The Six Five Pod | EP 285: AMD's $100B Data Center Vision, SoftBank's $5B Nvidia Exit, Government Shutdowns and GPU Shortages

On this episode of The Six Five Pod, hosts Patrick Moorhead and Daniel Newman discuss the tech news stories that made headlines this week. The handpicked topics for this week are:

1. AMD Financial Analyst Day Breakdown: AMD presents long-term growth projections with over 35% revenue CAGR. Pat & Dan discuss AMD’s 10-15% GPU market share projection, emphasizing Lisa Su's track record of execution and credibility. 

‍

2. SoftBank's Strategic Repositioning: SoftBank sold its entire stake in Nvidia for $US5.83 billion ($8.9 billion). Masayoshi Son, Chairman of Japan's SoftBank Group plans to reallocate capital to OpenAI and other AI infrastructure investments. Hosts discuss the potential of ARM-based AI chip development.

‍

3. Anthropic's Infrastructure Investment: New $50 billion data center construction commitment with FluidStack. Claude Code is driving significant revenue and a path to 2028 profitability. Comparison with OpenAI's infrastructure strategy and independence goals.

‍

4. Cloud Infrastructure and Capacity Deals: Nebius secures $3 billion deal with Meta for GPU capacity. Meta's strategy of risk-sharing and outsourcing during demand peaks.

‍

5. The Depreciation Debate: Patrick argues there’s a 6-year depreciation period for GPUs based on historical usage patterns, citing continued use of A-, V-, and H-series GPUs. Questions are raised about reticle limits and performance scaling sustainability.

‍

6. Government Shutdown Resolution: Senate votes to reopen government after 43-day closure, leaving in its wake and estimated $11 billion permanent economic loss and $16 billion in missed wages. Hosts break down the market’s mixed response with AI sector concerns overshadowing the reopening.

‍

7. Cisco Earnings Analysis: Beat on revenue and earnings with solid enterprise performance. AI infrastructure orders are expected to triple to $3 billion in 2026. Hyperscale AI orders are at $1.3 billion with a strong growth trajectory.

‍

8. CoreWeave Market Position: Stock down 33% from three-month peak, but still up 16% over six months. Data center build-out delays appear to be impacting capacity and revenue projections.

‍

9. Applied Materials Performance: Beat expectations despite revenue decline from the China market loss. Future growth potential from TSMC, Intel, and Samsung US expansion.

‍

For a deeper dive into each topic, please click on the links above. Be sure to subscribe to The Six Five Pod so you never miss an episode.

‍

Or listen to the audio here:

Disclaimer: The Six Five Pod is for information and entertainment purposes only. Over the course of this webcast, we may talk about companies that are publicly traded and we may even reference that fact and their equity share price, but please do not take anything that we say as a recommendation about what you should do with your investment dollars. We are not investment advisors and we ask that you do not treat us as such.

‍

Daniel Newman: The best part of my day on Friday is always the beginning. It's the workout, it's the podcasting. And now I know for sure. Pat, in one hour my day is going to just perpetually and pervasively get worse. So that's just what I want everyone to know out there is this is the best part of my day. And you're listening to this on Monday so you have no idea how bad it might have gotten. I don't know yet either. I just know it's not going to be better than right now when I'm here with you. So you're looking good this morning. I got to tell you, I put out a post last night that you know after taking a seven figure drawdown yesterday that I took an extra pump this morning so I might be a little over the top. And I actually did. See the good news is pump is not caffeinated. It's like all the pre stuff without. But then I actually took caffeine on top of it. So I took that and added some caffeine, took some creatine, and popped the peptides. I did a killer workout. Killer workout. And I'm feeling a little shaky. I'm not going to lie right now. But I'm super happy to be here with you. How are you doing?

‍

Patrick Moorhead: I'm doing well. You know, I am trying to figure out like you're an adult and you do adult things. You're taking a pre workout. That looks like what a 12 year old would have gorillas on. It was awesome.

‍

Daniel Newman: Like mythical with horns and it says pump.

‍

Patrick Moorhead: Yeah, I mean I'm trying to rationalize that, that, that, that whole thing and I'm also wondering like, like where does it end? I mean you look great. You know, you sent me your new body. I mean where does it end? Dude, you're adding more muscle, losing more than 10%.

‍

Daniel Newman: I gotta break 10%.So I'm at 11 right now on the Embody. It's 10 on the Embody. Cause I don't think I'll ever get there on the deck. So the deck is really the real honest truth of where you're at. But like you know, it's like you gotta have a purpose and the other purpose I was trying to get is to break 400 on the bench. Now I will be candid on that one. I'm getting closer, but I'm getting scared and getting scared because like I'm, I love being fit. And if I learned anything from watching your journey while chasing those kinds of numbers, there's risk, you know what I mean? Like, is it worth a six month injury to try to push that kind of weight? And I don't know. But like the ego, man, the ego takes over sometimes.

Patrick Moorhead: Numerics too. It's numbers, right? Which is, it's one thing to look at yourself in the mirror, which I know we both do a lot these days and maybe we still do share topless picks with each other.

‍

Daniel Newman: Mark, not everyone's gonna know about that.

‍

Patrick Moorhead: Well, that's okay. I mean, listen, you got to show it off. So there's that and then there's the quantitative dynamics of it and that's, that's where I broke. But I'm going for reps now, right. I'm not moving up in weight until I can do three sets 12. Right. And I'll move up five or ten pounds. I'm going up slow and steady these days, which you know isn't as fun. Doesn't hit my ego or my aggressiveness right where I'm listening to the absolute worst possible music you could be listening to to get, to get pumped up. But I'm going for, I'm going for longevity. And my new KPI with my trainer is injury free growth. It's called profitable growth, Daniel.

‍

Daniel Newman: I think there's some lifts that I know are just more likely to cause injury. Deadlifts, squat back, squats, bench press, shoulders like as you get old, you know, like I have soreness when I don't lift. So you have to imagine what happens when you do. And then there's other things like obviously things with cables, things that are more natural, body weight things tend to be a lot lower risk. Now I do tend to hurt myself sprinting. That is my kind of Achilles heel because I try to sprint and these hamstrings just ain't what they used to be, let me just tell you that. But anyways, it's all good. Pat, what a week. We're going to be able to hit a big week on all ends. You know, we've got, we had a little bit of the markets excited. We were shocked. The government reopened. By the way, that was kind of a big deal after being closed. One of the longest closures in history. Looks like flights are going to return to normal for Thanksgiving. You know, we're not a political show. But I'm sure that'll weave its way into the theme of the conversation today. There were some big events this week. I mean earnings were kind of calming down. It was a little calm before the storm. There were a few earnings this week and we'll talk about them. But we have Nvidia next week so we won't have to talk a lot about that. But you know, maybe have a few words in the preview. But we had AMD's financial analyst day. Softbank made big announcements exiting its position. Nvidia again visited the quiet. After being quiet for so long, Anthropic makes a big capex bet. Like, you know, we're kind of thinking, oh, they're getting to profitability. IBM shows a path to useful quantum computers in just four more years before they become useful. Nebius gets a big deal with Meta and then there's a few other items if time permits, we'll get to Pat, you and I. I think we just have to do a flip today and talk about this depreciation thing because there is such a great opportunity to debate both sides of this conversation. Everyone's probably heard about Michael Burry, the big short, the guy that shorted housing and is now an expert in shorting AI. Where is he? Did he shut his spun down? Is he exiting the space? Who the heck knows? We'll come back to that too. By the way, am I talking fast? Because it might be all that pre workout and pump bulls and bears. We'll get to what's going on. We talk about the shutdown, we'll talk about earnings at Cisco, Core weave. We had Nebius. We'll have a preview, some other things, Pat, depending on time. I know we both have a hard stop at some point, but that doesn't matter to you all. You just want us to get through the, through the docket. So Pat, that was a lot. All right, we got a big decode. You heard me, run it down. Hey, let's get to it. All right. I'm starting to wonder. All right, Pat, I actually went to this event but you probably had a better seat than me because it was great. So I actually, believe it or not, I had the stream going on my computer as I was sitting there watching it. So. Because it was easier to see it and get better images to share. But give me the rundown on AMD's financial analyst day.

‍

Patrick Moorhead: Yeah, so it was a big day for AMD and I think what this was was really putting numbers against the open AI announcement. If I look at the analyst ratings beforehand and afterwards. Before FAD it was about $283 and after fat it was $294. Right. So that's you know, a $10, $11 adder at her on that. So for the most part it had been baked in but we saw AMD go up, you know, 10% which is a lot higher than the re rating of about 4%. So you know, what did they say? I mean they, they said a lot and I think the money slide was over 20 non GAAP, EPS and long term plan and that Lisa and the team gave themselves a lot of wiggle room with 3 to 5 year, 3 to 5 year look over 35% revenue CAGR. You know some of the things. Again I'm not a financial analyst but I really love to see the market share claims and data center CPU, you know, over 50%. Probably the biggest surprise to me was client PCs at over 40%. One thing to, one thing to say, one thing that was interesting in there is the server. First of all, it's all revenue share. Okay. And because I looked a little cross eyed at the ARM server in particular and I think that would be very hard to do with the amount of ARM cores that all the hyperscalers are cranking out. And obviously somebody in a back room at AMD had to apply a revenue to cores which is probably the design, the design margin or, or whatever Google, Microsoft and, and AWS is paying for those. But anyways it was very. Those are really big numbers, meaningful numbers. The street showed up, they wanted to hear about GPU.

‍

Daniel Newman: Okay.

‍

Patrick Moorhead: And while there wasn't a distinct data center GPU number, there was a hundred billion dollar data center number that got put out which is a combination of CPU and gpu. AMD is not prepared to give a split between there and I understand why everybody wants a simpleton comparison to Nvidia. I get it. But yeah, I mean AMD finally gave the street something to model and, and now it has to, has to ship. You know, had one of Nvidia's largest investors, you know, kind of. What's the right word? Asked me a lot of direct questions via, via, via text. And in the end, right. It's roughly 50.

‍

Daniel Newman: I'm just kidding.

‍

Patrick Moorhead: I can't say.

‍

Daniel Newman: Oh come on.

‍

Patrick Moorhead: 15. 15 shares for AMD in three to five years in data center GPU seems reasonable to me.

‍

Daniel Newman: Right?

‍

Patrick Moorhead: Nobody's modeling 50% market share. And I think probably in the end it could be, you know, AMD at 10 to 15% market share, XPUs at about, you know, 20, 20 to 30 and then video at around 60 to 70%. That just feels right to me. And again, I don't have the fancy spreadsheets that you and other people do. Quite frankly. I think they add limited value. No, I mean yours are good as.

‍

Daniel Newman: Long as mine are okay. I don't care about the other guys.

‍

Patrick Moorhead: And, but, but I, but the. Nobody is going to let any one company maintain 90% share. And then I look at, you know, fundamentally they're like, pat, why are you so confident in what AMD said? I said, you know, my response was, I don't Babe Ruth anything. Okay, let me be clear. There is risk. But tell me the last time that Lisa made a statement or a claim in 10 years that didn't end up being true. And I, I leave an uncomfortable pause.

‍

Daniel Newman: Well, I mean, listen, I think that's a great point in part of the. I think that she's a bit of a calming voice in all the chaos because you know, she doesn't go over her skis. She's just not a salesperson. Like she may be cousins with Jensen in some distant way, but like she is not the one that's going to come and say the more you buy, the more you save. That's just not who she is. She's very thoughtful. And while it was a good opportunity for AMD to come out and talk about their growth story with a more than one quarter guide ahead, here's a couple things I'll think about. I was talking to the Wall Street Journal yesterday about this but they asked like where, you know, what's the difference in this situation? Because within the CPU space, AMD has been incredibly opportunistic. To their credit they've built a great product. They've won a lot of market share. They won first by going after the biggest buyers and were able to pivot the biggest buyers that are able to make decisions and change faster. They did throw out that 50% market share. And what's going to kind of get them to that 50% market share? Well, it's really two factors. One is continuing to win sockets from, from intel, which is achievable still, although hard, harder in the enterprise. It's gonna be harder in the enterprise than it was in hyperscalers. But the other part that maybe isn't being calculated at 50%, I don't think ARM is being fully estimated here and I think that's going to be a challenge. On the other hand though, her success over the 10 years taking the company from where it was to now nearing a half trillion dollar valuation also came from just great execution. She used the word execution machine. That's absolutely what they've been, you know, but then all the question now on this next leg up in the company in its future is about can it execute an AI. You know, we have estimated like 3 or 4% of the GPU market today. I actually give them almost a. It's almost a shoe in for them to get to 8. I think they'll get to 8 just with the next product. It's good. I think they'll get to 8 just with the supply constraint. If they can build it, they'll sell it. We're seeing that really across the board with them getting to Rack scale. I still think software is a rate limiter. I think they're making improvements in software, but it's very sticky. That will take some time. But to your point, like 10 to 15% if they execute, that is not an unreasonable number at all that they could get there. And at 10 to 15% of what's going to be, you know, and if you include CPU networking and GPU in the whole Rack scale, I mean that is 100 plus billion dollars. I mean, no doubt about it. So I think they're giving a very realistic number. They're addressing 100 billion of what they think will be a trillion, you know, in the future. And I think it's very achievable. I think a lot of the other stuff they're doing very well just kind of gets glossed over. But I think that's because right now it kind of doesn't care. Nobody cares. Nobody cares. Like, you know, I talked to Dan McNamara, like a CPU guy. I'm like, look, you guys have done an amazing job on Epic. It's great business, it's. But like right now, like the market wants instinct, they care about instinct. The investment thesis for Financial Analyst Day was all about what's going to happen, not just instinct, but Helios and then what comes after Helios. Because that right now is where all the upside is. But give them a lot of credit, Pat. In client, in, embedded in different areas, they've continued to execute. Not just execute in the short run, but they've executed over the long term. I will.

‍

Patrick Moorhead: Daniel, I'm wondering is, this is just a conservative piece, right? Because nobody really knows the Rack scale performance on inference for certain workloads on an Mi 450. Because it's not ready yet. Right? It's not going to be deployed till Q3. And Lisa did pull it in from a second half claim to Q3. I think that was a positive, but we just don't know. Okay, and imagine if OpenAI, you know, they're getting a sweetheart deal, let's say it's a, you know, a 30% price discount and they've got the equity part of that where they can go in and get a piece of AMD. I mean, you get that flywheel going, Daniel, because, you know, this, this very large investor, you know, said, hey, basically that, you know, this person and that person say that Mi450 is not going to be competitive per token. And my response is I don't think it has to. And oh, by the way, anybody making that claim, they don't know. They don't freaking know. Their engineers went in and did an architectural analysis on a spreadsheet. Trust me, I did a ton of those when I was at AMD. But until you have silicon in hand running the software, you just don't know. And if reality shows up, I could see OpenAI going, accelerator pedal to the metal down. Because not only do they get a better price, but they get equity in one of the highest flying companies in all of AI.

‍

Daniel Newman: There is a lot of commonality and a lot of mutual benefit for those companies to succeed together. So, so, and by the way, Lisa handled the question about OpenAI. She was hit by that on that one about, you know, where the money comes from, how they analyze risk and stuff. And I think the industry is answering it well. But I think the simplest answer, and most people think this is a blow off answer, is there is enough liquidity chasing the opportunity to invest in OpenAI at basically any price right now with its growth. And as long as that continues to exist, they can continue to raise money. And I think that's, you know, as sort of terse as Sam came across in that BG interview, that Brad Gerstner interview about it. Like, you know, it's arrogance and you know, you listen enough to him and you get that sense. But like, he's like, you know what, Sell your shares, there's someone willing to buy them. Like if you don't, if you don't want to be in, sell your shares. And you know, it's because there is a buyer, a bid at every price. Right now I don't know if that's the same for the CEO of Core Weave that kind of had that meltdown on the, on the Internet right now on his CNBC interview. But like right now all these guys do have the commonality of knowing that the demand for what they can offer is a lot higher than what they can supply. And you know, unfortunately that does allow you to piss people off maybe for moments in time. Just hope that that never, the equilibrium never comes and you can be that way forever. So a little, a little psychology there, Pat. But anyways, let's, let's, let's move on SoftBank to talking about, you know, exited Nvidia and this became a bit of a headline story for a day. I think I saw you go on TV and talk about it. I know I talked about it a few times, you know, did Masa, is he about to, you know, I think he did lose out on like 150 billion by selling when he did the last time and you know, this time I think he turned like 180 million into almost 5 or 6 billion dollars was the outcome. So he did really well on it. A thousand percent gain from his time and investment and he's basically saying he needed to reposition. He also sold a big chunk, of T Mobile in order to get the cash it needs to invest in OpenAI. A lot of criticism of this decision, but I think when you put that criticism in context, SoftBank is a company that invests in holding and portfolio, actively manages companies. They're not purely passive investors. You know, they're in arm, they're in graph core, they're in ampere, they're, they've got a big bet on Soft Stargate. They believe in this sort of build out of this large scale AI infrastructure of the future. And I think what they're probably looking at is at a 4 to 5 trillion valuation. What's the chance of it doing 1000% in the next 5 years versus what's the chance of OpenAI seeing really meaningful returns from where they entered? I believe that $250 billion round was the last big commitment that they made and I'm guessing they see that trillion dollar path. So they say what's the likelihood of Nvidia 4Xing versus the forex that's going to come from OpenAI, not to mention their need to support it. So I think people are kind of making a little bit of that wrong. It's like a false comparison. I mean if you're a pure passive investor and you are wanting to be a little more secure in your money, there's no way you wouldn't look at Nvidia as a much better bet than Open AI. But if you're willing to be active and you're willing to have a higher risk for a higher return, I think MAS is on the right track. You know, I also think there is a little bit of a subplot here. I don't know how true this is, but we know ARM is going to get into AI chips. We know they've made those investments in graphcore, the investments in ampere. They have been building a bit of an IP and chip stack that they want to be able to support. And at some point, I think they want to see if they can access that market. The same one we just talked about with Lisa, the same one that Nvidia is in, the same one that these custom chips are in. Because that's going to be, we think, somewhere in the 100 to 200 billion range over the next few years and that's a lot more revenue. And with their positions in those other companies, this may not just be about Nvidia. It might be about the fact that they could build a chip that then supplies these Stargate and other deployments that would directly benefit a lot of revenue.

‍

Patrick Moorhead: Yeah, great commentary. Yeah. I was on CNBC Asia where I've somehow become the Masa son guy. They brought me there to discuss that for unknown reasons. But yeah, I mean, Masta San has a better investment that he needs to make to get a higher rate of return. He's looking for maybe a hundred Xer. And it's not that that makes Nvidia or T Mobile bad choices. Both stocks have been doing great, especially Nvidia. But I did take pleasure and walk watching the videos of Jensen ribbing Masasan on stage and then giving a, you know, a nice group hug right after that. So, yeah, I, I'm really looking forward to Masasan's next step, how he's going to use ARM and all the IP blocks you talked about to, you know, know, maximize his portfolio in the future. So, yeah, big, big stuff, man. Masasan didn't flinch. He just switched tables.

‍

Daniel Newman: He just switched. And obviously he's been a big supporter. You know, it's funny. Everybody wants to lecture you about when you have, you know, unrealized losses, but then when someone sells it, makes a lot of money, they somehow also make that a negative thing. Like you'd made a thousand percent. Jesus. Like, give the guy credit he did good, you know, and let him move on. Pat Anthropic. You know, and I said this interestingly. There's been a lot of comparisons with OpenAI. They're going to get profitable first. They're growing faster, growing faster relative to getting profitable. Maybe they're doing things the smarter way. And then all of a sudden out of nowhere it's like, oh, they're also going to spend 50 billion, which 50 billion by the way is nothing compared to what will happen. Trillion and a half of spending commitments that have come from OpenAI. But like, what do you make of this Anthropic announcement? Why, why are they doing this instead of just, you know, doubling down on Amazon and others?

‍

Patrick Moorhead: I mean listen, this is exactly like OpenAI and let's be really clear, Anthropic makes most of their money off Claude code. And you know, whether it's code completion, doing complete end to end apps now it's got like security checking mechanisms. The original Vive coding was trash because it was just a security nightmare. But man, is it getting good. And you look at the amount of computer science students unable to find jobs and we don't have to debate that, but I can't imagine it's anything other than using these tools. Somebody in my family, it's very close to me who happens to be a software developer, has turned the tables on that and he uses cursor to become a capability of a five to eight year experience programmer. And there's also, it's funny, I saw you opining on Anthropic and enterprise. Let's be really clear, that's code completion. They're doing nothing else there. I'd love to see them show up with some sort of an on prem or hybrid multi cloud agentic orchestrator at some point, at some point the ability to create small models, vertical models. But what I love is, is everybody talking about how do we, how do we, how do we monetize AI? They're the perfect, perfect candidate here. It's like one application they're good at and they're driving a ton of freaking revenue and will be profitable on paper by 2028. So of course I had to come in like OpenAI and make a bunch of multi billion dollar commitments. This is a $50 billion US data data center construction with a company called FluidStack, which I'll be honest with you, I'd never heard of before. And the same rules apply to any other OpenAI focused data center. How are you gonna get the power, how are you gonna get the water? And how are you gonna have all of that as cheap as possible, right? And with a three to five year lead time on natural gas power generators, I still don't know how all of this, all this adds up. So, yep, open AI had it. Anthropic needs it too.

‍

Daniel Newman: Yeah, I mean, I think Lisa said it, I've quoted her a few times, I'll give her credit. But the future of intelligence is going to be directly correlated to how much compute you have. And of course she's going to say that and Jensen's going to say that, but I tend to believe it's true. And so you know, the dependence on for compute from companies that have the ability to constrain compute or that have a limited supply of computer anthropics. Just like OpenAI doesn't want its entire future to be dependent on whether or not Azure or Amazon has what they need. So they have to build for themselves. To your point about the cloud code, to be clear, I just don't care. I don't care what the enterprise use case is. My point is it's an enterprise use case. Enterprises updating and modernizing and improving and building applications and code using cloud code is a great application for enterprises and they're paying for it. And they're paying for it and they're getting augmentation, they're saving money, they're getting upgrades and updates deployed more quickly. And cloud code works really well. We use it in our own enterprise. Like our entire platform is built using Anthropic and Claude as the code backdrop of it and we pay for it. So I'm a believer, I think it's good. I think when people say there's no enterprise use cases, it's an enterprise use case. So that was just something of note there. But I mean otherwise, like, you know, you're exactly right. You know, we are in this exciting era where everybody needs to buy a boatload of compute. And so Pat, you know, if it's a bubble, it's got ways to go before it can burst because we got to get all this built out and then what we have to do at the other end of this is find out that there is no use for it and nobody wants it willing to pay for it. By the way, there was an interesting quote I heard on CNBC this morning. It was about like they had an event called Delivering Alpha and they said a lot of people are conflating that we're in a super cycle. And they're conflating super cycles, like tech super cycle and bubble, meaning that super cycle, you are going to have exuberance and you're going to have frothiness and you're going to have pockets that will be potentially in bubble esque territory. But it doesn't make the whole thing a bubble. And I think that's a lot of what we're up against is we are in a true technological super cycle right now. So. Interesting. Pat will pivot a little bit to something that you and I can talk about limitedly, admittedly I need to go get my PhD before I can talk any more about it. But is Quantum IBM? You know, they basically announced that, you know, IBM, Quantum Nighthawk, it's their processor built for quantum advantage offering 30% more complexity. I think the big thing is that they see by the end of the, you know, they're going to get to the quote unquote quantum advantage, which again is an interesting debate about qubits and number of qubits. But they're more importantly talking about getting a fault tolerant quantum, which really means bringing use to Quantum somewhere around 2029. So that seems to be starting to be more of a, well more agreed upon timeline. Seems to be the end of the decade. And you know, they think that they are the, you know, they're kind of saying hey, we're the best positioned company. We have the whole quantum stack. They have the most computers, they have the most developers, they have the most uptime, they've got the software Qiskit, you know, they like to liken their Qiskit software platform to the CUDA of Quantum. That's kind of, you know, it's a great analogy. I don't think it's anywhere close to scale yet. But there was a time by the way back about eight, ten years ago, Pat, word the same. We could said the same thing about cuda. Right. So you know, one of these interesting things. But they're basically saying that, you know, Quantum Advantage, just to be clear, what it means is that a quantum computer is able to solve a problem better than pure classical. That's how they're defining it. And you know, I'm just looking to see if there's any other, you know, heavily noteworthy things here. You know, they're, they're basically, you know, the other thing is just about getting to the error. What do they call it? Fault tolerant. And the fault tolerance is basically just the machine can stay in a state of usefulness long enough to solve a problem. That's one of the biggest restrictions or rate limiters of Quantum. You and I both talked a lot over time that IBM has definitely gone big here. The market doesn't credit it much for the research part of its business, but it certainly does seem to be showing progress. And while we may not love some of the comments that come out of these zero revenue Quantum companies and all the exuberance and frothiness around some of them, IBM is a very real company with a billion dollars in revenue with thousands of real developers building and some really good use cases, including the recent HSBC bond use case where they saw meaningful advantage to using Quantum to improve trading algorithms. So some interesting things here, Pat. This seems to be more of an ongoing journey than a true inflection.

‍

Patrick Moorhead: No, I think you covered that. Great. You know, the punchline on this was having a useful quantum by 2029. Right. Focusing on error mitigation and, and, and scale. It's interesting. I'm starting to wonder. It seems like this perpetual pushing things out four to five years. If I'm, if I start seeing this anymore, I'm going to have to start really questioning this. And really. And the funny part is I am not qualified to do that. But Paul Smith Goodson on my team is and I think we're gonna have to wait for you know, 26, 27 benchmarks as interim proof points. I do think that people should start POCs now if they're not already. IBM has a ton of them going on. I've seen, you know, a lot of things going on out there. By the way, have you seen the cratering of stocks in, in Quantum recently? It's kind of nasty.

‍

Daniel Newman: I think some are down anywhere from 30 to 80% depending on the run and the, and how far out in front they got. Yeah, but they're also still up like a hundred percent. So like, you know, it's kind of like I always say, zoom out. You know, you got to kind of zoom out. But yeah, the last few weeks the zero revenue or loss making growth companies have gotten absolutely smashed.

‍

Patrick Moorhead: Yeah, I mean listen, my tiny little company is making more revenue, driving more revenue than some of these companies by a lot. Yeah, it's, it, it's great.

‍

Daniel Newman: Some of those companies had 2 billion dollar valuations.

‍

Patrick Moorhead: True, true.

‍

Daniel Newman: You'd probably take 200 million and walk away right now.

‍

Patrick Moorhead: Yeah, I'm wondering what my problem is here. So anyways, so.

‍

Daniel Newman: Were we done with that one? We're done with that one, right?

‍

Patrick Moorhead: We're done with that one, buddy.

‍

Daniel Newman: Let's, let's, we'll have a quick one here. This one's pretty quick. But like Nebius did a $3 billion deal with Meta. I mean is this just more capacity, basically instant capacity type stuff?

‍

Patrick Moorhead: Yeah. The short answer is yes. And Meta has been very open about their strategy, which is we're going to share risk with other companies and we're also, when we need GPUs, we need GPUs and Nebbys has it. They have the power, the capacity, the water and this is just an element of it. Meta has also said that as soon as they get a better idea of consistent demand, they're going to try to pull it in house. And this is what part of that bond float was to build more of their, more of their own. More of their own. So yeah, they're not just buying the GPUs, they're kind of buying the guaranteed racks and watts and power.

‍

Daniel Newman: Yeah, I think this, you know, these, a lot of these kinds of neo cloud plays and AI utility plays is going to be about borrowed capacity. It's all about the fact that they're going to build and use and you know, it's just because they can only put so much cash up and so much time and you know, they can also float some of this over and have these other companies build it, use it, expand on it. By the way, not a totally new concept, it's just being done at a much larger volume. I mean doesn't Oracle and Microsoft have all kinds of like co-located data centers around the world and 100.

‍

Patrick Moorhead: Mean all of them are outsourcing to everybody. I mean CoreWeave and Nebius and all those companies that you're shelling out there. So yeah, sorry, educating not showing.

‍

Daniel Newman: Well, you know, look, I have conviction. Conviction. I'll, I'll take my lumps. I'm actually having so much fun with my trolls now that I'm starting to troll my trolls. It's actually been quiet.

‍

Patrick Moorhead: They do it, Daniel. What a perfect way to do it.

‍

Daniel Newman: No, no, but because like I used to just purely ignore it and now what I'm doing is like troll of the day. So like I pick like one post a day or someone trolled me real good and I troll them back and see like, you know, most of them never show back up which is a bummer because I'm kind of looking for the fight.

‍

Patrick Moorhead: Yeah.

‍

Daniel Newman: But you know, it's, it's fun stuff. It's, it's all good. You know, you got to have a little bit of a sense of humor about it because otherwise, you know, these people, it'll make, it'll ruin your day. I used to really let it bother me. I don't, it doesn't just like, you know, I don't even like, man, I wrote seven books and my parents never even read my own books. I have all kinds of therapeutic issues. I don't care what some avatar on Twitter has to say about me. Like, you know, remember what you told me yesterday? You're jacked. So none of it really matters. You're jacked, you told me and I was happy about it. And I basically took that to bed and I slept like a baby. All right, Pat, listen, you know, there's a few other topics for the deco, but we've got some time constraints, we still got a decent amount of market stuff to cover. So what I want to do is to have a little debate. Everybody out there, you know, the flip. The flip is where Pat and I simulate a debate. So when we share these. You have to remember, we may or may not agree with our positions. We may, we may not. We'll talk about that at the end. But this week the topic du jour was depreciation. And whether or not, you know, five, six year type depreciations is basically accelerating, you know, is pulling forward earnings, is manipulating longer term earnings, you know, and we're also, you know, looking at the situation of is this reasonable based on use and based on the usability of these pieces of hardware? Pat. So let's have a look at this topic this week: how long before a GPU depreciates? 5, 6 years is perfectly reasonable. You're such a shill. I figured you'd be the 4 guys.

Patrick Moorhead: Now listen, if I look at. So first of all, to get a definitive number on exactly what it would be, you'd have to know the next six year roadmaps, the neck of all the chip makers, all the designers, founder roadmaps, prediction of cooling technologies and software, and really have a good model of the diffusion of data center inference across the millions of apps, different platforms. The only thing I think I need to justify this is first of all, looking at history. Now, there's only been three to four generations of Nvidia AI accelerators out there and all of them are tapped out at about 95%. Okay, a, a series, V series, H series, they're done. And even though Jensen said that H2 hundreds are worthless, his customers beg to differ. And there's a very easy explanation for that, which is that every application does not need 100,000 bleeding edge GPUs to develop a leading edge frontier model to do video creation and things like that. You do, because there's a lot of capability and a lot of innovation that has to go in that. But for everything else we really do not need this. And the thesis here is that if you look at all the enterprises, you look at all the industries, you look at pretty much every enterprise application is not going to need bleeding edge. So where are they going to get that from? Where are they going to get their GPUs? Well, they're going to use the GPUs and the systems that were either primarily done for frontier model creation or they're going to buy, you know, GPUs that, or, or even XPUs that, that, that fit fit to purpose. The other thing I want to throw out there, and this could be especially controversial, is do we really think that we're going to have a reticle limit? So H200. You see a series was a card, each series was a tray. B series is an entire rack. And sure you can buy the piece parts but for the most part once we get to let's say Vera Rubin, where the fungible AI capability is an entire rack, where does it go from there? Daniel, is Nvidia going to be selling fit for purpose complete and total data centers that are all, all created by them. So there is a potential that because of the limitations of power and the limitations of foundry that being able to hit those performance elements the highest level of performance and 10xing it every single generation, I think it's going to be a stretch. So diffusion of inference combined with the potential slowdown and, and the we're not going to be 10xing every generation I think very well supports a six year depreciation.

‍

Daniel Newman: Pat, you know, ironically, I'm going to agree with a lot of what you said, but I just think six is wrong and four is right.

‍

Patrick Moorhead: Right.

‍

Daniel Newman: Six years is just too long when you know the amount of compute and the intensity of compute lead CEOs of companies to say things like my GPUs are melting right now, these workloads are using the least amount of compute that they're ever going to need to use. And as the token windows get larger and the volumes increase, it's going to just put more strain and more demand on these GPUs. Having said that we do have good, good accord and good history that companies including Google and, and Nvidia have had GPUs that have been in the wild four or five and six years. We know that there's still V series and a series that are out there that are being used and that are being paid for at a good rate. So I don't think this is a case where we're debating whether or not a longer, longer term is okay. It's just how long the term is okay. And there's a couple of errors. And what really creates this problem for six years isn't purely just six years and purely pulling forward some of the earnings by, you know, spacing out the depreciation. But it's also the fact that in many cases these GPUs are being used as leverage. So when you add the fact that they're, they're taking long depreciation and they're using the, the GPUs, that'll likely be end of life within three or four years to then basically finance the future buying of more GPUs, it does start to create a situation where only one thing needs to happen for these businesses to implode and that is any material slowdown in demand. So as of right now, there's no equilibrium. So right now you could do a 20 year depreciation schedule and look really smart. It's just like being a good investor this year. Anybody could have bought stuff, watched it go up and felt like a genius. It's basically functioning in the bear market. It's the same thing here is, look, we have probably four or five years before we're going to meaningfully see supply and demand hit any sort of equilibrium. And again, we don't even know if that's true. But we do know for sure it probably won't be before that. In that case, what's going to end up happening is very likely a company like Core weave that's using 60 year depreciation is going to be levered up and they're going to have to, because their entire business model is going to be based on being able to offer the leading edge. And most advanced Nvidia chips are going to have to keep buying. There's no situation in which CoreWeave is going to be able to depreciate these over long term, deprecate them down for lower use cases and continue to put cash in and then continue to lever off that cash. And that's what makes this company potentially in a bubble-esque situation, despite the fact that AI as a whole is not really in any sort of demand, limited demand situation. So having said that, I think four years is more reasonable because it's a better accounting mechanism. I think it takes some of the risk off and maybe makes earnings look a little bit worse. But if you're like a Quarterly, you don't have any earnings anyway, so you may as well just look a little bit worse. Get yourself in a situation where you're not putting as much pressure on your books longer term because this is going to be a snowball for you because each generation we know is coming faster. Right now it's twice a year. You know, if you look at each mid, mid generation that's coming out from both AMD and Nvidia, it's going to be too much, the pressure is going to be too high. But having said that, in some cases there are workloads that I certainly see lasting and being utilized on older equipment for more than six years, it just creates too much risk for most of the companies that are taking this.

Patrick Moorhead: So how do you really feel?

‍

Daniel Newman: I actually think, I think based on what we're seeing, four years is probably smarter. You know, having said that, like there's, you know, we've certainly, you know, this, like there's certainly a series and H series stuff that's been running, you know, four years plus that's still running, still selling and still driving a pretty high value hourly price. But I do think the workloads are getting more intensive and I think the efficiency gets really bad on some of these most, you know, advanced workloads. And it could definitely put a lot of strain on this stuff. But I also think like, you know, depreciation's always been one of those things that like the, the, the accounting standards gives you some wiggle and you do it and you basically always do whatever your accountants tell you is the best thing. Like, you know, when my, when I, we don't have a lot in our business, Pat, but like, I don't ever really make that decision. I let the bean counters tell me like, okay, how long should I appreciate this thing? You know, so I 'm pretty sure they're taking advice. They're looking at the market, they're looking at the situation. There's enough instances where GPUs are running for more than four years and they're saying it's fine, you can do this, and you know, it's making their numbers look better. What do you think?

‍

Patrick Moorhead: I think anybody who claims they have a definitive answer is a clown and doesn't know what they're talking about. That's where, that's where I sit on that and listen, was I definitive? No.

‍

Daniel Newman: Do I get to be called a clown?

‍

Patrick Moorhead: Okay, no, no.

‍

Daniel Newman: Somebody called me a clown earlier.

‍

Patrick Moorhead: No, but there are a lot of clowns out there right now who claim they know the answer. And they just don't. So we can debate it. I prefer a sensitivity analysis, which if I could make money on it, I would, I would create it. But maybe you should get your bean counters to create one. That'd be, it'd be valuable.

‍

Daniel Newman: Yeah, yeah. All right. All fun, all good. Fun, fun debate as always. We, we survived. I don't think either of us have to put our tail between our legs. And I think we both could, could win either side. So. All right, Pat, let's, let's, speaking of, you know, bean counters and markets, by the way, let's see where the markets are at right now. You know, we're not a, we're not a pure financial show, but the markets are just, well, the NASDAQ's slightly green, the S P is up just a tiny bit right now, and the Dow is down. So after yesterday's violent selling, it looks like today could be a mixed day. But again, it's early. It's definitely early. So just wait. Maybe President Trump will tweet something.

‍

Patrick Moorhead: Yeah, I mean, listen, you know, a lot of the typical AI infrastonks are up. Micron Coherent, A lab, pure storage oracle. They've taken a beating lately. They're up.

‍

Daniel Newman: Yeah, everything's up a little. But you can see there's been a bit of vacillation. It's been up a little. Down a little. Yeah. What it doesn't look like, at least it doesn't look like that it's going to be, is another massive capitulation day like yesterday. It looks like it might be kind of a low volume Friday. A lot of selling was done for the week. But probably the biggest and most noteworthy thing coming out of the weekend was the Senate voted to reopen the government. Not fully in the sense of like everything is not agreed upon. There's still a lot of health care things that need to be solved and settled, especially with the aca. But they did get most federal workers back to work. I think it was mostly based on the planes and flights, Pat, but like, you know, kind of close. What do you think about the whole situation? I mean, you've been through more of these than I have. Because you're like 20 years older.

‍

Patrick Moorhead: I'm not 20 years older, but we didn't have a lot of shutdowns. It was the longest in history, 43 days. There are claims that there was $11 billion permanently lost when it comes to the trillions of dollars that we flushed down the toilet. You know, I don't know if it even matters. There was a claim that 60,000 private sector jobs were lost. I don't believe it. I mean, come on, give me a break. I trust nobody's data anymore. It's nonsense. I mean, the only thing I trust now is when I look at, you know, the stock market and see what the price is, right? Is it up, is it down? It's real time, you know, but there were a lot of unpaid federal workers. There was about 16 billion in missed wages, around 6,000 flights canceled. But we're back. And you know what, man? The stock market really loved it. But the cross between the AI meltdown and the government reopening, there really wasn't really a there there. So.

‍

Daniel Newman: Yeah, I don't have much to add on that. I mean, I kind of hit at the top of the show. There is some, some impact of likely companies, private companies that served government that probably had some temporary layoffs. I'm guessing as the government turns back on, they'll rehire a lot of those people if they did furlough or lay anybody off, you know, I mean, companies that's, that's, you know, it's the accordion, right? How you add, how you grow, I expand, you know, and there was a little bit of uncertainty wrapped around it. The GDP number coming down could matter, but to your point, like, you know, nobody care. The headline numbers these days don't matter that much anyways because the revisions are where all the real truth is, in data. And then the fact is no one reads the revisions, so no one actually really knows. So they'll go in, oh, we grew this much. And then two weeks later it's like, oh, they actually didn't grow that much. But, but, but I didn't read the second one. So I just remember that we grew. Or how many jobs ones were there, Pat, where it's like, oh, we added 800,000. Actually, we lost a million. There was like a 1.8 million down revision, but that came out at like 2am on Thursday. Nobody saw that revision hit the wire. And so all things up anyways. And by the way, then we have a president, you know, that will mostly debunk any data he doesn't like and at least get half the country to think it's fake, even if it isn't, you know, so what a world we live in. I watched Idiocracy again. It is really. It is really that stupid. We are really. We have really hit that point now where I do believe, like, you know, it's. What do you say? It's over, Rover that's what I like to say. So in sorry, Cisco, you know, like this was, you know, some of these late cycle earnings hitting, Pat, did the business beat, beat, beat. It was a, that's all folks. So like the rest of earnings, I know it's a lot of pump like the rest of earnings, Pat. I mean, Cisco just does what everyone else has done, which is done well. And this is another I, I like this one as like a bit more of an enterprise indicator because this isn't a company that's 100 dependent on like a few customers. Cisco has a really big breath. They serve a ton of enterprise customers. They really saw good results across most of their business. You know, the parts that aren't growing haven't grown at all in this like collaboration, I mean that's been a falling knife for a long time. But they've seen service continue to grow. Observability, which was a big bet with Splunk, continues to grow. You know, security, you know, has seen more mix shifts. The overall security business because they're seeing Splunk down. Some of the legacy hardware and network of course is doing great because you need a lot of networking to do AI. You know, they weren't massive beats, but they were solid. And the biggest thing I probably took out is that they expect to almost triple their order book for AI in 2026. So they were like 1.3 billion hyperscale AI orders and they're expecting something in Q1 and they're expecting like 3 billion in 26 of AI infrastructure revenue. So a really big uptick there. I mean, solid numbers, like I said, from another sort of bellwether company that's not just hyperscale, that also has a really meaningful enterprise footprint.

‍

Patrick Moorhead: That was a great, great breakdown there, Daniel. And I like to focus on the re ratings out there and the percentages.The re rating was up about 9 bucks a share, which is up about 11%. And I think the stock was up about 4%. But yeah, I think it's probably half cloud and half enterprise. If I look at a lot of the discussion, you know, in the, in the analyst segment, there was a lot of discussion about, there was, listen, there was a lot of campus talk and wireless talk. But in the end, hyperscaler AI demand was a huge, huge driver. A lot of discussion about security decreases year on year. Legacy products declined. Splunk mix shift to the cloud, which Chuck did a great job on CNBC Squawk talking about this. You know, once you go from a sale to a service, the revenue gets spread out more as an annuity as opposed to a one shot. So yeah, it's good to see. You know, isn't it funny once a company gets very definitive of pointing out what they do in AI, other stock moves. Huh. Seems like a recipe.

‍

Daniel Newman: Right on. So Pat, another name for a company that's done well, one that I've, I've probably been a bit challenging of, but you seem to to be a big, big, big buyer on Monday. What do you think?

‍

Patrick Moorhead: Money. Not a big buyer. Dude, they had disastrous earnings. I mean they crushed it. They crushed it. You know, they had a double beat. But their guidance was an absolute miss. Right? Straight, you know, not off, not off like in a giant manner. But if you're not thoroughly talking about AI and how that is helping you, you're going to get smacked. And whether you're SAP, whether you're ServiceNow, whether you're Salesforce or SAP or less SAP, if you don't communicate that well and the company did not communicate AI well and any hiccup about growth and the whole meme of enterprise apps are dead. Plays completely into that. And you know, Daniel, I did a, you know, big CIO round table and the, when I was in London last week and the top takeaway for me that was sort of shocking. And you know, not only did they want to dial down GSI and they were doing a bunch of the projects internally themselves, but the desire to stop paying the tolls of all these enterprise apps is just going up and up and up. And in the end I hope. Listen, I have a bunch of enterprise apps in my company. They suck. Okay names but like you want to move a column, I need to bring in consultants, right? You want to have something new on there, I need to pay a bunch of consultants. That is ridiculous. Like the age of AI where it's going to be creating the UI itself, where, where the new UI is AI. I hope those companies who are rigid go out of business, right? They deserve to. Once they get you locked in, they can keep incrementing up and up. You know, one of our JVs, right? I look through the line item and I see a certain enterprise SaaS vendor that just goes up and we get like no value out of it. And you know, aside from, I'm thinking let's just go back to spreadsheets, right? It was a lot easier. And let's put a generative AI and agentic top end on that to do what we need.

‍

Daniel Newman: Put Gemini on top of a Spreadsheet, it can do a lot of magic.

Patrick Moorhead: No kidding, man. I'm going back to spreadsheets. The good news is I've still got the spreadsheets going.

Daniel Newman: So just watch those long term SAS contracts. I can tell you for one.

‍

Patrick Moorhead: Killer.

‍

Daniel Newman: Brutal. Brutal dude. Monday's a feature. You know, I, I, you know, it's a feature that's going to land inside of an application somewhere. I mean you need project management for sure and project workflow type stuff. But like, you know, I just, I just, I don't have anything else to say about that. I mean, I've said my piece on SaaS. You know, we sort of already talked court. Let's talk Corey real quick. And we kind of already talked about this. Stock's just getting banged. You know, they were up a ton. They went down a ton. I want to pull up what it's sitting at right now because it's, I just want to look at its weekly chart. Hold on, I'm gonna pull this real quick. CRWV. I know this off top of my head. So it's trading at 76.88. It's three months down 33% from the peak, but it's up 16% in the last six. So to be clear, like it, it spiked when a lot of this, the deal making, was out there. You know, the challenge is, this is a company that really is bare metal. It is a company that has a lot of capacity. And if capacity is the game, they have the demand for the capacity. Their revenue is good, their order books good. They were very defensive of that. They had the real pullback though. The real pullback is that they're having some delays on the build out. So when you're 100% dependent on other people to build out your data centers and that is the capacity that you have for revenue, you are going to be rate limited. And I think people are coming to the realization that building data centers is actually hard. It does take a lot of time. And by the way, it's not just hard technically to actually build the racks and install them and the cooling and it's concrete, it's dirt. It's, you know, it's, it's all the things. It's physical construction, it's labor. There's a lot of things that are tied to this. So I think this is a company that's had a confluence. It had a great uptick. The stock has gone up a ton. It's pulling back on the fact. I think people are starting to look a little deeper from the very beginning of its IPO pad. I've been kind of calling that. I was concerned about the depreciation plus the levering of all of the GPUs as a fundraising mechanism because it just doesn't have enough other revenue streams. You know, like when I, when you talk about Meta or Google, like I don't care how much debt they raise. These are companies that make like $70 billion a year of cash. Like if they want to raise money against their, they have core businesses with products that are super profitable. When you're coming that this is your only business and you are not generating cash profit and you are continuing to lever up more and more and you don't have a high margin to build into. That is really my concern here. I think they're going to do fine because I think the capacity issue is real and I think as long as they build. The biggest thing I worry about in this company is what happens when the music stops. If the music stops because they have, they continue to buy, buy, buy, buy, buy, buy, buy. And then continuing to take that, and then continuing to take these long depreciations. If Burry's thesis is right anywhere it might be here.

‍

Patrick Moorhead: I think you nailed it.

‍

Daniel Newman: Thanks.

‍

Patrick Moorhead: And I think they're good for three years or as long as the gravy train continues. I just know that companies like Meta and Google want to pull everything inside because they don't want to be paying a century to anybody to do what they can do better than somebody else. So I see it as a, as a temporary stopgap and they need to come up with their long term sticky play. You know, I look at, you know, the background of a lot of people. They're not technologists. They're either from hedge funds or financiers. And that I, that's not a knock because you do need a ton of money and to be able to explain to people what you're doing. But I haven't seen the stickiness yet for the long term.

‍

Daniel Newman: Yeah, we'll leave that one there. All right, Pat, you want to hit Navius quick?

‍

Patrick Moorhead: Sure, sure. I mean, revenue up 355% negative. Well, $5.2 million in EBITDA. Nebbys is very much a technology company. They got spun out of a Russian company who, I forget their name, but they're creating real technology. They're doing ODM style work. Their cost of deployment is a lot lower. So yeah, they're looking good and they're a, they're a cloud player, bare metal cloud player who I think does have some stickiness to them.

‍

Daniel Newman: Now we only got a minute right? You gotta run. So you know we'll, we'll, we'll hit Nvidia after the print. The only thing I'm gonna say on that one is I think they'll beat. There won't be any surprises there. Do you agree or do you think there's a surprise?

‍

Patrick Moorhead: I don't think there's a surprise. I mean I haven't even seen the channel where the folks who want to poison the well come out on their supposed channel checks.

Daniel Newman: So yeah, there hasn't been a hit piece yet from the information that's correct. It could come yet. I'll just give a 30 second shout. Amat did a good job. They are still fighting the fight with China. The fact that they lost access to that market certainly slowed down. They beat the expectations but their revenue has actually declined which is an interesting bellwether. But remember this is not the company that's providing the real time. They are providing the equipment to build the future. A lot of what they're, you know is the 2nm, the 18 as et cetera is where they're 14 days is where a lot of their stuff is going and not having access to China has impacted them but it does feel like the business is stable. Yeah.

‍

Patrick Moorhead:  Daniel, what's not even plugged in to the forecast is what tsmc, intel and even Samsung are going to have to do when they move in here. And I would posit the value of asml, the differential value of ASML versus applied materials comes into question. I think applied materials with some of their, you know, non, what I would call non shrink types of technology and value outside of that including packaging. I think it's going to be a huge surprise in the future for everybody.

‍

Daniel Newman: Well remember Pat, no amat, no AI period.

‍

Patrick Moorhead: Hey, that's exactly what I said in my tweet.

‍

Daniel Newman: So I think I said something similar. Maybe I read yours. I don't know. I, I generally try to have my own BS taglines but that. Yeah, I might have stolen that from my buddy.

‍

Patrick Moorhead: I gotta, I gotta go run.

‍

Daniel Newman: All right everybody, thank you so much for joining. Pat's gotta go be famous. I'm going to go get on calls of death. This was the best part of my day. I hope it was the best part of yours. Thanks for being part of the Six Five. Hit subscribe. We'll see you all later.

‍