134 The AI Bubble Debate, Neo Cloud Signals, What the Markets Missed on Cisco

Welcome to the Cube Pod episode 134. I'm John Furrier with Dave Volante AI salute to you. >> John Frier, what do you say, brother? >> Okay, we had quite the um conversation this week about the AI bubble and uh your post on LinkedIn got everyone Brian Bowman, me, >> Sar Jeet was involved. We were having private texts, a lot of comments. You called us out, we called you out, >> I love it, >> you know, and I love Rob Hope loves it, of course, front page of Silken Angle. He loves the bubble conversation. Um, but, uh, so much going on. So, before we we'll talk about the bubble bursting, there's two sides to that coin. Uh, I'm on the pro nobubble side that it's not going to be bursting anytime soon. But to set the table, >> come on. You guys are just misrepresenting. Go ahead. I can't wait to get into this. The day after the post, he wrote, "Jensen had a pow-wow in New York with CEO of Goldman Sachs, KKR, Blackstone, Brook, um, Brookfield, all the hitters. $500 billion commitment, financing vehicles." >> Yeah. >> Really talking about the AI demand. So there's really kind of two threads here, Dave, this week. I mean, so much other news happening, but I think to me the biggest news was the impact of Jensen. I mean, really, it was a power panel on CNBC. Um, they should have come on the cube after here at the NYSC, but uh they were in New Jersey at their other studio at CNBC, but if they were here, I definitely would have been down there. This is huge. You had you got you got Jensen basically saying, "Hey, we're going to be supporting $500 billion." and everyone was kind of reading into it. A lot of hot takes, but it was really kind of like validation at the same time. Nvidia has been criticized for kind of circular finance, you know, uh since I think maybe last year Bloomberg wrote that story. Uh we know that they're propping up and funding propping up in a good way um the funding for a lot of the NeoClouds and the and the buildout on AI infrastructure. So here was more of a revelation. And I wrote a post in uh April uh featuring Argentum saying hey the bounded function is not not energy it's financial. This actually proves my story was right but also proves the fact that this is now a new financing dynamic. Okay. Because if doesn't happen, the scarcity, the pricing, bubble bubbleicious behavior >> could be >> could be very much a negative, hence your post. Now, I don't you kind of said bubble, but you did say that, you know, in the midterm most likely scenario, it would delay the bubble, but that that you didn't say no bubble. You said bubbles delayed. And I actually liked your analysis actually. I thought it was good. I think it was flawed in one area. But you know, first talk about Jensen, then we'll talk about the flaw in your in your analysis. >> Well, but let me clarify. I'm not saying I'm saying there is a bubble. I mean, I think there's little question that we're in a bubble. I mean, what is a bubble? A bubble happens when the market prices rise exceptionally fast and prices are unusually high and supply is is or demand far outstrips supply and asset prices you know rise exponentially which is what happening what's happening now so there is a bubble I mean the balloon is growing there's absolutely no question in my mind about that the question is will the bubble burst bubbles most typically do burst there are examples that don't burst. I mean, I'll give you one, which is sports franchises. I mean, the Lakers just sold for what, 12 a.5 billion. I've been hearing that, you know, I think the Lakers, you could have bought the Lakers for like, I don't know, pep pitants. I think >> is the NBA bubble gonna burst. >> Steinbr No, it's I don't think it will. I think sports franchises are proving that bubbles don't always burst. I think that um I think Steinbrer bought the Yankees for $7 million. I mean, we could have probably put together a syndicate and bought the Yankees back in the 70s, but uh but but so so that's that's a clarification. I think you and Brian are saying there's no bubble. I don't know how you could say there's no bubble. I mean, with these bubbles, it's bubbleicious. It's beautiful right now. The market is expanding at unprecedented rates. The question is, will it burst? Wait, wait. The question is, will it burst? And what would cause it to burst? What's the likelihood it would would burst? to what's the likelihood we get a we get a soft landing and that's try what I tried to do last week and that leads us to Jensen. So my number one >> hold on hold on hold on hold on hold on hold on >> ahead let's stop there let's ar let's debate the bubble or no bubble because you say there's no bubble right >> Jensen Jensen just basically um addressed what your main the main thing but one of your main things was you're you're a I mean look we've done so many coupons I will say that you are a bull when it comes to AI >> course of course But you were very bearish on the durability of the current AI infrastructure capital cycle. >> Not necessarily. Not necessarily. I just said no. All I said was that the the the the first thing that could happen to burst the bubble. I was just trying to, you know, drive scenarios is that the capital runs out before the end monetization. I I called it productive monetization or productive utilization to use the economic term before that occurs because right now it's it's hyperscalers selling to open AI and and anthropic and so and and c and and neocloud selling to hyperscalers. So if the financing dries up before and end before JPMC monetizes and others you know and buyers that could be problematic and Jensen just addressed that to your point. Well, yeah, he did address it. We'll come back to that in a second. That's kind of like a big [laughter] way. >> Brian and I doing victory labs, but I you you didn't really say the bubble was going to burst, but you did say um delayed reckoning, which kind of implies bursting, but >> Well, no. And I and I did put I mean, I was the classic analyst, like a two-handed lawyer on the one hand, on the other hand, but I I did say at the end of my post last week, I said, you know, I'm a wuss if I don't put a probability on this thing. So I did >> and I stuck my neck out and Bruno Ziza said, "Wow, you're pretty brave sticking your neck out." But I I I felt like it would have been a copout not to put some probabilities on it. So I did. >> So So I guess in that sense, >> the way I read in that sense, I read the post like three times and it was a great post, but it's a conversation which we're having because I think it's important. >> You clearly believe that the AI demand is real and transformative. >> Oh yeah, of course. And you explicitly said strong AI demand exists points to Nvidia. You point to Nvidia, Broadcom, AMD revenue, substantive evidence. You're not I don't think you're arguing it AI hype. It's AI hype and will collapse. But you were kind of bearish on the capital formation and overbuilding. Okay. Which was a supply centric scarcity pricing, you know, enormous commitments. The question is will that fail to translate quickly enough to profitable utilization cash flow? >> Yes. BGO. You're right on. Right on. And and and I'm just trying to build scenarios like what could >> what could make this bubble burst? Like sleep with one eye open. What could happen? >> What are the factors? >> So I asked AI. Is Dave a bear on AI? >> What did it say? What did it say? >> I'm going to read it. Okay. Okay. If you force me to put Dave on a spectrum. Okay. I'm going to read the spectrum. Okay. AI technology. You get full bull marks on that one. Green green dot bull AI demand green dotbull 2627 infrastructure spending actually full green bull constructive you were constructive on that >> okay >> current scarcity economics green bull long-term infrastructure capex sustainability red skeptical >> okay >> yellow skeptical yellow okay that's fair I be I would put that as a Yellow. I think that's fair assessment. >> Yellow. That's a yellow skeptical. >> That's really good. What? Where was I? Red. >> This is just the grades from my AI agent that goes out and looks at So, I'm going to get tell you. Okay, the next one. 2028 2029 capital cycle red bearish. >> Last week I was last week. I was I think that's fair. >> AI in itself after correction. Green still bullish. Um, >> yes, 100%. I'm I am bullish after correction. Yes. >> So the most revealing then then it says this is the next thing in the AI just and I'll move on. >> The most revealing sentence in his analysis is this is this thesis AI doesn't have to fail for the bubble to pop. >> Right. >> So I think you know to be fair to you is very provocative. He definitely took a good approach. Um, I actually liked it because it made me think about things a little bit on one point because I really liked that supply side velocity uh versus this intelligence demand because what I when I read I'm like oh man Dave's all wet on this and I first went not negative but like I I got can't wait to debate him. Of course, Brian [laughter] coming everyone's going public. Um, but in if you're an investment banker, you could read into it as the cash flow piece might be dangerous. But here's what I would say and I've been asking people on the cube since this. The demand curve is there. You address that. You're green and bullish on that. The question is the supply side assume that works itself out. Does the pricing make sure come in line where it gets rained in? But the demand question and Jensen brings this up a lot for intelligence. So intelligence is hard to model up is what does that even mean? It's going to be highly elastic. Inference is a big part of it. Agents generate machine to mean workloads. Robotics, high compute, falling prices will stimulate demand faster than supply will expand. So the question is this. I think you brought up a good good analysis to ask another question from your post that if demand isn't just about the gear and it's about the intelligence it produces >> yes >> we might be at the beginning of the largest consumption phase of AI okay so I think and the way and the way I looked at it was okay let me think about what we've been researching let's just say we're in a 2026 2027 super cycle for scarcity which you pointed You agree? Yeah, >> you agree on that. Okay. HBM packaging, >> all the pricing, I mean, solid dime and if you're in the memory business like Micron, I mean, prices are through. You mentioned NAND on the last podcast. I think it's through the roof. >> So, I think let's just call this next year scarcity super cycle. The question on the enterprise that we've been exploring is when's the money going to hit the table? When's the value from intelligence going to be? So I think that the economics of intelligence is going to go through a discovery phase next year and into 2028. And I think this is where things get interesting because if compute supply expands and entrance costs falls falls in a big way and agent workloads come on board and production hits from these pilots then we will start to see the economic modeling and remember Jensen said on I think it was a podcast or on GDC it's hard to model out the value and I think this is kind of what I'm reading into it. So you know >> yeah he said don't worry about the ROI okay at some point people are going to start worrying about it >> 2028 2030 it's either going to be a capital reckoning or a second AI super cycle so that's a fork so you're you know you point out the supply the monetization all that's right on the money but what if the discovery phase of the value and economics again compute supply is expanding And if inference costs drop and then workloads kick in. >> Yeah, that's the soft landing scenario. >> That's laid out. >> Soft landing. That's a freaking >> soft landing. But so so back up a little bit. >> How is that a soft landing? That's freaking That's a launch. [snorts] >> Well, that's no bubble bursting. So yeah, it's like sports franchises. Yeah, I call it a soft landing. Meaning, you know, the economy keeps cranking. By soft landing, I'm trying to take an analysis from, you know, the Fed with interest rates where, you know, the the you never get inflation. In this case, you never get a bubble bursting. But I want to go back to the bubble because this is where you, Brian, and I I remember at AMD, you're like, Dave, you're out of your mind. Brian's like, "The IPOs." And so, but the the point is that so Floyer and I did a forecast in 2024. We said that the market for silicon will be about a trillion by 2028. Well, this year um the the uh the WSTS is forecasting that the market will be 1.5 trillion 2026. So, it was pulled forward or blew through our trillion dollar forecast. At the time, people thought we were crazy forecasting a trillion dollars. But here's the thing, John. Half of that, more than half of that, 800 billion of that is memory. I and because high bandwidth memory and the interesting thing of my research last week was that and I don't know if you you've read my post so you know this the unit pricing or the unit volume of mic for microns high bandwidth memory increased like 6%. But the sequentially quarter to quarter increased 6%. The average contract values increased like 65%. So this was not a unit driven demand cycle. this was a average priced derriven you know demand cycle. So that is why I was saying there was a bubble and so but but to your point and you made this point last week on the pod, you've got GPU shortages, you've got HPM shortages, you've got a you know energy issues, you've got uh uh advanced packaging like co-was is is still tight even though it's loosening up a little bit. You got obviously financing was a big issue. You got to light up the land power shell. All these things are in short supply and that's governing the bubble bursting and the timing of all that. And then the other the thing I pointed out in my post is there's there's two clocks. There's the IT infrastructure clock, compute, storage, networking that can go pretty fast. You can build that stuff quite quickly and deploy it. The long cycle clock is all that other stuff. It's the data centers, it's the power, it's the getting past regulatory issues. You know, all of that takes decades sometime, a decade plus. So those two different clocks, they do, to your point last week, govern the the timing of that potential bubble bursting. So I was just trying to say, okay, you know, one of those is the the cap the market's appetite for capital or for, you know, will will dry up. Um, but then Jensen to your to this week's pod just just attacked that premise. He's no longer selling technology. He's creating a financial asset class around AI compute which is >> I think I I want to get into that in a second because you just pointed out was interesting because the GPUs are going fast. So there's a supply constraint not just on memory but GPUs too. If that's going to be the demand uh service layer of tokens, you know, tokens per cost per watt, there's going to be not a lot of Vera Rubin around, right? or maybe sort of secondary market. But the bubble I was thinking about and this is why I was so pumped to see Jensen go on CNBC with the round table of the players. I mean, these weren't like junior people. These was the CEO Goldman Sachs. >> Larry Fink was on there. I mean, this is like >> I mean, they're all there. But to me, the bubble that I think is more dangerous um to think about is it's the financing layer between hyperscalers and the speculative AI infrastructure. Okay. And let me explain. The hyperscalers can absorb mistakes because they have huge profitable businesses that can subsidize their AI investments. I mean both all are taking out debt. The danger >> hold on hold on sorry sorry to interrupt. Microsoft is the only one that's c that's guiding cash flow posit positive. The other two are not and nor is Oracle. So just that's an interesting little sideline. >> AWS is pretty profitable. >> No AWS is cash flow negative. >> Well on the numbers but that's Amazon. They reinvest. You know how that works. I'm just saying it's a interesting signal. This is We got to watch these signals, John. >> Okay, I'll rephrase. Hypers scales can absorb their mistakes because they have massive amounts of cash putting to work on the C. >> Yeah. To use Frank Slutman's terminology. They could turn on the ATM machine anytime they want. >> Yeah. They I wouldn't worry about them, but it's the speculative AI infrastructure, which as you were you were pointing out, the dangerous part of the hyperscalers relationship to the uh speculative AI infrastructure. They're all the number one customers. Okay. So, look at the dangerous part of the ecosystem. It is what we've been covering. Neoclouds, uh, private credit, projected financing, data centers that have 10-year real estate, three-year contracts, GPU backed back stops financing, power commitments, long-term capacity contract, highly leveraged infrastructure vehicles. That is a very that's a tinder tinder box. one match could you don't know what's going to happen. So why I like the Jensen thing with this week is he's put that to bed. Okay, he's basically saying, "Hey, we need to have credit systems, finance, data centers for the growth, for the for the buildout. Of course, he's supplying it." The dangerous ecosystem piece Dave is that we covering this ecosystem the neoclouds the credit and everyone that's leaning in and doing CUDA working with uh all the work that Nvidia is doing they're they have an ecosystem look at open AAI and anthropic major ecosystem investment if those players in the ecosystem are you know impacted >> that's a bubbleicious like impact so the Jensen financing of half a billion dollars with all the all these big banks is essentially showing commitment that okay we need to raise and support all this equity and financing debt equity financing vehicles you Oracle you pointed out on your post okay so they're putting a lot of money in >> yeah with Stargate but but I think it's it's worthwhile to before we get into the Neocloud because I think that's a really important piece to understand because essentially Jensen is funding the NeoClouds of course you have Tensor Wave which is AMD's Neocloud. But to understand what Nvidia is doing, you got to think like a banker, right? So the structure is like these guys, they they finance, you know, power plants and, you know, fleets of airplanes and, you know, these big infrastructure projects. So what's happening is okay so these six investors that you just mentioned they're going to provide debt and equity to to dedicated you know SPVS special purpose vehicles. You hear that thrown around a lot. And then what are they going to do with that? They're going to use that capital to buy or or maybe lease Nvidia infrastructure. And then they're going to get the site. They're going to get the power. They're going to build the AI factories. all that physical infrastructure but that's only part of the equation. So the assets you got Nvidia you know systems AI factories then you have a customer contract you got the land power and shell but then you got you got what what I called you know productive utilization the monetization profile and then very importantly you got the residual value of the equipment that's what everybody's talking about after the first contract ends and when we talk about the neoclouds we're going to see they're still selling you know A100s and and ampers this is six years in. But but but the the point is the the lenders, they don't even care about the AI hype. They don't care about the bubble. They want to know who's going to pay, how long is the commitment, is it a take or pay contract, which means take or pay means you either got to take the product or if you don't take the product, you got to pay basically a penalty, you know, pay a minimum amount. And can the customer cancel or can they delay? And that's that's what the lender cares about. And so my point is the the whole thing is shifted to to your point. It's no longer about is the money there to fund this stuff. It's all that other downstream uh those downstream monetization issues which ends with residual values. Well, I'll tell you one other thing. When I was at IDC, I I I inherited the what was called the leasing planning service. We had a multi-million dollar business that just tracked mainframe residual values. And what happened was in 1986, the Congress killed the investment tax credit which was propping up mainframe leasing. You had all these leasing companies, guys making tons of money, buying boats, big cars, wearing Rolexes. The investment tax credit killed when the when the Congress killed it, the freaking mainframe leasing business went in the tank and the whole, you know, aftermarket died. And then that just coincided with microprocessor based computing and was like a double whammy just on main frames. But but you know it's hard to be I'm not saying quantum computing is going to do that but hey you never know. My my point is >> not even on the radar yet. I mean Jensen >> of course of course but this is a much more sophisticated equation now because of all those downstream effects. But but what you just but what you just said is right on at the top. Jensen attacked that one weakness of my, you know, my thesis last week and it to me pushes the potential for the bubble bursting. It derisks that, but not completely. It just pushes it downstream. >> Well, the the I looked at this like I first of all love very provocative post and it's really going to get a lot of credit for making people think because it was good. To me, I looked at what makes a bubble collapse. So, you brought up memory pricing, compute pricing, utilization, and financing. The sequence of a bubble would be okay huge memory pricing uh compute pricing is still high utilization is low financing is there and if the utilization and financing roll over simultaneously meaning they don't happen. So if utilization doesn't hit and the financing is not delivering the cash flow that's a bubble break. That makes total sense. My argument would be the supply velocity that you pointed out is today's next 12 months issue. Nvidia is going to say we're going to put half a billion dollars with all these banks. We'll try to make as fast as we can. Engineers will build new alternatives, compute, it's going to be a bigger role with with pre-fill and decode. But if you look at supply velocity versus intelligence demand, the elasticity of that, that's key because let me give you a scenario. What if the intelligence demand for intelligence tokens produce intelligence? That's the goal of AI. You can inject intelligence into finance, accounting, supply chain, any partwhere. Yeah, >> that's what people want. Again, we can't model this out because it's still in discovery mode in my mind. So, if intelligence demand grows faster as prices collapse, it looks more like AI version of cloud computing, meaning repeated over capacity followed by new applications that consume everything that can be built. So the reason why I'm b bearish bullish on this is that coreweave's earnings this week point to the backlog. So it is the supply and demand. Yes. Is utilization in demand? We see it. Is it fully understood? That's the debate. I think it's going to come in. If that utilization of intelligence happens, financing works. If it doesn't, then we got a problem. So >> that's a big to me a big thing is intelligence the utility and with Jensen's kind of move you're looking at an asset class here. So >> I'm curious to see how the bankers do this because if it becomes an asset class that's completely different animal. >> Well so I want I want to pick up a couple things. So I've thought you you you just made me think of something that I've been thinking about which is the cloud. I I never felt like the cloud was a bubble. I felt like the cloud was more of a share shift. The cloud was just better it than onrem at lower lower cost, not lower cost anymore. >> Uh with, you know, capex to opex and it was sort of a sharehand and shift if you will. And I never felt like it was an asset bubble. And so I let me interject real quick because remember go back to when we were having conversations on the cube when we first started doing the business the cube there was a point in time where Amazon didn't cross over to the enterprise okay you had all the startups on there we used to call it the junkyard dog right you build your own and we used to have conversations with the oracles of the world like why would anyone want to go to the cloud and we were we were bullish on the cloud but there was a moment where it didn't cross over >> but once it crossed over I would agree share shift, but it never felt bubbleish. It was just revenue. SAS apps >> uh and revenue kicked in. So, I like that analogy. I think that's what I was trying to get at. So, continue. >> Yeah. But the difference is that the difference is well, actually, it's interesting. I mean, the difference is that Amazon Web Services hid under the losses of Amazon and Bezos was raising all this money and and funding Amazon. But it was never that massive capital buildup build in 2013 2012 20 2011 2012 2013 I remember vividly this conversation you and I had many of these on the cube there was a point where it was just startups because the alternative was get a data center buy a box from super micro put it into a cage and then you don't know but you know companies like Airbnb Dropbox they put their credit card down they build it Twitter was built on a all the web 2.0 0 SAS were built on on Amazon. what really crossed over and that was not obvious and people were squinting saying hm is that the real deal once it became a reality was good and remember Adrien Cochroft okay said on the cube why why did you move Netflix the first real company to move to AWS that was the big win CIA came right after that with Terresa Carlson and public sector but Netflix I asked him Adrian what was the big decision Remember remember remember his answer he said >> he said it's easier to teach a developer to run operations IT operations than taking an IT operations person and teaching them how to code that was devops that was a big moment that crossed over and two it followed everyone else came on and then you saw the CIA and the public sector once Amazon nailed security and made it reliable it grew but there was questions I feel like AI I and your post was similar. Now, what if this doesn't come home? What if this pony doesn't come in? >> Yeah, but the difference is it's not wasn't that was infrastructure is code. This is a huge capital buildup, right? And infrastructure buildout. I mean, this is I I I there are similarities. I'm not saying there aren't, but it's just this is of such a much larger scale than the cloud. I feel like the cloud was a ratchet game almost like Amazon. They tested the market. They got product market fit. They didn't have it at first. They won the CIA deal. they got security, they sort of ratcheted their way into the enterprise and then obviously it took off, but it wasn't like, hey, we're going to drop like a trillion dollars on capex and see what happens. And and and so there in I think lies the difference. I want to actually mention two things if I can take a a little side trip here because I think the media has gotten a couple things wrong. I was listening to Fast Money last night and they were like, "Backs stop, boy, that's not a really positive sign. That sounds really negative." So, everybody's talking about the Nvidia backs stop. I want to clarify what that is. Nvidia has the option to backs stop up to 25% of that like 125 billion of that of that half a trillion, they don't have to do it. If if the if the banker feels like that that the thing is too risky, then Nvidia can come in, but it's at its option. And if any of us says, "Nah, this is too risky or we don't see the monetization or we don't like the residual value profile and the assumptions, then we're going to tap out. We're not going to backs stop it." And then the the deal dies. So that's one thing that I think they got wrong. The other thing is I'm hearing a lot about like mortgage back securities. This is not MBS. It's not securization. It could be some days, but you're not seeing pools of loans being, you remember the big short being divided up into tanches and then raided, you know, double A, AAA, B, Crap, and then sold as diversified in a secondary market. You're not seeing that. Uh they what they've announced is financing platform and dedicated pools of of institutional capital. And by the way, we should say, oh, this is speculative because the deal it's no deals have been signed yet. But still, I wanted to get that out there, John, because I think the media is all this frenzy about these are mortgage back securities is is absolutely off base. >> Yeah, that's totally wrong. And there's no layer to anything other than just giving confidence that there's a capital markets and there's an asset class that people will recognize and and and you made a good point about the cloud, right? The cloud didn't have the same problem. Well, I was comparing more of the the situation of unknowns on demand and no one could predict Amazon's demand for SAS was going to be an IT share share shift as you pointed out because no one ever built an app SAS app that said look how great this is and everybody want it unlike AI everyone sees AI and they go I get this I'm using it there's demand for it it's a user experience shift so I think it's a significant uh shift in the marketplace for user behavior and business value that's easily to identify. So I think the demand that we're seeing for AI and and intelligence is clear. Hence the buildout. The question is can that actually be leveraged into monetizable? It might be monetizable directly every company but if you believe that intelligence can be injected into systems and companies process then you got to believe that that's got to be thought through. No one knows what the demand curve on that looks like. We know people want it because they're doing it. I'm using AI all the time now getting, you know, agents going. So, you know, there's a lot of value in AI. I mean, everyone's using it. So, it kind of came before the scale, right? So, like Amazon just had scale and it's like, hey, why don't we just run our it? And that was risky. That was a slow ramp. And then it became obvious like wow I can save money and shift to the cloud and not have to have all these data centers that have energy costs and all this stuff. So >> yeah and and and Microsoft financed it with its software estate. Google financed their buildout with with search you know Oracle is financing look what's happening with Oracle but it's you know it's OCI originally was financed through its cash flow. So all that that is to me the big difference from cloud but but to your point about neoclouds we have some evidence that this is working because and it's happening it's not mortgage back securities but you're seeing some secured financing that before weeave has has financed AI infrastructure through through syndicated loans uh they've got some some customer prepays um they've got short duration contracts you know Nebius has a like 700 800 million dollar facility that they secured by because their GPUs were deployed. So the point being that you know Jensen's right this is an asset class this is monetizable today. Now whether he he emphasizes how it's fungeible right which is what what that means is there will be residual values you'll find today's training workload becomes tomorrow's inference workload and then and then the other big question John is will will software efficiency help improve performance per watt in other words you know let's say there's an older GPU everybody's saying oh these these these GPUs Michael Bur they're their short-term assets. Will will software updates allow them to extend the life cycle and be and get more utility out of these things? If CUDA becomes more efficient, you can apply that, you know, to the N minus one or N minus 2 or N minus 3. So the the interesting thing I just want to bring up one more point you know Ben Thompson is amazing. He his latest piece in strateery or whatever he calls it. [snorts] One of the things he said was at some point and this was he wrote this before the Jensen thing. At some point though if if if financing becomes tight again for whatever reason people are going to care more about the initial check that they write versus the total cost of ownership. My point the point being this is the be bare case for Nvidia and I'm a huge Nvidia bull but Nvidia stresses performance per watt and I have always believe they are going to continue to have the best cost when you look at because of their volume you look at performance per watt which is what matters but if you can't afford to write the original check who you going to call you got to call AMD you know you're going to call the alternatives and so that that again I think Jensen you know, took took that off the table with this announcement. >> I think that's you're going to see a lot of software innovation and hardware innovations, but the demand's too high. Edge is coming. I think you're going to have big AI factories. You're going to have demand for, I should say, on, you know, ontake or vertically integrated. We talked about that on the last pod, vertical integration versus, you know, being the intelligence refinery, if you will. So, I I think there's going to be a lot more action and I think it'll normalize. I think there'll be a power law like we talked about the models but I think you know the data growth the quality that's coming out of the AI right now is amazing I mean other data points just this week um data bricks okay okay growing 80% year-over-year surpassing $7 billion run rate okay raising huge another huge round of financing $5 billion in funding five billion >> at 190 billion ion dollar valuation. Okay. So again, they're are going to be a big supplier in the AI intelligence race on the public side for companies. So you know you and I you did a great analysis on data. You and George Gilbert wrote a great posts many series of posts on you know why you like lakehouse and lakebase and ging unity AI gateway these these the performance is incredible. Okay, so they made by we predicted that their open table move was a burn the boats moment a couple years ago. We kind of predicted >> this. >> Well, the iceberg acquisition numbers like we're like this happened. >> The the tabular acquisition was a great chess move. I gotta say, I mean, Ali Goatsy, I mean, data bricks, what they laid out at the data bicks data and AI event, they they in some ways somebody posited that they could be worth more than Anthropic. They're they got a better way better software stack. I mean, obviously Anthropic has the AGI and the LLM and the Frontier that that Data Bricks doesn't have, but in a lot of respects, I think what Data Bricks has is potentially more valuable. They've got the system of intelligence. They've got the user surface. They've got the business user. They've got the ontology. I mean, they've they've got it. They've got it in their vision and in their roadmap. And they ultimately, you know, usually deliver. But are we done with with Jensen's deal? Because we haven't really talked about the the Neoclouds in in depth. And I just wanted to cover their I mean, Cororeweave and Nebius both >> Yeah, definitely announced. >> Let's get into it. >> I mean, Corweave announced 2.6 6 billion in quarterly revenue was up over 100 was up 112% and they had 104 billion of backlog. I mean that and that didn't include what the Michael and Trader said on the call 25 billion of additional customer commitments that were signed just after the quarter ended. So, I mean to your point about demand, I mean it's like and near near-term capacity is like completely sold out and it's unbelievable and and half their backlog is already attached to contracts where we've started to deliver and and they expect like three three quarter twothirds of that by year end and so >> I mean I a lot of people were down on core many years ago. Oh, they're taking on all that debt. Again, Nvidia obviously helped a little bit there, but their vision of building a vertically integrated system is phenomenal because we again we we compared the two approaches last week, but I think they're positioned perfectly to be the next hyperscaler for what they're doing. They have a lot of ex Googlers and a lot of smart people in cloud and over there. Hen Goldberg's running engineering. They're targeting the enterprise. They're targeting these big needs from the hyperscalers and big buildouts. But one little tell sign that might give you some some uh some some faith in the enterprise side of their business, which I think isn't even in the numbers, that's just on the AI infrastructure side, is that um CNBC broke a story today that OpenAI CFO told investors that the enterprise business is now generating more revenue than chat GPTled consumer business. Enterprise business on on on enterprise customers on OpenAI grew 32% in July. Okay. Now, there's a lot of token maxing, but that's a direct momentum point to Anthropic, which is clearly being recognized as the leader in the enterprise. And you know, they're on they're on track to do 40 billion plus right now, doubling its run rate from the end of last year. They're already doubled. >> Yeah. And Anthropic saying it's going to be 100 100 billion, but I think that it's not apples to apples. I mean, clearly Anthropics got, you know, more ARR than OpenAI, which I didn't expect, but I think Anthropics also like doing a little double counting like including some of the Amazon end revenue in there. So, we'll see when they when they actually go public. Nebius um announced I think a billion dollar deals for like four customers. And most of that or half of that was was was paid by, you know, upfront by customers. So it's like amazing the support that that these companies are getting that and they've talked about how A100s and hoppers are still selling. But there are John some cautions in there. I mean what we these are all being I mean there's negative cash flow, there's losses. What we want to see and this is going to be whether or not the bubble bursts is we want to see the full life cycle where in companies like Cororeweave and Nebius and Crusoe are able to pay for these cycles through their own cash flow. So they're really the evidence that we have today is at the front end of the curve meaning you're seeing um customer prepayments, you're seeing you know the debt financing um and that's converting but what we want to see is that full life cycle from you know that you get these capital events, you get you know the customer is monetizing, you're getting productive usage, you're getting the end of life of the contract and then you get you know the residual value with a fungeible asset. that and then that and and and then after the depreciation cycle, they're still monetizing and that pays in funds for the next cycle. If they have to keep raising debt and keep giving away equity, um that's problematic because they don't want to do that. And so that's what we have to watch and you you're not going to know for a couple years. And that's why I put 2029 2030 as that sort of new window as when this thing could pop or when it turns into sports franchises. >> Well, we'll see. You know, reduce, you know, slow that slow that B roll. I think that the that's going to come into the I think 27 28 window. You're going to have visibility on the unit economics of where intelligence goes. And I think to me all the data points to to to revenue in my opinion. So, I mean, we'll see. I mean, it's a great conversation. Again, you know, people love to talk about bubbles. That's like saying, you know, it's like talking about sports when you have your favorite team, Red Sox versus the Yankees, you know, like people love to riff on what the future's going to be. And we'll see. I mean, we'll see. >> Okay. Wait. Are you saying we're not in a bubble or are you saying you use the term bubbleicious? Are we in a bubble or not? John Furrier. So, >> bubbleicious is not saying there's a bubble. Bubbleicious means it's frothy in market. >> Are you saying we're not in a bubble? >> No, we're not in a bubble. We're in a demand curve that's underserved. Hence the backlog at Coreweave. The intelligence is not penned out yet. People are betting. So that I look at as educated bets. I interviewed the CFO today from Lumen um who had their earnings. They bought Alkira a company that was on our super cloud event. If you remember them, he said absolutely no bubble. They're going to they're putting more fiber down than ever before. They have conduit. They're turning it on. There is massive demand for what they're doing. They reset their financial capital structure. They're growing. They're kicking ass. Why? Because they're in networking. Okay, we haven't even gotten into what comes after inference in terms of growth cycle. Right now, it's HBM semiconductors of the darling. Well, my prediction here is not not only we not in a bubble, that you're going to see other sectors light up like freaking rocket ships. Networking. Look at networking in 2027 20. I I predict that's going to be the one of the hottest sectors. KVash set the table a few years ago that's going to move forward. Everyone sees the value in intelligent networking, not your conventional network. So, we know it's it's other stuff. >> So, we're not in a bubble, right? You're saying we're not in a bubble. >> No bubble. No bubble. >> So, what's the definition of a of a bubble? I just asked I just asked. >> Bubble is a bubble. Okay. >> What's a bubble? How would you define bubble? And I'll tell you what AI says. Well, I'll tell you what AI says. >> The outcome of utilization of a utility. >> Wait, say it again. What's your What you cut out? You cut out. You cut out. What's your definition of a bubble? >> My personal definition, simply put, is if you're investing money with expectations and they could be over inflated about how the future will look. It's an unknown future investment. If the cost to deliver that don't drop and the utility of that value doesn't go up faster than the cost drop, then the financing collapses. And if the utility and the finances roll over at the same time, meaning we're expecting my payout, it's not going to work. A great example, unlike this example, >> that's your definition of a bubble. That's that's your that's not a definition of a bubble. That's a that's an indication of when the bubble pops. A definition of an asset bubble happens when the market price of things like houses, stocks, or gold goes up much higher than their real or true value. This happens because people buy them out of excitement and hope to sell them later for more money rather than for what the asset is actually worth today. And I would say that defi that's a classic definition of what's happening now. HBM is artificially high because you can't get it. you people aren't making and customers aren't making money. >> That's not a bubble. That's e it's called economics. This curves you learned that in high school. So what I would say a bubble but I would what I would say a bubble is is this a tulip. The tulip craze. Tulip was worth the same it was the day it was priced X and then it went up high. Everyone knows that story. >> That was definitely a bubble for sure. >> Tulip craze. That's an extreme example of what a bubble is. I was getting more specific around things like the fiber buildout during the internet phase. That was a bubble. The disc drive of the 80s, you you and I chatted about that the other day off camerara. That was a bubble. The the there wasn't enough demand for hard drives, but everyone thought we need hard drives. The PC revolution. So that was over inflated. The value of the hard drives, the supply was available. The supply exceeded demand and there was no utility of value. The prices dropped like a rock bubble pops internet. The idea that we're going to make so much money from people buying online and web commerce just didn't hit the demand curve because the online population of the internet wasn't matching the economics of what they thought it would be. That's a bubble. That's just miscalculation. That's kind of like a reality version of bubble in the tech world. Now, there's examples. I laid one out. Fiber being laid down. Let's build dig trenches and lay fiber onto dark fiber. There's no demand for it unlike today. There's massive demand and the backlog on coreweave is is is a sign that that's demand that could be economically driven by demand and supply. That's why I'm focused on the utilization because if it doesn't come home, they've overpaid because the prices are high on supply scarcity which you pointed out in your post. But that'll normalize that'll that'll figure itself out by you know better comput architecture. >> All you're saying is the bubble won't pop. But you're not saying that doesn't mean we're not in a bubble. The definition of a bubble returns >> I think the returns on the capex that people are freaking out about the billions of dollars to spend for say a nebas or argentum or core weave these companies have to be worth more than what they paid and >> but they aren't today. You a you you acknowledge they're not today. In other words earnings. >> No, we're not we're not nearly seeing the revenue match the the the capex buildout AWS negative cash flow. Oracle negative cash flow future revenue and that's >> negative cash flow, right? But so but yet asset values are climbing very very quickly and they're at unprecedented rates. People have FOMO. Everybody's, you know, diving in to the pool. >> You're going to build these data centers. They're not cheap. >> So that that to me is the the classic definition of a bubble. The bubble is expanding. >> Whether or not it bursts is a different question. >> Okay. Well, we have to continue this debate next time because we can go in and salami 10 ways from Sunday. But, you know, I stand by my position. >> Well, >> you're anti you're probubble. I'm you know, >> well, I'm just saying the facts support that we're in a bubble that we're decoupling from from actual revenue. >> I think there will be a massive renaissance of economics. >> Oh, I agree. Bankers see cash flow and real value across all businesses because every business as you said trillions of dollars of value spread across every single sector will be serviced by this new kind of utility like electricity that will be rolled out like a refinery like oil and you'll have retail you have vertical integration and I think that revenue that's spread across all society and business have to be served by intelligent brokers. >> I I I agree with you. I I think that ultimately you're right >> and the bankers will see great returns. >> Oh, I think the bankers are going to make money off this deal. I But we'll see what happens with the bubble bursting. I've laid out my scenarios. I I I I we we've we beat that pretty well. I want to say something about Cisco. I think the market has Cisco totally wrong. I thought Cisco had an awesome quarter, by the way. They missed the gross margin target by like I don't know like the freaking hundth of a basis point. I mean, slim margin. They grew the company like 18. This is Cisco. They're growing the company topline by 18%. Their product revenue grew 35%. That is an awesome quarter for a company that is a huge company. They're like right in the middle of the AI mix. If to your point about bubble, if we're in a bubble, Cisco should have gone up because that was a great quarter. >> I think Cisco is a great buy. If they're plummeting, they're going to bounce back. They're too >> The stock got crushed. It's down 9% down 11% in the past five days. I mean Cisco >> I would be quarter they had >> if I was a stock adviser which I'm not so I'm not don't take my advice I'd be all over Cisco. Here's why. If you look at Lumen again I mentioned the CFO is in here today. I forgot about their earnings today. Um but Cisco has conventional networks. There are so many advantages to lower level technology that they've already perfected. And if you look at um what's happening with Nvidia KV cache again four years ago we saw that we saw what infiniband did we saw what they did with KV cache they created networking protocol around the GPU to bring all the resource together that's because it's highly dense in the factory if you project out what the computing architecture will look like with distributed computing more nodes are going to be on the network AI nodes that means you have to connect the factories together and Cisco is and Lumen and companies that are in networking are perfectly positioned to take their conventional networks, their pipes, their dumb pipes, not only put intelligence in there, but have intelligence run over them and connect factories. That's going to be the next wave, right? You're going to start to see a lot more things >> uh networking. So Cisco has the exal base of conventional networks which by the way even in the stack of AI they have lowlevel you know physical layer technology that's already perfected >> their optics their bet on silicon silicon one I mean I I I mean I love Dell Dell's 300 plus billion dollar market cap company but they're three4s the market the valuation of Cisco Cisco's got I think you know far superior financial model and Um I mean they're going to put intelligence I would watch Cisco very carefully and I would look at them through the lens of if they take those conventional networks that are already connecting resources and intelligent resources going to sit on top of it and as you pointed out in your research and with the cube research and silicon angles coverage of the systems of intelligence a new layer on the stack I'm going to say it right here on the cube a new layer in the stack is emerging the systems of adaptability Okay, because your systems of intelligence thesis is so right on that all this talk about harmonization layers, semantic layers, har um control plane, resource management, that is in essence a systems layer for resource management and scheduling. That's an operation. >> What did you call it? Systems of what? >> Systems of adaptability. >> What is that? >> That means like under the covers of intelligence, you look at FPGA for instance. I was talking to Alira this week um team there. the other pure play FB FPJ company. You got um AMD has >> XYlink F that's perfect for smaller clusters where computes involve maybe a little bit of GPU. They could use the FPA it's programmable. So you're going to have this programming layer managing underneath the systems of intelligence that's going to sit above the AI factory infrastructure and also maybe even be primary edge nodes. So you start you're going to start to see this new adaptability where I got to react to certain situations. Workloads could be deterministic but for ones that aren't there's going to be intelligence around adaptability and I think that's an unknown area and I've been watching it very closely. I just coined that term because your systems of intelligence assumes and I think it's the right analysis that that's going to enable agency value execution capabilities through headless systems AI agents but under the covers it's like cloud technology meets data right there shit's going on under the covers right so I think there's going to be a whole another intelligence now I think that's hiding in plain sight because Nvidia and others are doing it at rack scale they're already doing it it's already adaptive So I think when you start to get into disagregated infrastructure and disagregated resources that are connected by networks, you're going to have to perform stuff at the edge. Look at robotics, look at manufacturing, look at telecom. They're all going to have intelligence injected into them and they're going to have to respond. So I think I think there's going to be another layer of, you know, resiliency uh that's going to be needed. So I think we're we're starting to see signs that okay, assume you have a 100 AI factories connected of different sizes. What does that look like? Well, you got to network them together. >> Yeah. And it becomes the it becomes the digital twin of an enterprise that we talked about in real time. Real time streaming becomes, you know, really, really important. >> Um, but yeah, I think you're right. System of intelligence is you've got without a system of intelligence, you're not going to ever trust an agent to take action. And and is going to do all this under the fly too. Remember, look at all the stuff we're talking about with agents. Oh, I can just write the SQL queries. Am I going to do this over here? I got an agent going to write code on the fly. So, if this things need to be managed situationally in any kind of dynamic workload >> or relationship on on resource, it's essentially operating system layer of the intelligence system. So again we used to talk about this as a semantic layer on the on the basic database data layer but everyone's talking about control plane again >> ontology I mean Alex carpets you know everybody wants ontology of course we've been talking about ontology for five years more back to big data >> ontologies have the graph look let's take um any scale for instance ncale they'd bought any scale and I was talking to um the founder um on any scale and think about like prefill and decode right just the concept steps that are going on today. It's a little technical term, but you know, you know what's going on. We at AMD, they talk about all the time and so does Nvidia. You separate the prefill and decode in the cluster. That means a compute engine is going to do the prefill and then decode handled by another resource. Well, if you're going to have workloads and many of them working across these systems, you got to track it. You can't, okay, I got to talk to the compute node. I got to get that in. So you got to bring all that together, manage the resource, do the scheduling. So you have to have a mechanism up and down the stack to manage that. That to me is the adaptability concept. And you got to have software to do it. So whole another ball game, Dave. Whole another level. It's next level distributed computing in my opinion. >> Yeah. And and I think actually I think a lot of the general purpose function today that is being managed by you know x86s is going to get absorbed into what you just described a lot of the functions of all that management that control plane and and and that and that's going to be drive a lot of demand, right? And so you're already kind of seeing it with CPUs. Everybody's CPU crazy these days. Um yeah, what else is going on? What else are we talking about? >> I It's Friday night. It's getting late. DJ Snake is playing with Zed in New York. Little music concert tonight. >> Oh, really? Zed. We saw Zed one time at at Amazon. Really? >> Yeah. And DJ Snake. I got a selfie with DJ Snake in in the Amazon, you know, luxury box when I was hanging out with the execs there. >> Oh, that's DJ Snake stand there and I'm like I was sitting next just standing next to him having a beer and all these people coming over doing stuff. I'm like, "Hey, what who are you? Are you a developer?" He goes, "No, I'm the DJ." Like, oh. I'm like text who's DJ Snake to my kids. They're like think no way that's DJ Snake. So I took a selfie with him. So I got a selfie with DJ Snake from like what >> and Zed I didn't know who Zed was but he was now he's all bald. So I'm going to go check him out. >> That's good to see you. 134 is in the books. Great debate. This is not over. >> That was fun. I I I'm publishing this weekend. I'm gonna I'm gonna keep pounding my my bubble scenario and up I have to update it. I mean, Jensen just basically he must have read my post and said, "Ah, watch this, Balante, spun up his uh his his uh his finance seers." But wow, what a move. What a chest. >> He timed it perfectly. We better get it out there before Dave's post gets traction. So, [laughter] >> he knew it's going to go supernova. >> Awesome announcement, though. >> All right, Dave. We'll see you later. See you next time. All right. Thanks, everybody. Say bye.