also a $200 11 billion company. It's up about 5.6%. I've got exposure to Marll. I don't have exposure to Broadcom, but I really like Broad. Like I'm really tempted by it right now. $1.7 trillion company. But these are your core sort of ASIC producers uh as as uh inference plays basically. And that's what OpenAI is saying here. OpenAI is telling you look we're making this chip not to compete with frontier bleeding edge training for the next GPT56 model or you know the next anthropic fable model. What we're doing is we're trying to create a chip that reduces our cost structure for when we're providing inference solutions and we're trying to get those wherever we can get our hands on. If we can get them from A6, great. If we can get them from SRAMM based chips like the Nvidia Gro that was just announced for ultraast token generation, super low latency, imperceptible latency for like voice AI customer service. That's an Nvidia product. the Gro they they acquired essentially all the IP from from Grock or Gro G R O Q that's another uh competitor to the Cerebrris inference chips which are really good I hated Cerebrus at IPO I thought $370 was way too expensive but now that it's come down a lot like half from there I actually think it's a pretty good deal just little sidebar notes on where my head is on all of this you can see there's a lot benefit from the now Frontier Lab saying, "Hey, you know, we're we're not just focused on the greatest leading edge frontier chips from Nvidia anymore, like the Vera Rubin chip. We're also really focused on how can we fill out our inference stack and Nvidia is realizing this and that's going to be one of the risk factors for Nvidia. And it's probably why Nvidia is selling for such a dirt cheap valuation. Nvidia's PEG ratio right now sits at about 69 which is an insult to Nvidia how cheap this is. This is why our stock AI indicates this has like a 4x upside at our price target is $886 for Nvidia uh at at a fair value just because their margins are so high. Why then are they selling for so much lower? Like what's going on? Well, let's simplify it and understand uh why Nvidia would potentially be selling for this much of a discount. Okay, so first of all, uh you've got Nvidia. It unfortunately has this large company deficit. You know, it's $5.7 trillion. It takes a lot for a $5.7 trillion stock to move. So, that makes it hard. That's one thing it has going against it. The other thing is it is not the sole competitor in uh or competitor in ultra fast token generation because you have Cerebrris who's actually ahead of them. This just entered production with Nvidia which it's an Nvidia product now so it should do well but Cerebrris is ahead of them. Cerebrus just needs to scale up how quickly they can manufacture which Cerebrus expects to get to scale in 2027 and 2028 probably around the time that Gro does. But the issue here is you have almost two equal I would almost say Cerebrus is ahead competitors whereas Nvidia has this big moat in the bleeding edge right you think about like the Vera Rubin uh space compute which in fairness you've got Elon Musk talking about how Vera Rubin space optimized chips are going to get launched into space in Q427 and then scale in 2028. That's great, but that's a bit of a long shot. That's sort of your uh Vera Rubin for space. I would call it uh hope right now. It's it's a little early for that. And so that's getting discounted. Grog is getting discounted. Uh its size is getting discounted. Like can this really 2x from here? Right? That's the problem. You know, Forex is going to require this being like a 20 trillion dollar company. Based on how much money they make, you could justify it, but can you actually get to it is the other question, right? Then robotics are still very early. Nvidia is uh just announced uh yet another uh sort of robotics release, if you will. It's the Jetson Orin Nano2 robotics computer for sort of lower power edge computing. That is the future. But robotics again, it's a future hope that could be a decade out before we really scale GPU heavy vision-based training for robotics. You're more lab-based, edge, edgebased. Right now I should be careful with the word lab since the frontier labs use a crapload of GPUs right like anthropic and open AAI that should not be confused with lab-based robotics which is more you know very very early stage mostly because you still have issues with actuators and the actual mechanics of keeping these things running for more than you know a 30-minute demo. So anyway, so you've got some of these issues. Large size. This is maybe Cerebrus might be beating them here. I mean, they already are to market. They've got Cerebrus has customer concentration risk with OpenAI. Some of their uh IPO issues have derisked, especially now that the stock has come down as much as it has. So I like it better now than I did then. Vera Rubin uh space is more of a future optionality. Robots are future optionality. So, a lot right now comes down to margins. And this is going to be the focus from folks is if you're raising prices 15% but memory prices have gone up substantially more than that, how stable are your margins going to be going into earnings? And so there's some nervousness that maybe we'll see some softening of earnings especially as some of the focus from uh enterprise companies goes towards inference chips rather than these bleeding edge frontier chips. Think about inference chips for a moment. This sounds crazy, but the gaming PC uh that operates or any gaming PC that operates on a 5090 GPU is pretty much good enough to run a lot of enterprise artificial intelligence. You don't even have to buy what we did, which is a Blackwell 6000. We've bought, you know, a few of these Blackwell 6000 chips. We bought these back when they were $7,000. We 2xed our money on the damn chip. Like, I don't even know. I don't even think the stock 2xed in the time that the chip 2xed, right? Uh, so it's kind of crazy, but a lot of enterprise AI does not need this memory level. You don't need 96 GB of DDR7 headroom. You can handle a lot of enterprise uh, inference with a 5090, which operates at a third of the size of this on a 32 GB uh, uh, you know, VRAM set. So what does that mean? Well, what that means is you might end up seeing companies preferring A6 from Broadcom, Google TPUs, which are expecting to scale next year, uh, or Marll A6, uh, or Nvidia 5090s or 6000s over the bleeding edge Vera Rubin chips. That, in my opinion, is why Nvidia is starting to say, "Hey, we're actually going to downsize how much memory we're putting on these massive frontier uh uh chips because the enterprise applications don't actually need that much memory on those big chip stacks." So, there's been a lot of rumoring in social media over, oh, they're reducing how much memory they're putting in the chips. This is shrinkflation, you know, they're they're cutting corners. I think it's so they could keep the cost down and actually make it useful for enterprises because the chips are getting too big. We don't need these massive stacks because you could do so much work with these little chips. You could literally take a $4,900 I think is the retail price. That's not even what you could get them at bulk pricing, right? You could take an Nvidia 5090 RTX chip right now. Oh, they've gone up another $200. Are you kidding me? These were literally $4,900. one left in stock at least on this store. Uh but anyway, you know, this was $4,900 like last week. What? And the delivery, you can't even get it until September. That's stupid. Uh so anyway, yeah, I mean, these are selling out. One of the reasons these are selling out is because you can take a company like Mona who's doing fantastic work with using artificial intelligence to maximize their speed to therapies for their clients. You could, in my opinion, do what they're doing with artificial intelligence on a 5090 GPU and you don't need these crazy uh Vera Rubin CPUs. I'm not GPUs GPUs. What I mean by that is, and I want you to visualize this because I think it's really useful. Back in the day, it used to take Mona, like back in 2022, uh 2022, let's call it, it would take about 60 days to sequence a cancer patients tumor and then select which mutations they wanted their therapy to target. And so then in 60 days, they would target maybe one to five mutations in these cancer cancer cells. That was not really ideal. You want to target a lot more mutations so you have a more durable solution. I'm going to make a healthcare video on this and explain that separately. So subscribe if you haven't yet. You'll see that. Now in AI in 2026 or with AI, you could actually do this in hours and you could get to 32 m mutations that you can target in these therapies custommade for a cancer patient. and you could literally run it on a 5090 gaming GPU. The problem is these gaming GPUs are lower margin for Nvidia than the big $4 million racks. Now, one of the reasons this is so relatable to me, this this Mona example, and I I don't like to sort of sound like a broken record, but that's what we do with our Reinvest AI. It's our, you know, our homes AI where you go into a a county and you're like, I want the best deal in this area and it sorts deals by here's how much money you can make. That's proprietary AI. Uh, our stock AI, proprietary AI. Uh, we're not making chat bots. We don't care to compete in that space. We're trying to compete in a in a practical AI place. Just like Mona, they're not trying to make you a chatbot. They're trying to give you practical artificial intelligence. And their practical AI could work on 5090s. And you don't even need to stack them together. You know, there's this idea that, oh, you could stack four 5090s together and operate them together. You don't even need to do that because you could just do four patients with four 5090 GPUs. Rather than stacking four together to solve one problem, you can solve four problems. The point of that is enterprise AI is going towards smaller chips, not bigger chips. Uh you've got Gavin. Gavin uh just had a post about his outlook for uh this paro frontier. I think he's trying to sound fancy about how uh artificial intelligence uh ends up getting allocated. He called it the paro optimal level. You know, he was just on the all-in pod not that long ago, but anyway. Uh I think it was like 10 days ago he was on there. He says, "Suspect we'll start hearing about a paro optimal balance of computational efficient humans, cheaper open source tokens and frontier tokens." Let me just explain that very simply. I personally think this is again my opinion that you can break down what he just said into a pie chart of three things where a like spending goes. So you have labor, you know, so people with human labor, human labor IQ. Then you're going to have frontier tokens and then you're going to have enterprise tokens. I personally think this right here, frontier tokens will represent like 10% of the slice. Labor will represent like 45% of the slice. Enterprise 45% of the slice. You know, it's probably something like that. And this would only be about 10% of the slice. That's my take for enterprise. I I could be wrong about that, but that's where my head is and that's where my investing thesis builds around uh AS6 rather than trying to sell the frontier level chips. My take. We'll see. We're seeing a lot of companies do that and the payback periods are honestly like one to two years on a lot of those chips. We were calculating some of this in uh in our course member live stream. Remember, you can always get our trade alerts over there. See the long-term portfolio, how we're investing for the next decade, how we're allocating. But, um, join that over at meetke.com before Friday. We got a big expiration there. But, uh, you know, we were calculating some of the break even periods and we're like, yeah, you know, our chips have basically already paid for themselves. So, there is some reasonleness to the idea of short payback periods. The question is, how long are those pay payback periods going to last like that? To be determined, right? Those are the spot prices we're monitoring. How do the spot prices turn into long-term contracts? We'll end up seeing. But Thomas Reuters is doing this as well. They just announced uh that they're using Quen, which is Alibaba's open weight model, and they think they're 3 to four months behind the big frontiers at a fraction of the cost. And they keep their own data. So, they get to in-house their own data. Mona gets to in-house their own data. Reinvest gets to in-house its own data. We don't have to send it all to the labs. The problem is that's bad for margins for Nvidia potentially in the future if they're now selling fewer of those Frontier chips and more of the enterprise chips. You get compression there. Now, those are some of the issues that people are worried about with Nvidia. I'm not really worried about that just for this next earning set because I think Nvidia is just so cheap and I think the stock has already discounted a lot of that. Again, if you're trading for a 69 peg, it's dirt cheap. Let's actually go look at the financials so you could see it. So, these are the Nvidia financials right here. And these are the numbers we have for Nvidia. The current growth forecast for August 25th are that uh at the end of the year they will generate $9 of earnings per share and their growth rate is expected to be 44 24 87 and then negative 22%. So you do have sort of a a curving out on growth at the end of the decade. That works out to about an average growth rate of about 33%. which uh means they're trading for about a 69 peg right now which is very cheap. Part of that could be because there is negative growth forecast in about four years. That could be as we see Frontier trip chips become less critical and the inference chips much like the jalapeno from OpenAI created in partnership with Broadcom taking over. That's bullish for a company like Broadcom. But bear in mind, and this is in our stock tab as well. You can see this, join us, join the courses. But this is literally from their earnings call. We do a lot of research and we try to provide this every single day. Uh within semiconductors, we've always said our A6 and TPUs have lower margins. So as TPUs continue to accelerate, there will be pressure on margins. Broadcom's margins are way better than Marll's, but Marll's growing into the Broadcom style margins in my opinion. Now, what's like what are some things that are really bullish for Nvidia right now? Well, $500 billion of financing for data centers and the SEC changing the assetbacked uh securization rules. That's freaking crazy. Okay, that's like 2008 kind of stuff where the SEC says, "Hey, you no longer need 5% skin in the game if you're going to sell asset back securities for data centers because we think this time is different." Oh, we're also going to loosen the rules and say you don't actually have to disclose what the components of these assetbacked securities are or what the utilization rates are inside these data centers. To me, that sounds like 2008 all over again. I don't like that. I personally still maintain that, you know, you don't really see 2008 style tops until you get uh like Anthropic rollover. You know, we led this segment with this idea that Enthropic is projecting $30 trillion of potential revenue. I mean, that's like the greatest TAM sales pitch you could potentially ever come up with. Uh but whatever. Those rules are short-term bullish, more debt, but also more spending. And that's where one of the weird ironies comes in. A lot of people are talking about, "But Kevin, credit default swaps are skyrocketing on these plays. Like, isn't that isn't that a risk?" Sure it is. But there's an irony here. Hedge funds who were exposed to Meta, Oracle, Google, Nvidia, because Nvidia credit default swaps are also rising. hedge funds who are buying Nvidia or Meta and Google and then they're hedging with credit default swaps are actually winning on both sides right now. Think about how weird that is. Usually, usually a hedge would mean if Nvidia stock goes up, a put contract would go down. But that's not what's happening here as a hedge. In this case, as Nvidia goes up, credit default swap hedges are actually going up. So, if you own Nvidia and you own credit default swaps, Nvidia goes up because there's more debt going into the system creating more spending. Nvidia stock goes up and that extra debt then makes people nervous. So, credit default swaps go up and then your hedge actually goes up, which would be your insurance plan in case Nvidia defaulted, right? So, weirdly, we're at a place right now where hedge funds can literally hedge and see their hedges and their stocks both go positive together at the same time. Uh, if Nvidia or Oracle or whatever go down, those CDS's, ironically, might still go up. So, it's sort of like the best hedge. The stock goes down, your hedge goes up. The uh stock goes up because of more debt financing, your hedge goes up. It almost makes you just want to buy credit default swaps. I mean, maybe that's not a bad idea. Obviously, that's derivative based, trading based. It's a little different. But think about the other things that are bullish for Nvidia. SpaceX spending, $80 billion of spending or $80 billion raised on IPO. They're probably going to raise another three to $400 billion. Uh Nebas has $8 billion of cash on hand. Google had just raised $86 billion for uh for Capex. You've got the Anthropic IPO coming. Yes, everybody wants their hand in the cookie jar to get a piece of these chips. Yes, we're going to see more movement towards inference chips and enterprise chips versus these Frontier chips in my opinion. But so far, Nvidia looks pretty freaking discounted for all of that uh you know in mind. So, let's look at some more more of these takeaways here. Uh and let's look at a little bit research from a little bit of research from Edgewater. So this is a uh a bunch of server based research for Nvidia and CPUs. GB200s to 300s had were expected to increase by 20 to 30% in Q3. Those actually ended up getting increased by about 15%. So a bit lower than expected. Server CPU deliveries are expected to rise 50% in uh 2027 which is really fantastic. Uh that's right here. uh Nvidia or Intel CPU capacity is sold out through 2027 or into 2027. It says price increases are still expected on Blackwell products, but they haven't happened yet. They just happened. This was published like days before the price increases. The cost increases are mostly memory as GPU prices are unchanged and Nvidia has secured enough supply. So that does indicate we expect all that attention on margin. Did you guys raise prices enough to offset the memory cost increases? That's where there's going to be a lot of focus on Nvidia earnings. We'll cover those tomorrow after the market close. This uh this idea about package shrinkage. A lot of people talking about how this is just inflation from memory. I personally also think it's just staying competitive with prices for that enterprise AI. We touched on that. We've got um the RTX 6000 right here, which we I just showed on Amazon, has gotten an increase on its MSRP from $8,000 to $11,000. So, you are increasing the prices 37% on enterprise chips because that is where the demand is going, which is pretty impressive. I agree with that. Channel demand for servers is declining and gaming demand is weakening. gaming and PC deliveries. Not a surprise that we're seeing weakness there. I do think it's interesting that Edgewater literally says we don't know how customers continue to get more of a budget for AI servers, but standard compute servers are in decline. So, I think when they say here channel demand for servers is starting to decline, I think they're talking about traditional servers because AI servers so far are still booming. uh as they say here standard compute our customers are only upgrading if they have to. So that's your classic cloud compute rather than AI compute. Uh the other notes that they make is that they expect DRAM prices to cool in 27. This is also not surprising. We could see price increases. So that second derivative, the growth rate gets sharply negative into 2027 with DRAM prices only expected to grow 0 to 5% in 2027 depending on which stack of DRAM you're in. Whereas at the beginning of the first half of 2026, you know, these were all up 60 to 120%. A lot of that is expected to cool. Edgewater also argues that Nvidia might be starting to feel some pressure on uh openw weightight models which is why they're getting in. You know, they're spending $6 billion to get in with poolside to launch their own openweight models. But they're also possibly trying to compete against the Deep Seeks and the Moonshots as they mentioned here, open weight Chinese LLMs. And I mention that I think what's happening is there's an economizing of spend going on. So, hey, why are we going to overload these chips with all this memory if people don't need it? Let's cut the memory, cut the entry cost for these chips, get these GPUs into the hands of people, uh spend less on memory as a proportion of that and then let them go apply their enterprise uh AI. Again, though, the question is ultimately what is the margin for a Vera Rubin versus an RTX 6000? And that's probably why we just saw the RTX 6000 go up 37 and a half percent in MSRP because that's where enterprises are moving to. And it totally makes sense. It's what we're seeing as well. Uh and I always like to try to provide that insight of like, hey, this is this is what we see going on. You know, when when we were buying a bunch of CPUs to process uh some of the data rich applications we're doing, we um uh you know, we we ended up uh calling out AMD in our course member live streams. We're like, man, this is going to be good for AMD. The freaking thing doubled. So, uh, you know, sometimes we can we can get a little bit of insight onto where the trends are going just by sort of being in the trenches of of building out AI software, which is really cool. Uh, so, um, anyway, uh, what else? Uh, I mean, that's that's kind of the core set of info for for Nvidia. Uh again, they're expected to beat across the board as they historically have, but that's not what's going to move the stock. What's going to move the stock is what we hear about margins, frontier chip deployment, competition against open weight, lower margin enterprise. Those are the red flags we're going to look for. Are we really going to care right now about space or Groot for the robotics AI or the robotics chips? Not really. Do we think Gro is really going to definitely be better than Cerebrris for an SRAMM inference chip? We don't know. We'll see. Remains to be seen. All those things are really a sideeshow. Gaming is expected to continue to decline. That's also a sideeshow. So, not jumping up and down about really caring about any of that. It's all about that frontier spend. And what's really bullish is that you do have a lot of circular financing going on. A lot of money flowing into the um uh you know anthropic IPO likely at least. I think that S1 drops they'll raise a lot of money and they'll just spend it on chips, buy their own ships even or or spend it on uh the Neoclouds like an NBIS or or whatever. Another thing we'll be looking at is Nvidia's investments into other companies. You'll be able to see that not as capex. It doesn't show up as capex because you know capital expenditures will show up here in PPE. Where you'll actually see it is right above that in purchases of non-marketable securities and marketable debt and equity securities. And you can see this about $26 billion of spend right here in investments so that those companies they're investing in can go buy more Nvidia product. So that's where some of the circular allegations come from. So we'll be watching for that as well. But it's going to be a big day tomorrow and I think a lot of folks in the market are a little kind of like tenuous and nervous. I actually think that creates a buy the dip opportunity today and tomorrow because I, you know, I don't think we're on the turn of a recession uh anytime soon. So I think there's an opportunity to buy the dip over the next couple days. We'll see. Uh and uh when we go buy the dip, we'll send you a buy sell alert. Make sure you're part of it over at meetke.com. >> Why not advertise 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 have done so much. People love you. People look up to you. >> Kevin Praath there, financial analyst and YouTuber. Meet Kevin. Always great to get your take.