135. The AI High-Wire Act: Trillions in Spending, Rising Backlash and the Race to Prove the ROI

Hello, welcome to the Cube Pod episode 135. I'm John Furrier with Dave Volante for our weekly podcast. Dave's in the salute position already. You're on vacation. You're fast. >> You do, man. No, I'm on a vacation. I took a day off. You >> happy birthday. >> Thank you. >> I saw you with the pitchfork digging clams. >> First time ever, John. I mean, I used to go clamming when I was a little kid. You know, you just do it with your hands. But we uh we got a lot of clams yesterday. Now I got to figure out how to eat them. >> I'm jealous. I'm like, "Are those cohogs, little necks?" I'm like, "Mam chowder be great." >> I got probably two dozen little necks. I got a bunch of fatties and then you know cherry stones and I got a probably a dozen cohogs. So >> the uh the it looked like a fun time and you got tools. Great. You can go into the into the water chat in the cape. Great place to get clams. >> Hard work. >> It's really hard work. >> Yeah. I mean >> the workout >> I was doing in all day sitting in a chair. So I wish I was digging for clams in a beautiful day at the beach. Happy birthday. That's all right. Well, let let's get into it. >> Today though, pillar breaking analysis, bunch of meetings. A meet is and I would did a deep dive on sovereign on the financial pillar. That's going to be good. >> Yeah, we had I've had this week probably a dozen sovereign conversations. It's interesting. Sovereign, sovereignty, sovereign cloud, sovereign AI. Dave, it's a mashup because no one knows who owns it. every everyone in the organization is fighting for it. I own sovereign. But if you if you're in the cloud group, you own sovereignty. But if you're in the AI team, you own sovereign there, too. If you're in the enterprise, you own sovereign because the data is about sovereignty. So, you know, sovereignty is coming up in all areas because it's a platform issue. And I think that's a big theme. I want to drill down with you on that because the the massive infrastructure spending we're seeing continues to to go great. And one of the key things is sovereignty because the it's not like a marketing thing. It's really technical architecture. It's a lot to do with the things we've been covering in the weeds on the cloud native side. Old school data center, application performance management ties in what's the boundaries? What's the governance? How do agents run? Because AI is driving uh revenue. So the AI side of it, if you're a country like France, you want to have that AI in the country. One, privacy, but two, the revenue, and two, Mestral, for instance, is a model in France. Maybe you want to prioritize that one. I mean all kinds of corner cases are coming up with the infrastructure that's enabling. So I mean let's just start there since you did a breaking analysis and we're all over sovereignty and we're getting more data from the suppliers, the entrepreneurs. What is your take? Because I'm seeing it being an awesome opportunity almost like an AI factory moment where you get the definition just start doing it and you can get data but it's not about who owns it. It's going to be it's a horizontal thing, right? But it's it's it's each use case is dependent upon who's running the environment. What's the software stack? So it sounds complicated but actually isn't at least from my perspective. I'd love to get your thoughts on one the five pillars that are coming out of the research team on Silicon Angle and the cube. What's what's what's your analysis tell you? >> Yeah. So we did a deep dive on the most important pillar which is the financial pillar. That's the fifth pillar. And we kind of pivoted off of a story that the information broke earlier this month about Canva. I don't know. Did you see that? >> No, I did not. >> Here's a company. They're like they're about a billion dollars a quarter. >> Well, I know Canva. I know the company. I didn't >> So So just for the audience, you got you got a company that's almost a dollars a quarter and they had to cut their revenue growth forecast from 30% which is obviously very good to 20% which is still good. But the reason was that their AI costs were way more than expected. So Canva has this premium model and it used to be cost them peanuts to serve the free users but now it doesn't anymore because they were routing all their queries to a frontier model. So you know the fix was it wasn't like negotiate a better price with OpenAI or Anthropic. they had to rebuild their stack and um you they used their Leonado AI acquisition and which gives them you know you know routing LLM routing and they cut the cost of AI by like 90%. And so that slowed them down. They had to do the pivot. And so the point was that like enterprises, yeah, they're all talking about do buying tokens and token maxing and everything else, but of course, as you know, John, they want outcomes. They don't really give a about >> tokens. You're not like don't wake up in the morning go, oh, maybe you need tokens, but you want a result. You want more revenue or you want, you know, less tickets or and so that's what we dug into pretty deeply. We called it from we called it from token maxing to sovereign alpha. who controls your AI economics and it was a good deep dive. >> Yeah, I mean that's come up a lot. It's interesting. Ironically, you made me think about this because the token maxing or the token utilization, the cost of tokens when people are building AI into applications, it's actually calling essentially production systems like the frontier models which cost money andor uh neoclouds hosting stuff or other like production grade high cost infrastructure. And what's interesting is if you remember the server days, Dave, back in the data center preverirtualization, um, and this is this there's a point to this because VMware is having their big event next week in about a week and a half. It's actually the 31st it starts where if you look at virtualization, without virtualization, there'd be no AWS because AWS was built on Zen that created server virtualization. But remember, server virtualization took the server and virtualized it. So you had more servers that reduced the cost for buying another server. So there's an interesting trend of foot called virtualizing AI layers. Meaning stuff that's either premium like in Canva or a test bed or test environment. People are actually building AI into everything so fast that what's out of control is the fact that they're using, you know, production grade AI, aka stuff that cost million dollars a day. In some cases, some companies were spending a million dollars a day on test testdev. So what's happening is there's a trend to virtualize the AI layer and send all that remedial first wave testing andor premium AI to an AI tier and then move it to the production environment. So it's like running testdev in production. you know, if you were building payment rails, for example, on an enterprise, that's like testing your code on the production system, right? That no one does that in in a traditional software sense. So, this is an emerging opportunity. And then concurrent to that, the Linux Foundation this week launched the tokconomics foundation. >> Yeah. >> So, the FinOps group. So, what's starting to happen is this real momentum around cost, control, and governance. This is sovereignty. And that's why I brought up the enterprise because I was doing an interview. I forget who it was. I think it might have been NYSC or here. I can't remember. It's so many interviews this past few weeks. I brought up sovereignty and the expert that I was interviewing said, "Yeah, great point. Actually, the enterprise." So, he correlated the word sovereignty to a talk track around the enterprise. So, I'm I said, "Okay, what do you mean sovereignty for the enterprise?" He goes, "Well, they're a sovereign company." So sovereignty is now broadening to what was once a geopolitical boundary and then moved to network boundary because hyperscalers like I was have regions in Europe and serve France for maybe another region or whatnot to now company boundaries. So it's interesting the definition of sovereignty is moving into the enterprise because from their standpoint think about it if you're an enterprise you've got data it's important to you it's a competitive advantage they want to have local models to do that so they need things like route this to there that's why we're seeing AI factories being installed in on premises close to the dev teams and have unlimited tokens and starting to see that with Nvidia so it brings up a question sovereignty is not what it was a year and a half ago or even two years ago or maybe even six months ago even a year ago was about geography. Now it's about enterprise network boundary country boundary. So this tells me that sovereignty is an infrastructure reference architecture challenge because if you were going to start your network and you had certain boundaries and conditions, you'd want to control it. >> Yeah. >> And if you can control that, then you control the AI. So governance controls the the AI, data controls the AI. So you're starting to see a new layer in the control plane where you separate data from models. We already saw that with a lot of the hyperscalers. Now you're seeing once that's separated, this contextual layer emerges and that context layer could be graph database. It could be a control plane. So you start to see this emerging. This is like nerd heaven. Like what do you do? You know, it's not. And by the way, there's no one control plane to rule them all. >> So, so you and Amit started this research with the five pillars. You guys wrote that um that research note and then I chimed in and added a bunch of things so I stuck my name on it. But um the five pillars were territorial like where the data and compute physically reside and then you defined operational like who runs the environment technological who owns the IP and the stack legal obviously what what what jurisdiction governs everything including access and the number one the most important one is financial like how do you get freedom from vendor lock in and there's there's some interesting things here. I mean, you know, on the financial front, there's there's there's a number of examples. Um, and we picked a couple. If I if you if I could briefly just sort of go through some of those. There was Microsoft, Uber, a company called Lindy, which an AI company. I think they're based in San Francisco in this the Swiss government in addition to Canva. And you've heard Microsoft say they ultimately want to, you know, kind of eliminate their reliance on, you know, frontier models. You know, uh, Mustafa Sllamon said they're paying Anthropic too much and so they're moving stuff to Microsoft AI. Uber spent a a year's budget in one quarter on tokens. Lindy said their pay payroll was smaller. Their token budget was bigger than their payroll. So they had to shift a 100% of their anthropic traffic to to an open model. And I think it was deepseek. And then the Swiss government was on this path toward sovereignty, AI sovereignty. And they had defined it. They were using the five pillars. But then at the end of the day, they decided to make a decision to pay Microsoft about $150 million, $125 million contract because, you know, they just needed the seats. They needed 54,000 seats to run their business. And so the point John of all this as you well know is you know it's every there's no such sovereignty is not a purity test. Every company has to make some tradeoffs and absorb some risks. The problem is a lot of companies aren't understanding those risks. They're letting the vendors sort of dictate those risks and then all of a sudden an invoice shows up and they say oh crap you know we got to do something. I mean you were talking about Broadcom and VMware earlier. You know people were caught flatfooted. What happens if a company like doubles their price? Okay, that's potentially a problem. So, who who controls your economic sovereignty is probably the most important factor. And but just make these decisions about whether it's technological, legal, or financial. Make them with an understanding of where the risk is and then have contingencies or reject it outright. >> Yeah. I mean if you look at our editorial weekly uh summary Dave that we put together with the team and um Silicon Angle uh team and the research team it's interesting the stories are kind of hitting the same thing but you know the the topline narrative continues to be trillions of dollars of infrastructure commitments are are are out there but they're meeting questions. Last week we talked about a bubble. We had some fun with that but the the real questions are what is the return on investment? What's the energy impact? Our jobs impacted and public acceptance is an issue. So this is on top of the competition increasing among AI companies. A lot more Neoclouds. Um since writing about um Argentum that brings up a financing question about energy. I'm doing a post inspired by your breaking analysis last week because you talked about in your breaking analysis you actually mentioned the word utilities. You talked about electricity and you talked about, you know, CLA asset classes talk commenting about Jensen's $500 billion uh commitment. He had all the top bankers. I mean, we're talking the the the CEOs of the biggest names, you know, JP Morgan, KKR, Blackstone, Black Rockck, Brookfield, they're all there at the table, but that's that's Wall Street. That's the financial. So, yeah, they'll find >> JP Morgan, by the way. Sorry to interrupt. JP Morgan wasn't in there, interestingly. But >> oh I thought well they they should have been in there but anyway the level of >> Goldman was Goldman >> Goldman I mean Goldman Sachs was Goldman Sachs not JP Morgan correct okay so what does that do and I think you nailed it okay and what I'm going to write about is on the other side of the coin so okay Jensen's basically saying look at here's my banking friends we will prop up support backs stop whatever word finance and and give confidence that the physical buildout of the capex is going to happen we all agree But in every single example of a utility or major infrastructure buildout, oil, electricity, airlines, they all had a hedging function. Okay, they had a futures market. So, I'm I'm dialing in a post for Monday that's going to talk about the energy futures because we're starting to see future prices of GPUs that ties to energy. So, there's literally no hedge to the execution risk, Dave, of this. So let's just say that all that money goes into the buildout. What if half the people fail, don't deliver. Maybe there's a force major or some sort of factor that says, hey, they just made the wrong choice on power and cooling. It could happen, [laughter] right? You know, or the energy is not there. So all these execution risks come in. But in every single, you know, you know, example of utilities in those asset classes had a hedging function and that was the future price. those future prices of oil, future prices of of electricity, airlines, if you're running an airline, you know, there's a price as an options a hedge on fuel prices, right? So all these execution risks are hedged. So to me, the missing picture of the Jensen story is that piece. Now, how did I come to that story is because at our NYSE studio, their parent company, ICE, uh, and Brian Bowman and I were talking about this, his really his idea kind of gave me the seed, too. his his idea, his conversation with me and your post. It's like, oh my god, that's it. Because the financial system can certainly finance things, but there's still risk. What's the ROI? That's a risk. Energy, that's a risk. Job thing, I don't think that's going to be a problem. I think that's going to just make people better or more prosperous. But look at the public acce acceptance. No more data centers in my in my area. There's a real backlash right now on on data center construction. It's gone political. I mean, who would have thought AI infrastructure would go political? Dave, this is a huge deal. So, risk management at this scale is is made me think about our conversation around does it pop. Certainly, there's demand. There's no debate. There's demand. How that demand curve looks like, what it is, that's going to be always be a debate. But when you get down to the money, where's the risk? There's execution risk. And what do financial players like Goldman do and all these guys? They hedge it. What's the hedge for AI infrastructure? That's a question. >> Well, and that's part of part of the Jensen backs stop. Although, as we talked about last week, it's that is option. Did you see the Wall Street Journal had an article today or yesterday um just about this off-grid power with the hyperscalers and the risks are there? I guess Bloomberg originally broke the story earlier this month saying some data centers that were, you know, building off grid combustion, you know, engine capabilities were failing like cracks because, you know, they're taking air jet engines and they're putting cobbling together bubble gum and bailing wire this data center off-rid power and, you know, shit's breaking. Um, but I I my So there's hench your risk scenario. I do think that the industry is going to solve that problem, but there's definitely friction to your point. I think there's a lot of misinformation about, you know, the amount of power that data centers consume. I mean, it it's it's not as large as a lot of the people claim and how much water they use. It's true that that a lot of the legacy data centers are you know poorly constructed just in terms of their efficiency but you know these modern data centers we you know we heard from CTO at Cororeweave that those interviews we did with those guys and it's true I've researched it you know to seed these these closed loop water systems and these data centers it's like two days of what a golf course would would use you know so you know imagine people coming >> explain that two days golf courses two days of two days of data center equals >> no two days. Yeah. To seed a closed loop water system for these modern data centers. It's about the equivalent of watering an 18hole golf course for two days >> which is you know you know people don't want to hear that. And you know it's really not >> Yeah it's not that bad. I love golf like good greens you know. >> So so stop golfing if you're really concerned. >> Yeah. Protest golf. We'll never see that happen. Um >> yeah, protest on there zillion other energy things, but it's interesting that it you mentioned that because the capital cycle, again, this comes back on some of the editorial and research we've done this week is continuing to get riskier. You know, you kind of tabled the first conversation last week, but you know, the demand curve is interesting. So, if you look at the news, say Broadcom, by the way, I saw Charlie Kawaz last night at the Rosewood. Um I also saw Ronnie L >> party at the Rosewood last night. You saw Curtis Martin. >> It was a normal day. Curtis Martin was there. We had Troy Vincent. there. He's the executive director of operations, EVP of operations for the NFL. Um, I showed Lonn Ronnie Lot the picture of me and him at Cloudera in 2010 and he laughed. He looked young people. I was 46, had black hair, you know, [laughter] when we had the cube there. But if you look at Broadcom, right, they announced a hundred billion in debt financing. That's now on top of Amazon and Google. So if you if you if you say to yourself, are these guys dumb asses or why are they doing this? Why is Andy Jasse committing Amazon to spend all that cash or take on all that debt? So, you have to assume that they're smart for a second and you know, we know Broadcom is super smart, too. We know the management there. We know the management Amazon and Google. So, they're making that bet. So, to me, that's a signal of relevance on the demand side. So, let's just assume that they know what they're doing. But, but if you look at that spend, that's interesting. Now I'm getting some scuttlebutt from our community that Anthropic is on a trillion dollar revenue number pace. They've 10xing their revenue and Broadcom has a deal with them or in process of getting a deal to supply them chips. So if you're anthropic and you're really sandbagging the revenue or because the numbers that they're reporting don't match the numbers that I'm hearing. So, you know, there's a lot of unanswered questions of, you know, where's that cash coming to? Now, I think they might be discounting the cash because of the reasons you mentioned about Canva. So, the question, does Anthropic and these guys know that people are really kind of just spending because they don't really know how to manage it and is that repeatable? It's a lot of questions. Again, back to the ROI piece. So, you have all this buildout and again, that's why I'm focused on this hedge piece because how do you hedge against all this capital? And you know, I think the financing is going to be in the trillions and trillions of dollars uh on the capbacks. I don't think it's going to stop building. But with if they don't have some sort of vehicle to manage it, there has to be some sort of futures market. And energy is tied to that's why I'm t tuned into the energy. But you know, anthropic might be doing more revenue than they're reporting. That's new information. No one's reporting that. I think we're the first ones to actually talk about it on camera, but if you look at their numbers, it doesn't match the 10x growth that we're hearing. So, well, and did you see the um the Wall Street Journal article that came out, I think Monday on on big tech spending $3 trillion more than people realize. Did you see this? >> Yeah, I did. >> They had that pyramid with the offbalance sheet. >> Um >> that's what I'm talking about. That was that that's exactly what you're talking about. I mean, you got Alphabet, you got Amazon, you got Meta, you got Microsoft, Oracle, Nvidia, Broadcom, SpaceX. I I think AMD's, you know, doing this. They're committing, you know, trillions and but but it's it's not in the balance sheet because they're setting up these special purpose vehicles, you know, like the, you know, the Meta Hyperion deal. They set up an SPV. Blue Owl is basically owns 80% of the SPV. Meta only takes, you know, 20% of it and then a holding company from Blue Owl raises a bunch of dough. Yeah. >> And then they they finance the construction in a bond sale [laughter] and then Meta signs on as a as a tenant. They agree to lease, but they don't have to put it on the balance sheet because, you know, it's an off-balance sheet transaction. So, but they're committing all this capital and so but they're unc unccommenced leases >> so they don't have to record it as as a as a liability. >> So that's three trillion. Whoa. I mean because we always talk about 6700 billion in capex and we're going wow that's incredible but there's far more being committed. >> Yeah. And that's why you're the the argument of like where the revenues come in where the profits come will land justifies the capital spend being deployed. Right. So the buildout that looks like too big to fail actually can fail. But then how would you hedge that? So you'd say, okay, we know energy is involved. We know tokens and GPUs are tied to revenue. Is there a future value of oil? So if you're drilling an oil, well, say you and I go to your backyard and we we drill for oil because we we think there's oil there. We don't know the value until it comes out of the ground, right? So again, you're going to invest in the capex. So there's a futures market on oil. So these are things I'm looking at because like okay I think you can say there's intelligence demand hence the numbers from the big companies that are committing. So you assume that they're smart for a second but there has to be an other side of the physical funding and I think that's what I'm going to start poking at because you got open AI and you got anthropic going public. Okay you got anthropics numbers that I question are accurate. Um, OpenAI, you know, gets the narrative that they're way way behind Anthropic when you look at their numbers this week that they put out a leaked all hands from Sarah Frier said that they're 50% of their revenue is now enterprise. Okay, so they're coming into the enterprise. So, you know, they're one model upgrade away from the top spot. You know, they are behind compared to anthropic, but I think anthropic just cracked the code, but who knows what's sustainable per your Canva example. how many people are building AI and paying too much. So I think cost control and cost efficiency isn't an audit issue. It's more of an architecture issue. Don't do test dev on production systems or >> that's interesting that that's VMware's play. That's that's better than the data play because they don't really have >> Well, I don't think that's their play but I synthesize it from the virtualization layer because you know I don't think they even have a product there. What virtualization could do is and this is what I think we're hearing from Don and Vidia is that you can have an AI factory on prem that just gives you unlimited tokens pay the cost for the system and then just use the tokens to develop and then scope it then put it in production then you lock in the forecasting. So I think I think we're going to see a different build and operation strategy around AI. And I think you know my takeaway from your Canva comment is that no one yet knows how to operationalize injecting AI into their business right did they get the right workflows going to be a lot of test dev experimentation moving that into you know a pipeline of you know productization and oper operationalizing it just is new. So you know Jensen's remember that time Jensen said they haven't spreadsheeted it yet and he was I put that in quotes because he meant no one's yet modeled out the value. So there's a lot of that risk on that side. And of course there's execution risk. If a data center screws up a piece of the execution over 12 to 24month build out of a data center you're hosed if you screw it up. >> Well and and you nailed the the trade-off if you're going to bring that on prem. You got to you got to write the check up front versus Yeah. Well, we have comps. You heard the you were saying some numbers like the payroll is less than the token cost. Like hello, buy buy a system from Dell or HPE or Super Micro, you know, whatever. Get it done. [laughter] Like that's >> if you have a steady workload, you know that then you can justify owning that capacity. Um, Amit says his modeling says it's about at about a half a million a year of steady volume. If you're you know your if your cloud bill surpasses half a million um you're you're probably should think about doing some of this stuff on prem if you have the steady workflow uh workflow and are there other benefits to consider? Yeah, I mean again we we have I mean I was talking to um an interview today was embargo because it's it's VMware um content for next week or the weekend after next and the conversation went something like this. Hey, you know, in the old days, if you had a desktop user coming online and they you put a desktop PC on their in their cubicle, you go in there, you put the software on, you have a Yeah. What they call it? Desktop virtualization. Um, >> yeah. And they get a lot VDI. VDI, I can't even remember. It's so old. It's like >> the most boring topic on the planet. >> Okay. So, VDI, but you know, users weren't crying a river, okay? They just go home, right, and have a life, right? It's like, oh I can't access the internet from my office with my apps I want, you know? Okay, no one was crying a river. It's an inconvenience. Okay, I'm working. Whatever. In this era, you're seeing people who are crying a river when they don't get their AI, right? It is so awesome for the user experience and as people figure out beyond checking the weather forecasts, when you actually do work with it, it's super powerful and no one wants to give they want more of it, right? They want they don't want less. They don't and they're not indifferent of not having it. They want it and they're pounding their fist. And with agents coming on, it's even better because you could get work done while you're sleeping, right? So like that's a dream scenario for a knowledge worker. So you know, all this is pointing to that. And again, this week, another data point to support this idea is that uh Stripe paid 7.5 billion to acquire Open Router, which wasn't around last year. Okay. Seven billion, >> right? So why why would Stripe do that? That's a great that was a really good example, right? That's that's a great example of sovereignty. They're trying to take control about how the stuff gets routed and they they want to control the meter. They don't want somebody else controlling it for them. They want meter intelligence and they're gonna that's going to feed into ROI, energy usage. Do I ship it to a lower tier compute platform or some maybe I mean I made that the virtualization layer up for AI, but it makes sense to virtualize something or like like network virtualization. Not that VMware is going to do it. Someone might do it, but Stripe's smart. So they paid 7.5 billion for a less than a year old company. Okay, that means that they're on to something. And Stripe's not dumb either. They're smart. So again, back to what what's happening. So payments, commerce, infrastructure. Okay, the framing there is I need to have intelligence. It's almost like their version of observability, right? So why would I want to do that? So again, this points to the fact that this is not stopping. And then agents are going to need to do commerce, right? cyber security vendors are already talking about moving down the stack into into um the control plane context layer with governance. That doesn't mean that the traditional RSA and black hat market is going away. That means you just got root security, if you will, into the infrastructure. So, I think the AI infrastructure is getting very interesting, Dave. And and I think the security is looking like it's going to be an AI infrastructure buildout. I think we rifted on this last week a little bit, but the vendors are, you know, all all over security and the Mythos thing, okay, is another thing that's going on. So, you know, the amount of um fear that Mythos is injected into every CISO has been completely insane. People are literally like, "Oh my god, this is too powerful. This is the word I'm hearing from experts." So you know the traditional networking vendors that are doing security paloalto you got um cribble >> strike >> crowd strike foret you know people are they going to build their own security so again security is now a big piece and it's got governance so all those things are in there and the agent is going to play a role you know patching faster real-time patching because mythos can find some longtail surface area in an enterprise um you know I'm I'm hearing from the the some folks that it's it's makes Swiss cheese look like bigger Swiss cheese. So like everyone knows they have holes in their in their vulnerabilities, but mythos has been it's relentless apparently. So it's like a war games that movie remember when we watched as a kid if you're over the age of 40 you probably seen war games. He just gets fixated on on gaming it right. So these agents like and this AI just goes to town. So, uh, it's going to be very interesting to see how the security space emerges because it it changes the resilience equation. It changes everything. >> Well, you saw that in IBM's earnings. I think they, you know, they said that their clients are distracted and, you know, we we heard that at, you know, IBM Think. We were talking a bunch to a bunch of their clients and they were like, h, we're dealing with this mythos stuff. It's driving us crazy. We're trying to get AI right and all of a sudden we gota, you know, worry about yet another threat. And then you saw Oracle respond with new guidance which basically changed their old guidance. Their first guidance was upgrade to autonomous. It's kind of self- serving but then they said if you can't do that we want you to change the way in which you you patch and your patch scheduling. Then IBM you know and Red Hat did that open source thing. Um and so yeah it's it's it's creating a lot of havoc. I think ultimately it's going to settle down, but there's that zone we're in right now of exposure and uh people have to deal with it. >> Did you see the Google fun um deal with Marll? >> I read about it. I didn't really pay pay close attention to be honest with you, but I saw that, you know, they're doing a deal. >> Well, did you see Broadcom stock price? >> Was it today? >> Yesterday. Yeah, it did. >> Marll jumped 9.8% yesterday on the news, but what ended up happening is Broadcom dropped. Okay. It was a 12.2 billion deal. Let me see what Broadcom's price is right now. >> Yeah, it was down 6% past 7% past five days. >> It's up 1% today, but you know, yes, let's go five days. It dropped >> because everybody's inferring that Marll is going to win that business. >> Well, Marll, I mean, Broadcom's in everything. So Broadcom's rising tide with everybody else. I think more if you're Google, you're just supplier diversity. Why would you put all your eggs in one basket? Makes no sense for Broadcom to drop from they were close. They were at 400, pushing 400 um August uh 17th. >> Um and then yesterday they closed around 36 363 364. Today they finished. >> You should be part of the Mag Eight, John. They're like the the eighth, you know, magnificent company. Yeah, >> I agree. That's I'm not worried about Broadcom's business with, you know, Google. You know, Google was probably trying to get a second source, maybe negotiate a better price. That's the AMD doll coffee cup kind of thing. But Broadcom is such a, >> you know, credible supplier. I just I wouldn't >> Monday Monday or Friday last week, Monday, actually Monday, they hit a high of 398. They closed Monday at 392. Okay, after the news on Google, it dropped to 359. >> Okay, >> so you know, pretty much down down on a downward slope significantly and then creeped up a little bit, popped up a little bit today around 9:30 a.m., then settled in down a little bit. So, it bounced around a little bit, but man, that's a that is a massive drop. I don't see Broadcom impacted by Marll. I I am definitely not agreeing with that if that's the reaction. Um and again like I said I think there's 20 to40 billion dollars of upside in Broadcom chips business because of the just just the activity on anthropic and open AI and all these buildouts because you know people are building heterogeneous systems and certainly there's a supply constraint as you pointed out last week on our podcast huge opportunity so I just don't see broadcom slowing down I don't see Nvidia slowing down either I mean what's your Definitely not. I'm I am those are the two companies that that I'm least worried about. I mean to me the worries, you know, Michael Bur, you know, obviously who he is. Everybody's down on Michael Bur in our in our industry because they think he doesn't know what he's talking about because he came out and he said, you know, GPU cycles are much shorter than people are putting on their income statements and their and their assumptions and and people say, well, you don't know what you're talking about. that he's it's proven at least thus far he was wrong about that. But he's actually a very thoughtful guy and he put a post out I if you read his newsletter it's called Cassandra Unchanged. He put a post out this week talking about h what you know why higher rates are not slowing the economy the way they used to in past cycles. And the reason is that the private construction buildout is being funded by these hyperscalers who will pay anything. And they're these companies are rich. Nvidia is rich. Broadcom's rich. The big three are rich. Oracle, you know, is rich. And they're willing to pay up for these, you know, the the the to get access to this infrastructure to build it out. And as a result, the economy, even though the rates are going up, is slowing less than it has previously. So that's, you know, again to our bubble conversation. To me, it's a sign of a bubble. Doesn't mean it's popping, but it could pop if all of a sudden, you know, that liquidity dries up. Like we talked about last week, um Jensen made a massive move, half a trillion dollar move to stave that off. And so to me it gets pushed down the line, pushed out if in fact there's going to be a bubble burst. And the the big thing to watch is productive utilization when end customers that are ultimately consuming AI actually start to monetize. And there's not blatant obvious evidence that that's happening right now. In fact, you know, more often than not, you see either people are just trying to still trying to figure it out or the big companies maybe are getting some wins. It's probably affecting earnings, you know, to to a modest extent, but it's certainly not across the board wins in in AI terms of productivity and monetization. So, that's that's one thing that I think we just got to keep an eye on that the old AI ROI narrative. Well, I think we'll have to see how these IPOs come out uh this second half of the year, fall for sure. You got Anthropic, you know, saying, you know, speculating. We had a report on Silicon Angle. The headline is report Anthropic hopes to surpass SpaceX record IPO raise when it floats. When it finally floats. Um, you know, that's significant. I mean, because SpaceX raised $75 billion when it went public and people thought it was going to, you know, pounce down. It's not. Um, leaked numbers came out from Anthropic. Again, I think they're lower. You know, I'm hearing rumors that, you know, with all the activity, they're on a pace 10xing their business, okay, over and over again. So, they could get to a trillion dollars in revenue if that continues. So, there's a path to a trillion in revenue. anthropic, not valuation, revenue. >> You're saying 60 billion now, right? AC be at a hundred billion dollar run rate by the end of the year and then 10x that is I don't know that would be obviously unprecedented. I mean 60 billion is unprecedented. 100 billion run rate by the end of this year is unprecedented. You know, people down on open AI at 40 billion, by the way. I want to see how much of that 60 billion or 100 billion or whatever they're calling it. I want to see how much of that is, you know, them counting, you know, other revenue from like hyperscalers, like distribution channels through Anthropic, you know, so you know, I want to see if this what apples to apples revenue numbers really look like between Anthropic and Open AI. >> Yeah. Well, we'll see what the what comes out on any more data, but you know, I think it's I think the the report that CNBC put out about their numbers, I'm hearing it's lower than what they reported. I mean, the numbers are higher than what CNBC reported for sure. Um, well, next week we got Nvidia's earnings. >> Yep. >> That should be a bell weather, right? On what's happening, you got Salesforce, Crowd Strike, Workday, Marll, uh, and others, Nutanix. you saw Workday under private equity, you know, lenses. Now the guns are out. Um, that story was out there. >> Yeah. And and Dell, I think September 1st they're reporting now. Is that right? >> Yeah. >> It's just talk, but you know, Silver Lake is in talks to buy Workday. Roy just broke that story. Um, >> well, somebody's gonna make some money on that deal. Silver Lake. >> I mean, I just interviewed Gabe Conroy. He's been a cube alumni. He was at Google and then Azure. Uh he's a cloud native. He was talking about some of the cool intelligence that's coming out of workday. Now, workday is a great example of, you know, the work that you guys are doing, you're doing around system of record, systems of engagement, systems of intelligence, and systems of agency, the stack of progression. they have record data, they have data, you know, so certainly they can easily turn on some AI for their own customers, but now also they can feed other systems. So that's where the agentic conversations coming out and a lot of these conversations I'm having with a lot of these um leading engineers around graph databases. So the word the knowledge graph sits as the layer between data platforms and say um models, right? So you're starting to see graph databases and other approaches. So if you if that continues, data from Salesforce, data from workday, data from any ERP or CRM can be input into AI. So you know, it it it kind of smashes the whole SAS cop apocalypse argument. Yeah, stuff some stuff can be coded from scratch if you're starting from scratch. I mean, if we wanted to build Salesforce for our team, even though we're a customer, we're not a big power user of Salesforce, but we do use it as a system of record and reporting. But, I mean, we could probably, if we threw money at it, could have someone replicate it the same functions, log calls, you know, log data, book business, use the tools, generate a couple reports. Now if you're a complex user probably a little bit different but you know I think there's use cases of you know old guard you know or modern legacy getting rewritten >> if if they if it's a tradeoff works like does it cost more or less >> do we get the same or better functionality >> it was interesting I mean I always thought 37 signals would come up with a better CRM than Salesforce but then you know I was a big 37 signals fan as you know we even tried to use their system but you know Salesforce was just better but yeah it's just you would think but you know they're so entrenched you know it's it's going to be interesting to see how this how determinism meets you know stochastic analysis and probabilistic LLMs and how that all comes together. There was an article in the in the paper about paper in the journal digital about remember that guy Dave >> he weaves in digital didn't you? [laughter] >> Dave Wright remember remember him service now? Yeah, >> they, you know, I'm sure it was like a I don't know, probably a placement piece, but um, you know, he's talking about the SAS apocalypse and, you know, what's coming around the bend and obviously Service Now is one of those companies that, you know, is under the gun here to prove that, you know, they can they can survive as is Salesforce, right? I mean, I think those guys will get through the the knot hole, but there's going to be some pain in SAS. I think a lot of SAS companies are in deep deep dudoo. Yeah. All right. Well, I'm reading this article now that David Wright wrote Cube alumni right there. Look at him. >> Yeah. >> Yeah. Yeah. >> Well, what's the takeaway from this? Is he saying um it's possible? >> You know, he I think I think his vision is right. It's it's it aligns with what George and I have been talking about for a long time, which is, you know, the the potential of creating a digital twin of an enterprise is is the future. And to do so, you got to have process knowledge. And they possess that process knowledge. And the question then becomes, well, you know, who's going to disintermediate that? If are the LLMs going to do that? They don't they're not set up to do that, but they could partner. They could, you know, invent. >> Um, they could acquire. But, you know, I think that I think what what's likely is you're going to see partnerships between, you know, the cognitive layer, the LLM you know, frontier models and the SAS companies and you know, the SAS companies, the leadership, they're not dumb asses, you know, they'll protect their business and and and they'll partner up, maybe they get acquired um or maybe they acquire. >> Yeah. And so >> I love I love David Wright's line here. He says he believes that one of the largest areas of potential is in creating a digital twin of the entire company that will allow leaders to simulate different business scenarios such as price cuts, new regulations, and business expansions. The idea is to model different future scenarios that inform their big decisions. I mean, Dave, we've been talking about this for three years. Digital twins, not as a metaverse kind of thing, but more of, you know, you can actually run simulations with >> AI and run through things if you have the right workflows. Again, the part hardest part is getting the data like what's the data feed? Who's the domain experts? Um, and I think that's what's coming out of the specialized intelligence trend that's booming, which is you can have domain expertise and you you don't need you don't need the biggest model. You need the right model. That's the number one thing coming out of all this agentic enterprise conversations is it's a mish it's a mishmash a mashup of models. Separate the models from the data platform. have a control plane, context layer, knowledge graph, something in there to manage resource and leverage the infrastructure and AI to go at it. And so you got to have that domain expertise >> and that's usually enterprise unlocked data or lock down data >> and that's where the tacet knowledge is and an enterprise if you're going to have a real- time you know digital representation of your enterprise you have to have that workflow knowledge you have to have the tacet knowledge of a human and that obviously takes some time to to to ingest and develop and learn and and improve but you know companies like service now have that capability they have that knowledge It's the knowledge existed within their clients, but they, you know, can surface it. And so to me, the ideal world is you marry the LLM's, the frontier models with that process knowledge, and that's what's going to be, you know, create that digital twin of the future. And I think that is Dave Wright's vision. He is a visionary. We've had him on a number of times. >> Yeah. And yeah, he was a visionary for Service Now is just a company that looked like nice, you know, his nice enterprise business turned out to be a monster machine, cash machine. It's interesting you he brings up that point. You mentioned the tacid knowledge. You remember Peter Ducker? He was the father of modern management. He wrote books about it. People would read it. Um what you just said essentially no one's written the Peter Ducker like book that says what's the modern man management technique to what Dave Wright and what we've been talking about. How do you lay that out? Because you got to document the business before you can simulate it. You you it's like saying if I'm going to build the company's brain with digital and AI, you got to have one the knowledge. You got to have the brain. It's got to work. You can't hallucinate. So you in order to make that highly functional, you got to know everything. I mean just know like the workflows. You don't have to know every single corner case. But if you're not getting the input, if you don't know the process and that knowledge, you're half brain dead basically. So you can't have >> working on it. you know he's working on is Jeffrey Moore actually he's working collaborating with George on this for the uh the economist article that George that feature that George is doing and uh so he's done some work on that but yeah nobody's written a definitive treatise on this yet because I think it's all being sort of created as we you know we're making it up as we go along. Yeah, I mean treat Peter Ducker is is the is the master of of ma of traditional management in the in the you know kind of like our era grow when we were in college coming out of college um all the business schools had his stuff um all about efficiency there's nothing so useless as doing efficiently that should not be done at all [laughter] this famous line um >> yeah all right Dave well We got a big week next week. >> Delete, delete, delete. Yeah, next week >> we got a huge week. Um, got we got quantum going on. We got some news going to be hitting. I got a post I'm going to put out on Monday on the futures around the intercontinental exchange that Brian Bowman and I see it was really his initiation genius idea and I didn't get it at first like I get and then your post made me connect it because what you were referring to electricity I'm going to get that out on Monday and then there's some big announcements on Tuesday and then through the week and we got earnings from Nvidia. We'll see how that goes. That should be very telling. Um, I'm expecting it to be a monster quarter. I mean, unless something [snorts] weird happen and then the stock will go down. [laughter] >> All right. Well, >> it' be a shock if it weren't a monster quarter and then a beaten raise. But, uh, and then the following week, we just we start the season, John. It's >> Yeah, >> we got VMware Explore, we got Crowdstrike, we got Go, go, go. We got, you know, an event we're going to do for the first time, this is the first time they're doing an event, is Coreweave. Um, and, you know, don't count out the IPOs that are coming. So, I think there's going to be an IPO wave coming. And there's rumors that Anthropic or Open Air might go out by the end of the year. Definitely in 2017. Sarah Frier on the record saying, you know, definitely will it be a public company in 2027. So, yeah, you know, the question is who goes first? >> I think Anthropic is probably going to go first. I would damn if I had to bet I would bet Anthropic goes first. I think their financials are probably more conducive to an IPO right now. >> The sovereign stuff is hot for us. Keep a track on that folks if you're listening and watching. Um and there's movement that we might be at the F1 uh in in October. And you know, we even talked about reinvent coming up in December. Like >> which where's that? Where's that one? Which F1? >> That's Austin, Texas. >> Austin. >> Yeah. We're going to definitely We'll have the cube there. NYC wired our wired brand will be there. Um Brian and I and the team. So I think there'll be might be some really good sovereign conversations. Certainly a lot of activity. Dell's a home game for Dell. Obviously Dell's from Texas. Um so and and Austin. So we're going to try and see if we can get some Dell AI factory conversations. >> Texas Texas is the data center central. You know, it's the one one state where people aren't throwing data centers out of the, you know, out of the locations. I mean, they're sort of embracing it. So, Texas is going to have low energy cost and great AI. >> Are you going to go down to the Dallas launch of the um the studio down there with NYC and ICE? >> I don't know yet. I just got I just got word of it yesterday. Got a I got a bunch of stuff on the calendar and I got to move. Yeah. >> A lot of things happening. We got Tech Exchange. IBM has got a whole new focus on Tech Exchange. I had a briefing from IBM on that. They're really got the Hashi Corp community. They got Red Hat. They have all that crowd and they're really trying to put that together. Again, we'll see how that goes. Um, >> they got something big coming coming next week. >> I can't talk about it. >> Yeah, I was supposed to be on that briefing. I'll have to call you after this pod and find out what's going on. >> Got the ear to the ground, Dave Volante. Well done. So, we'll see you next time. Enjoy the eighth weekend. >> See you later. Cool. >> All right. All right. See you.