On this episode of The Six Five Webcast, hosts Patrick Moorhead and Daniel Newman discuss the tech news stories that made headlines this week. The handpicked topics for this week are:
- AWS re:Invent 2024
- Intel CEO Departure
- Pure Storage Q3FY25 Earnings
- Salesforce Q3FY25 Earnings
- Synopsys Q4FY24 Earnings
- Marvell Q3FY25 Earnings
- HPE Q4FY24 Earnings
For a deeper dive into each topic, please click on the links above. Be sure to subscribe to The Six Five Webcast so you never miss an episode.
Watch the episode here:
Listen to the episode on your favorite streaming platform:
Disclaimer: The Six Five Webcast 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.
Transcript:
Daniel Newman: Hey, everybody, welcome back to another episode of The Six Five Podcast episode 242. Sorry, we missed you last week, it wasn’t without the best of intent. But Pat had a trip to fan hangover. He ate so much turkey he couldn’t get on that morning. I was there, I was ready, I was waiting, I was dressed, I was in my best garb and Pat was like “I can’t do it I ate too much turkey.” No, I’m just kidding. But hope everybody’s well. It’s been a busy week. In fact, it’s been a crazy week, a chaotic week, a nutso week in tech. Not only did we go to AWS re:Invent, which in itself is nutty, but on Monday we all woke up to the same crazy unexpected message that the CEO, or former CEO now of Intel, Pat Gelsinger had left the company. So that threw a wrench in our week. But Pat, best hour of every week for me is the hour we sit down and do this podcast. How the heck are you, buddy? I feel like I haven’t seen you in minutes.
Patrick Moorhead: No, I know, it’s great. And there’s nothing like hopping on the pod and immediately getting thrown under the bus for Thanksgiving.
Daniel Newman: It’s just the trip, man. Nobody can predict that.
Patrick Moorhead: Depreciate that.
Daniel Newman: Are you upset about that? Is this going to affect your ability to perform today?
Patrick Moorhead: There’s nothing that affects my ability to perform. As I said in the green room run up on this, let’s effing cook. Let’s go.
Daniel Newman: We’re cooking.
Patrick Moorhead: Let’s jump into this thing.
Daniel Newman: All right. We got a really great show. We’re going to talk Amazon, we’re going to talk about Intel and what happened there. But there was a lot of earnings this week. I know we have that earnings palooza period, Pat, but this week was crazy too. Pure Storage, Salesforce, Synopsys, Marvell, HPE all had earnings this week. We’re going to hit all of those. We’re going to talk about AWS re:Invent which, by the way, it could be a show in itself. You’re going to have to check out all the other Six Five covers because we did a bunch of episodes there, they’re going to be coming out live. We were there, our team was there. I think there was 30 people across our firms that were at this show this week. So that was great as well.
Now, before we do proceed, reminder everybody, this show is for information and entertainment purposes only. Stop it. While we will be talking about publicly traded companies, please do not take anything that we say as investment advice, even though it will be great. Pat Moorhead, when I host the show I never call my own number first, but I’d want to throw you under the bus for that too. Let’s talk about AWS re:Invent because there was so much there. And no matter how much you talk there’s going to be so much more to talk about.
Patrick Moorhead: I mean, listen, I like to spend max two days at an event. Spent two full days. We were in on Monday morning and left Thursday evening, landed about 8:00 at night. We all kept ourselves busy, AWS kept us busy. Their ecosystem kept us busy as well. I mean, there were literally hundreds of announcements boiled into … Across four days. The main keynote was literally three hours long. Yeah, let’s try to boil this down. To me there were six announcements that I want to dive into. The perception is that AWS is behind in enterprise generative AI. We can debate that, we can discuss it, we have on prior shows. But I just want to jump into the announcements here and maybe we can wrap this up. Maybe we’ll go back and forth on this. I’d love to start off with the silicon but I shouldn’t. So here were the top announcements for me.
Daniel Newman: Why not?
Patrick Moorhead: So, first of all, SageMaker, which is Amazon’s end-to-end machine learning developer tool, they integrated data into the pipeline. And this is a big deal. It’s all about simplicity. They created a … Called SageMaker Unified Studio, they added a lakehouse, and tools like SageMaker Catalog for data governance. That’s huge. The next one they did is Q Developer. Sorry, I didn’t move it here. Hang on a sec. Q Developer, think of that as Copilot, think about that as your chatbot, your base for agents. But three announcements that I just thought were super cool were essentially code conversion and setting conversion to A, move you off of VMware. That was one. Super interesting that the prior week AWS did an announcement with VMware, mainframe, and then going from Windows to Linux.
You and I sat down with the general manager and the leader of that group, asked him some very hard questions. We talked about customers, the reality. I thought his answers were really good, right? You and I have both seen these transition tools before, and quite frankly they haven’t been that good. I love the openness which was, “Hey, this isn’t one button push yet.” Whether it’s VMware, mainframe, or Windows environments, these all go through systems integrators. But I do fundamentally believe if there were a great use for generative AI it’s these transition tools.
Let me dive into chips. First announcement, there’s nothing new about Graviton. Graviton has been out for a while. What was new and what was astounding was that 50% of all new compute instance capabilities, not in the number of announcements but the number of chips is Graviton, okay? So that means 50% of what’s coming in the back door of their data center is ARM-based Graviton and not x86 Intel or AMD. That is phenomenal. I’ve chronicled this … Graviton even since it was Nitro, right? Graviton1 was actually a Nitro card. But now you’ve got two socket capabilities, you have scale up, you have scale out, and you have literally every type of workload with high-performance storage that you can do on it. I think it was nearly every customer that they have has … They’re doing some sort of thing with Graviton. That’s amazing.
And I really think this puts the X86 folks on notice here. Trainium2 is not new but it did go GA which was pretty incredible. And they also brought in this training of two instance called UltraServers. And that’s essentially 64 of these chips with a high speed interconnect, 83 petaFLOPS, quadrupling memory performance networking capabilities of a training of two instance. There was a thing called Project Rainier which is an EC2 UltraCluster with hundreds of thousands of Trainium chips in a partnership with Anthropic. You and I sat down with literally the design engineer. When he said hundreds of thousands he’s like “Multiply that by a lot,” okay? I don’t know if that’s half a million Trainium’s, Dan, I don’t know if that’s 600,000.
Daniel Newman: It was more than 200,000.
Patrick Moorhead: The biggest cluster I think now is xAI that has 100,000 nodes, right? Truly amazing. I’m going to shut up here and give you some … But maybe we loop back and I can talk again.
Daniel Newman: I mean, there’s no way in a Six Five format we could hit everything, it’s just not even within the realm of possibility. I guess I want to think about this through some themes. You and I had a number of sit-downs, and people out there can catch that through the Six Five. But one of the big themes here was all about the transition from sort of this AI exuberance to AI practicality. How do we remove hallucinations? What can AWS offer there? What can AWS do to make unstructured data more available and automated for utilization in gen AI workflows? I thought that was really interesting. They had Amazon Bedrock data automation. That was a great thing for that exact purpose. They offered a lot of advanced RAG features. Because companies that are going to want to implement AI are going to want to use RAG, they’re going to want to use Knowledge Bases GraphRAG. And so they did that with their new advanced RAG features.
A couple other things that I noted was they were very focused on accuracy in their AI. Because his was an AI-centric show. I mean, they certainly hit on some other things. What is it called? Automated reasoning, Pat. So they have something called automated reasoning. And they’re saying that it could catch 100% of hallucinations. I mean, this is what makes gen AI not work is that we’re still not confident all the time. And the other thing they did too, Pat, that I think you probably appreciated, I certainly did, was this new idea of distilling models. We keep talking about-
Patrick Moorhead: Yes. That is a good one, yes.
Daniel Newman: We keep talking about how we’re going to go to smaller models, but how do we get from these largest models to smaller models in a way that is more … I guess that you could say is more fluid, it takes less overhead. And that’s what they’re doing with distilling models, basically, down to only using the size of model and the amount of parameters required for the model to perform within the context of how you’re trying to have it perform. And guess what everybody? When you do it that way you reduce the amount of energy and the amount of cost exponentially. It’s great to run these giant frontier models. And it’s really nice to talk about trillion parameters, and million GPU clusters. But the average business wants to be able to run on a handful of very small clusters, they want to be able to fine-tune, optimize a model, access all their unstructured, structured data, be able to attach it to a model that maybe has very specific use case capabilities that’s designed for their business, and then have it perform with minimum hallucinations. And that was a big feeling of what I got.
The way I’d almost describe this is this was the event where AWS was very clear that yes, maybe we didn’t get out in front of AI as well as some other companies, maybe people think that we’re a little bit late to the party, but realistically from Sage to Q to their new Nova models … Which, by the way, interesting, Nova versus Anthropic, this seems to be a thing. They’re all invested in one and building another. And, by the way, separate from Titan so they still … Is that there is a full set of suite, a full suite of solutions for everybody out there that wants to do gen AI and it can all be done in Amazon AWS, in the cloud. The only other thing I’d say, Pat, is I got a real sense that managed capabilities are going to be very important. And this, by the way, is where I see non-Nvidia silicon, for instance, starting to take shape the same way AWS has really done well in serverless to drive to Graviton. You just want a fully managed infrastructure solution to run your workload. Does half a second versus 2/10 of a second of latency on a query for a token for a language or video output change anybody’s life?
This might be where Trainium starts to get momentum. Because where I see them succeeding is when they can write the software for something that’s going to be done in high volume, that can be done on their software, that could be done at a lower cost. And they were very focused on that. They were very focused at their partnership with Nvidia is very strong and intact. But you have no doubt that they’re going to go down this path. And for these managed workloads, of course, they’re going to want to use Trainium. It’s going to be just way better economics for the company. I could go on for a long time. But those were a couple of things that caught my attention, Pat. I don’t know if you want to boomerang back on anything here or we want to jump over to-
Patrick Moorhead: Listen, they were great comments. All companies have their own models is … I think when it comes to potential cost and ownership, but also the potential when models are undifferentiated, right? And then if they’re undifferentiated then what you want to do is you want to hit cost. A great commentary, Dan. Do you think AWS moved the needle on this?
Daniel Newman: Matt Garman was saying to the press, “Looking for some needle moving.” I mean, there was certainly a lot of interest. I can’t tell you how many press calls we fielded. Now the week was a little blurry because of our next topic. I mean, I think they moved the needle on silicon. But I think more importantly they moved the needle on services. Where AWS shines is services. And a lot of these services that make data more available, unstructured data more valuable, databases more optimized … You and I have talked about this endlessly, almost at nauseum. Gen AI only works when the data works. Those solutions to make … To expose the right data to your applications is the key to making this stuff good.
Patrick Moorhead: Yeah, man, I got to tell you, I’m as skeptical as ever of gen AI for enterprise. I’m just going to lay it out there. And this isn’t because of the Amazon event it’s us going to Amazon, it’s us going to Microsoft, us going to … And actually talking to the CIOs and then talking to partners. And it gets back to data, right? I firmly believe that generative AI is going to be domain-specific. And that’s not domain-specific to the data nerds or the developers. What I mean is inside of, let’s say, your CRM system or your ERP system. I’m wondering if Enterprise SaaS is going to make the big rebound but we’ll probably talk about that a little when we talk about Salesforce. I’m done on this topic.
Daniel Newman: No. There’s some reasons to still have some skepticism. A cautious optimism maybe is another way of expressing it. All right. So the main event, right? Everybody wants to know what is going on at Intel. Only maybe more so than Jaguar’s interesting ad campaign was Intel’s CEO leaving for Thanksgiving and then coming back on a Monday, but not coming back on a Monday, on a date marked Sunday. So on the Sunday following Thanksgiving did Pat Gelsinger retire, resign, leave? Was he forced out? There’s probably more than enough speculation of what we think on the record, I don’t think we have to spend a ton of time on that here. There’s links in the show notes. CNBC, Bloomberg, Fox Business, Yahoo Finance, our team, ourselves we’ve been everywhere on this topic. I’ll put the TLDR for everybody out there that didn’t look and read everything that we’ve already had to say.
There’s absolutely a 0% chance that this was just a woke up on Friday after Thanksgiving with too much tryptophan and Pat calls up the board and says, “I’m leaving on Monday.” Something came to a head. Bottom line is there’s a lot of speculation, some rumors are rising to the surface. We are here to try to provide our best perspectives. We’re also not here to give … Or lend credibility to bad rumors or even unconfirmed rumors. We, of course, as analysts I think have earned some right to speculate at times of what’s going on. But look, I still believe deep down, and this is pretty obvious from my commentaries, is this is a foundry-led CapEx and investment-led decision of how do you carry forward on the foundry business? How do you keep throwing this much capital into it? How do you do it without a customer that’s not Intel, or now named Intel products? Or at least a customer at scale that’s not named one of those things. I think there’s a lot of external pressure for a split. There’s a lot of external pressure for a split. And Pat has been steadfast that that is not something he’s interested in doing. And so my guess is that has something to do with it.
You and I had the chance, I think both separately, to briefly sit down with the new co-CEOs, David Zinsner and Michelle Johnston Holthaus. The meetings are under NDA. I know we teased out there and everybody want to know. But I did promise to give some perspective. And the perspective that I have is that, like every large company on the planet, there are no options closed off right now. Meaning that what we’re hearing and what we’re hearing from the leadership is really business as usual. They’re continuing down the path of products in a separate governance model for foundry. The plan is to continue to push ATNA and to put great products out on ATNA. Intel products largely. To continue marching towards that next node for ATNA which has some advantages in areas like mobile.
From where Michelle sits on the product side, in that new role of CEO of Intel products, clearly that’s a very … People will argue with me about this but that is a very healthy business. If there was no foundry and they just worked with TSMC … And, actually, we’re seeing this right now with Lunar. They’re able to put out a very good quality part. The world asked that question, “Well, if the products can work with TSMC and put out a part are they” … “Could they just be another AMD? Could they be another NVIDIA?” I think that’s a question. But let me be very clear, part of what makes Intel very unique has always been its IDM, has always been its capabilities to build a fully integrated design and then manufacture these parts. And this is something I took away from the meeting. I still believe the world needs a strong Intel and the US needs a strong alternative to foundry. My take is, there is a goal to come out with a plan. Now, could it be a spinoff, Pat? Could it be a new external capital come in? I did a lot of research on that by the way. There’s a lot of speculation that they can’t split, and they can split. But if they split and they want to keep the CHIPS Act money they have to commit to doing the exact commitments that they made to get the CHIPS Act money.
Patrick Moorhead: There’s no ownership, 51% in there, Dan?
Daniel Newman: Now this is how I read it, I want to be clear, because I’m not the attorney, I’ve not read all the legal contracts. But the way I read my research was that if the assuming partner or the assuming investment group is willing to continue on the exact same commitments that were made by Intel in order to get funding, they can assume it. Now, there are also national security requirements so it can’t be … Huawei couldn’t do it. It couldn’t be some Chinese company. I had people, Pat, asked me, “Would TSMC assume it?” And the answer is no. TSMC, it does not work. The whole idea of Intel being the national champion foundry was that Intel would be a US-based company that would add leading-edge capabilities, that would manufacture chips for leading-edge fabulous designers, that would be Broadcom, Qualcomm, Nvidia, and AMD, so that if … By the way, if anything ever happened between China, Taiwan … If that risk ever went overboard we wouldn’t be without our stuff. And, by the way, when I say our stuff, I mean, first and foremost it’s security, it’s servers, it’s data centers, national security, all that stuff. And then, by the way, Pat, it’s cars, phones, PCs, and all the things that we need to run an economy.
I think people sometimes don’t take the threat of the China, Taiwan situation seriously enough. Even if we have tons of great TSMC fabs here in the US, if something happens between China and Taiwan that will be impacted here in the manufacturing here as well. And also, Taiwan’s always been a bit hesitant to give their most leading manufacturing and bringing it here. And I know there’s conversations about that. So my take here, Pat, is we still need a solution. There’s still belief within the ranks at Intel that they can be the foundry of the US but there’s a lot of proof that needs to be done. And, by the way, there’s a lot of capital that’s going to be required. And the CHIPS Act money is just not enough. And I think that was sort of an insult to Intel, it was an insult to the whole situation that so many dollars went to non-US-based companies off the CHIPS Act. And I think that might’ve been part of what started the sideways relationship between Pat and the board, going back to the Lip-Bu Tan, was that why did Intel not get more? What is not being done? So there’s a lot more here. I’m going to let you chime in, maybe boomerang it back if it makes sense.
Patrick Moorhead: So first off, Pat had a tough job coming in, right? The company was three to four years behind in process transistor tech, and I would say two to three years behind in design. To call this a turnaround is an understatement. Pat had a lot of accomplishments under his tenure. And I do believe that he’s the only person on the planet who could have gotten this into motion and gotten most of this over the line, right? He did do five nodes in four years. I didn’t believe that was humanly possible. And it’s not just money. He got the money to do this which, of course, dragged the stock down. So he did this. We can debate the quality of ATNA. I mean, ATNA doesn’t go into HVM for another six months, but the fact that it’s booting silicon is a good sign. Booting is not representative of yield or defect density rate it’s just that it’s good. Things could fall off. I mean, things could go poorly we just don’t know yet. And on the design side, there wasn’t a single shipping design that was a distributed design they were all monolithic. AMD had been doing monolithic designs for three or four years. So those were the accomplishments.
Now, the biggest issue that I can put on Pat and the leadership team there was, how could they not have a tops Monster, low precision fielded in three to four years? We saw AMD do this with a lot less resources. Why couldn’t Ponte Vecchio be refactored from a high-precision FLOPS monster to a low-precision tops monster? AMD did it, folks. AMD went and they won all of these national labs designs, turned around and now it’s a $5 billion business. How could Intel not do that, right? So that hangs on his head. The dye had been cast before with Ponte Vecchio and on Gaudi. Gaudi is a good product, a very good product. Challenges is that the hyperscalers built their own, for the most part. And if they’re going to put work into the software they’re going to put it into their own silicon.
Let me move around to the conversation we had with the co-CEOs. While it was an NDA conversation, everything we talked about was very consistent with what came out at the UBS conference. No change on ATNA, no change on product roadmaps or timelines. Why the company couldn’t come out with a press release and say that, I have no idea. This is probably the worst CEO transition announcement I have ever seen in my entire career. And the only thing that comes close is when the AMD board, the inept AMD board, fired Dirk Meyer from AMD, basically a design God. And why did they fire Dirk? Because Dirk didn’t get them into the smartphone business. And the board was stinging that Nvidia won a couple of words at that year’s CES. Made no sense. And I believe that firing Dirk Meyer led to what was the … What ended up- It could have been the downfall of AMD. And it went into a very, very dark place. With that said, hey, 18 A and product roadmaps being on time, new milestones hit every week, okay? I think in UBS they talked about a major milestone hitting soon. I didn’t look through the whole transcript of what that was.
Here is something I am seeing. To this day I have no idea exactly why Intel and the board got sideways. Was it death of 1000 cuts, right? I’ve heard lack of product focus, too much focus on the foundry. I’ve heard not enough progress on the foundry, and you had laid that out. But with the new management team, I am hearing customer focus, product focus, and hardcore execution mindsets. Could we see a restatement of schedules that say, “Okay, we only publish 90% schedules?” I don’t know. Intel did that before. And, by the way, Lisa Su, when she came in, she reset everything. So I feel like we’re getting to best products. By the way, MJ is CEO of products. Hey, I’ve heard some people say, “Oh, MJ, she’s not a PhD.” Look at all the product leaders. We’ve got Lisa Su, we have Jensen, right? And maybe I’m biased because I do not have a technical degree. If you surround yourself with the best technologists but you inherently know and have a better feeling of what your customers want, and I have … You and I both have a lot of executive communications. You really want to talk, don’t you?
Daniel Newman: I was just going to laugh. What degree did Steve Jobs have?
Patrick Moorhead: It was an arts degree I think.
Daniel Newman: I don’t even know if he finished. Anyways. It keeps going.
Patrick Moorhead: Where was I on this? What I have heard of a criticism of Pat is he just didn’t listen to his customers. And again, you and I talked to the executives of all her largest customers, I hear very positive things about MJ. She deserves a chance to drive products. To be blunt, I think if she were running products before we would have a low-precision tops monster. Versus something that esoterically looks good on paper. I’m going to give this a chance. Pat is the only person who could ever get to the company where it is now. Quite frankly, those people who are just casting overall dispersions, you’ve probably never done this before, okay? You’ve probably never worked for a semiconductor company or you’ve never worked for a customer of a semiconductor company.
Am I passionate about this? Yeah, I’m pretty passionate. Does this give me flashbacks of my 11-year tenure at AMD? Heck yes, it does. Anyways. Tune in. Hit me on social media, hit us up on social media with any questions you have coming aside of it. What I loved about our conversation with the co-CEOs, it was consistent with the UBS. But they just needed to be more direct. Just say it. I hope they can do this. I mean, their stock is down 16% on the week, okay? Yeah, 16% on the week. This rollout was a failure.
Daniel Newman: The initial bump, I think people thought there’d be an announcement to follow quickly. And then they took a breath and then they went, “Holy crap, this is going to be really hard.” Now we should be cheering as a … As people. Unless you’re short then you’re just an ass. But the point is, we should be cheering for Intel to succeed. We need competition in products and we need availability and scale in production. There aren’t a lot of options. Pat M, little Pat, as big Pat used to call you, I’m glad you pointed out the accomplishments. Because I do think in many ways that the way this ended was probably very unjust to almost a four-decade career. And someone-
Patrick Moorhead: Pat’s such a good human. He’s such a good human.
Daniel Newman: Someone that gave so much of his self to the business. Like you said, people just don’t understand the hard decision sometimes of having to make changes, make cuts. I got asked twice this week, in two different programs, what would Musk do? And, of course, when you’re callous and completely callous to these kinds of things, you could go in and cut 50% of the workforce and not blink. That’s a different kind of leader. And yes, that probably would drive more returns to shareholders. But cultures matter too. It’s not right or wrong. Anyways, we could beat this one. This is another one that could be a whole show, we could go for a long time. There’s going to be more here, it’s not over. There’s a lot of work to be done over at Intel. And I can assure you both Pat M, little Pat, and I will be commenting a lot on this in the coming days, weeks, and months. Stay tuned. All right, buddy, we got to keep going. These will be a little faster I think. We got to hit four earnings so we’re doing seven five today.
Patrick Moorhead: Yeah, we’ll work with this.
Daniel Newman: Five earnings starting with Pure Storage.
Patrick Moorhead: Listen, Pure Storage is up 22% on the week. And after their announcement they were flying. Was it based on the fact that they did beat, beat, and raise? No. It was actually because they filled in some of the blanks of this big hyperscaler that they’re going to be doing business with. They had talked about this I think three quarters ago of this major hyperscaler that they were going to get. They couldn’t give details and the stock didn’t move, right? And this is not to be confused with CoreWeave. But literally, in about a year they will have a gigantic installation in one of the top four hyperscalers. It could be AWS, it could be Google, it could be Microsoft, it could be Oracle. My guess is that it is going to be Microsoft. But doing a mass deployment with direct flash. Hyperscalers don’t like to bring anybody else’s infrastructure in. By the way, my second guess would be Oracle. And we can debate this.
I just think this is huge and the stock market agrees. Should this drive a 20X percent increase? I don’t know. But doing something with a hyperscaler is huge, particularly if they’re going to standardize on that storage. I get the feeling that this is going to replace all hard drives. What everybody listening might not understand is that most of the storage in a hyperscaler is on hard drives, right? This is probably QLC. And it comes in and replaces it with all the benefits of density. And this is the thing that people don’t fully grok either. The amount of density, when you take out hard drives and move in dense SSDs, is absolutely incredible. And you can push that power. You can add more GPUs, you can add more compute, right, in the same square footage. It costs you less for energy. As Charlie said, when I had talked to him, right … Or I guess we talked to him. Did we talk to him at the same time?
Daniel Newman: We did.
Patrick Moorhead: “Your choice is you can sign up for more nuclear reactors.” Which I know was tongue in cheek but in a way reality.
Daniel Newman: By the way, it’s really funny that you say that. I had a conversation, I can’t disclose who but a major CEO from a big chip, and he was saying that, at some point, the optionality is going to be build more nuclear reactors or figure out a more efficient way to do this. Everything can’t be these big monolithic GPU superclusters. You can’t power these things, Pat. We have a lab performance business and we’re doing some serious AI testing, and just smaller clusters. I mean, you’re talking hundreds of thousands a month in power.
In one of our testing facilities, Pat, we’re going to be probably spending somewhere around that just to keep the … So you look at how that scales. Because we’re not even talking the biggest stuff, we’re talking about what we would call something in the enterprise realm. Where is this power going to come from? Every part of the stack from software efficiency to distilling models to using the most efficient infrastructure for storage to network to putting AI out at the edge, Pat, all these things are going to be super important if we’re going to actually be able to realize this AI thing. And I think at some point we’re going to end up getting to the … Coming to this head on how fast can we actually implement all this based on limitations that are really outside of what we can do with AI, and GPUs, and software but really the infrastructure gap. I think it’s a good opportunity.
I mean, Pure has just also just been one of those head down, steady Eddie … They are growing in the hyperscaler which I think excited a lot of people. Let’s not minimize all the growth they’ve been able to accomplish in the enterprise. They’ve always been a very customer experience centric, easy-to-use centric. They’ve made that big hard pivot over the last couple years to more subscription. One of the probably slightly weaker spots in it was on Evergreen on there … It was only slight and it’s still showing growth in the right direction. So across the board it was really good. But, Pat, you nailed it on the head, why are they up 22%? Just all the shares rotated out of Intel and into Pure. That’s not what happened, okay? What did happen was people are seeing the light. There we go, there we go. Look at that, it looks like a little mountain.
Patrick Moorhead: Look at that thing.
Daniel Newman: You know what? For sure when you see a line like that that goes up, it means I’m not holding the stock. Neither did Marsha.
Patrick Moorhead: I don’t know, you’ve done pretty well with SoFi and you and your-
Daniel Newman: Well, you know me, I can’t invest in anything I really know about so I’m always in these names that are … It’s like the casino. Which by the way, Draft Kings I love that stock because I think people are degenerates.
Patrick Moorhead: By the way, my policy is, as you know, direct investing in a company that I’ve signed any NDA with.
Daniel Newman: Great. Same. I learned that from you boss.
Patrick Moorhead: Hey buddy.
Daniel Newman: Where’d you learn how to do that? I learned it from you dad. Anyone remember that commercial? Anyways. It wasn’t about investing. All right. Let’s move on to the next topic here, Salesforce, or should I call it Agentforce?
Patrick Moorhead: Agentforce.
Daniel Newman: Agentforce. We all know it’s been an interesting and exciting couple months. Marc Benioff screaming into the ex-abyss at Microsoft for some areas that do warrant consideration. I’m not going to say right or wrong because I think there’s some … It’s not universal. But the challenge of going as fast as some of these companies are going is the data governance is complicated. And when you try to put generative AI on top of every data set across every platform and make it … There is opportunities to create risk. Which, Pat, this goes back to your sort of skeptical comments that you made about AWS. Not about AWS at AWS about enterprise AIs.
This stuff’s really hard. It is really hard to do this. To get all the data, make it all available, exposed, and utilized. And then, by the way, make it all compliant. Because even what you can do here in the US with an agent or with a assistant or a Copilot or a automation you can’t necessarily do in Europe, you can’t necessarily do in other parts of the world. So doing this, and then doing … Dealing with the sovereignty of data and stuff like this is really hard. We’ve been evaluating this very closely. There’ll be more coming out very soon. They missed on earnings per share but they beat on revenue, and they skyrocketed, okay? So how does that happen, right? Well, it happens when you make two claims out loud that people really wanted to hear. The first claim was that Agentforce is on track and the early adoption is strong. Yep, there it is. I also don’t hold that one. Congrats Salesforce employees and longs, you are wealthier today than you were before their earnings.
The second thing is that you announce is an up guide for the next year. You and I have had this conversation, I’ve written pieces and made suggestions that enterprise apps, as we know them, may be dying. And I’m not pulling back my belief. They’re not all going to die. What’s going to happen is you’re going to see consolidation and you’re going to see smaller SaaS companies, enterprise apps, companies, are going to become features. They’re going to become deprecated data services that are going to have bigger apps, agent-driven apps that are going to be pointed towards those deprecated data services to do agentic things. I know really complicated.
If you think about agents as a bunch of different workers with a very specific set of tasks and then they need to be orchestrated. Say you close a sale, and then you need to start a project, and then you need to deliver a project, then you need to invoice a customer. The way it goes around. You can have a bunch of different agents doing this. Well, Salesforce basically said, “People are happy with what they’re seeing with Agent Force so far and that I’m raising guidance for the next full year because we’re getting strong traction with Agentforce.” That’s the rotation, that’s the bet. We could pretty much talk about everything else but that was all that mattered. All that mattered was that people feel that this pivot, this very hard pivot, this very enthusiastic version of Marc Benioff that has been all over social, basically, challenging the competition and saying that he’s got a better mousetrap, is saying that it’s so far so good, they’re on track. And that’s why we saw this huge bump this week.
Patrick Moorhead: Just to put some numbers behind that. Marc had said that they had won 200 Agentforce deals, and I think, in the first month of availability. We talked about it as a ground transformation of digital labor. And I started to dig into that a little bit, I read the transcript. One thing that really popped out was they’re pretty unique pricing model, Daniel. It’s not dollars per seat it’s like dollars per conversation. The list price is $2 per conversation, you can buy that in bulk. But I thought that was really instructive and cool. The only thing that could be better, and maybe the Accenture’s of the world will put a layer on and charge, is charge by the outcome, right? Find a quality metric to be able to do that. Yeah, in call centers, you … They do arrive at a cost per incident. And I just thought it was fascinating way to do pricing. I’m wondering if it’s cause and effect the … Marc waging war on Copilot, or as he called it clippy and these 200 deals. Or they’re just completely unrelated. I have no idea.
The other interesting thing that came out on the call was this consolidation into a new core. And to me, this is the back-end work that you need between their data platform, and then the all air front-end platforms, to be able to more easily move around. It’s interesting, I thought they already had this when we went to their big tent event a few months back. Maybe that was just a reiteration of that. But I think I need to do and go a little bit more work on that. The way that I think this is going to play out, Daniel, is, the first generative AI implementations are going to be inside of specific data domains, right? It’s going to be in ERP, it’s going to be in CRM, it’s going to be in Salesforce automation, it’s going to be inside of HR. There’s a lot of work that goes into crossing the streams, right? Whoever’s watched the Ghostbusters Crossing the Streams. I have met with so many CIOs out there and I’m like “Give it to me straight, where’s this going?” And they’re just like “We don’t know. We’re not moving everything into the cloud, we are not going to do that. We’re going to move a little bit of it. No, we’re not moving our data.” I think they are going to move some of the data up into the cloud.
But it just reinforces to me that this is going to be a domain-specific world with some … With a little bit of crossing data domains. I just think that the enterprise SaaS, I don’t think they’re dead in the water. As we’ve seen with every major transition in IT, everybody tries to get rid of what they had before and do it on their own, but very infrequently does it work out. The public cloud is 17 years old. And, Daniel, how much data is still on-prem and on the enterprise edge?
Daniel Newman: Somewhere around 30. Or sorry, on-prem 70 to 75 I believe is the range.
Patrick Moorhead: That’s right. Are mainframes dead, Dan?
Daniel Newman: Not if you ask companies that make mainframes. But now I’m curious.
Patrick Moorhead: No, no, no, no, no.
Daniel Newman: They’re still very relevant, especially in transaction processing.
Patrick Moorhead: I think if you have perfect data, you can put an agent, or you can put an LLM on it and you can do that. But you know what? Nobody has perfect data. It’s ugly. I asked one CIO, “Hey, tell me, couldn’t you do this, just put something on SAP? And he’s like, “The data schema for SAP” … “And no, we don’t have 27 different instances of SAP in my company we have one, we’ve already done the consolidation.” But the data is so hard to work off of it’s almost impossible. Anyways. Interesting stuff.
Daniel Newman: All right, we got three more. I am supposed to be on another call in five minutes but we love you so much I’m going to go a little bit long to take care of our great community, Pat. Let’s keep crushing it. Hey, let’s go, let’s cook, let’s cook. What’s next? What’s next? What’s next? Synopsys.
Patrick Moorhead: Synopsys, right, had a beat, beat with record earnings but Wall Street didn’t like the guide. A couple things going on there. I think there was an extra work week in Q1 ’24 that won’t be in Q1 ’25. I thought it would get figured out the day after, it did not. There was also some skittishness about China. I think the company’s taking a very conservative approach to it as it should. Gosh, the details in the call are amazing. If you want to know about the future of semiconductors, and even platforms, go read the transcript, right? Design wins on new technology like PCIe 6, CXL 3, UCIe, ADAS, 800-gig demo discussions. Read it, it’s pretty amazing. There’s a lot of discussions seen spent on alignment with Arm, right? Even CSS was discussed and fully integrated. And as we saw from the 50% AWS just on data center. These guys are cooking. Final comment. Ansys is not a done deal yet. It has passed pretty much every hurdle but one out there. I think it’s a huge opportunity. I like to call this the melding of micro design with macro design. And you sprinkle AI and it’s an accelerant to both. That is the value proposition. It took me about 12 seconds to figure out what these guys were doing and who would benefit between Synopsys and Ansys and I think it’s going to be big.
Daniel Newman: Yeah, this one is pretty straightforward. I continue to get asked all the time about where are … What are the other plays in AI? Did I miss Nvidia, am I too late? I don’t believe there’s another GPU that can compete. The software companies are being too opaque. Well, here’s a play. I mean, this is really a play. You think that we’re going to make more AI chips? The two EDA companies that drive most of this is going to be Synopsys, it’s going to be Cadence. Synopsys is even more diversified, it’s got a very substantial IP business as well. It’s got some very advanced AI features that are used to help companies speed up the chip verification design.
If you’ve been following the cycle, remember three and four-year compute cycles for data center, meaningful compute cycles, now being condensed to one. And we could argue maybe even below one for some things. It’s going to come down to having the tools to speed up the process of designing the digital design RTL. Which as we all know you can get 60 or 70% of the way through verification very quickly, but getting from 70 to 100 is very important. And, of course, getting those highest yields is always going to be … We see how even BS rumors about low yields can crush stocks. Well, companies like Synopsys are an enabler for this. I really like the company in its periphery to all this AI action. Similar to ARM in some ways and different in some ways because it’s actually diversified.
I do think the one other thing that sort of struck the company, Pat, besides China is just people waiting on this Ansys deal. There’s been a bunch of sort of concessions made over the last year. These big deals are hard. And this is just one more indicator of how do bigger deals get done. Is the Trump administration going to change anything? Pat, you asked me a great question the other day is, even if the US has friendlier rules what does that do to the rest of the world where you can’t get anything out of England or the UK, you can’t get anything out of China right now? Getting China to approve anything is really, really hard. Is the big deal environment back on? Is it game on in tech? And I think the answer is still very touch and go. It was a good quarter for Synopsys but there are still some question marks around it. I really do like it as an AI play because of its … You just can’t make chips without them, it’s just not going to happen.
All right, Pat, so let’s hit a couple more. Marvell. By the way, we’re going to be next week in San Jose area, Santa Clara, wherever Marvell is, and we’re going to be getting their update. Now, this is a company, Pat, that was really, really crushing it through the pandemic. And then, all of a sudden, as some of the legacy semiconductor areas like communications started to fall, industrial out started to fall, traditional networking, they really saw a somewhat sizable downturn. Well, guess what? Marvell is back. Pat, show the picture.
Patrick Moorhead: I’m getting it, yep.
Daniel Newman: I’m just waiting for you to pull it up for … So I can look at another eye-watering cliff.
Patrick Moorhead: Eye-watering cliff that you and I cannot participate in.
Daniel Newman: And we do not participate in. But look, we can talk about all the different parts of this business, this is very straightforward. This company is one of two in the world right now that is meaningfully enabling the design and build of XPUs, or what we all like to affectionately call custom AI chips. They’ve been talking for a number of years about cloud-optimized silicon, this was what they were talking about. Now, an announcement this week about a partnership with AWS was out there. Of course, they never really fully say who’s making chips for who. Maybe there were some new chips that Amazon announced, or future chips, as well as current new ones that are going into high volume and being utilized by Amazon. There’s one called Trainium. I think it would be a fairly safe bet that there’s an influence and a … And some sort of alignment with a company that we’re talking about called Marvell. And this is really, really exciting.
So you look over the course of a year, how has Marvell, basically, found its way back? Its core businesses of two years ago haven’t found their way back. But what they’ve been able to do through accelerated data infrastructure and cloud-optimized silicon is, basically, double revenue in a year in their data center end market by being one of two companies able to provide the key support for all these hyperscalers. And, by the way, other companies interested in building homegrown silicon. And this is a trend that’s only going to continue more into the future. So you saw them go from 556 million in the same quarter one year ago to 1.1 billion in revenue in the same space.
Now we’ve heard the numbers from Broadcom in the area of $10 billion a year of custom AI silicon. 10 to 12 silicon and networking. Well, in some ways you can think of Marvell, in many ways, as a mini Broadcom. I know they wouldn’t necessarily appreciate that exact vernacular. But I look at it more as the opportunity. You look at all the hyperscalers, all the demand for custom silicon … And it’s not just AI chips it can be network offload, it can be all these other important chips that you’re hearing these cloud providers build that are used to enable them to build their own infrastructure. Marvell is going to be aligned in a lot of areas for connectivity and, of course, for compute. Pat, I know there’s a lot of other numbers in there but that’s my read on this. This was all about that data center number and it looks really good. The AWS partnership, very promising.
Patrick Moorhead: Not a ton more to add. What I’m going to do is try and simplify this when it comes to the AI. They operate merchant, of course, but increasingly this custom. And think of Amazon, and Microsoft, and think of Marvell as driving that forward. And, by the way, it’s XPUs but it’s also custom networking and custom security. I believe that the Microsoft announcement the last … Two weeks ago had a Marvell custom HSM chip, in addition to them being behind Maya and possibly Cobalt. On AWS, yeah, Trainium 2. Think about in the future, 50% of the new CPUs coming into AWS are Graviton. Imagine if 50% of the new AI acceleration across training and inference are going to be Trainium.
And, by the way, I don’t think there’s going to be new Inferentia. For those who are counting, Trainium 1 was also Inferentia2. We did not hear a Trainium 3. Sorry, an Inferentia3 come out. I do believe that they’re going to be focusing in on one chip. But imagine if 50% of that went to Marvell. And there was a press release right before re:Invent that talked about a … The continuation, the lengthening, the strengthening of the strategic collaboration between the two companies. I think I’ll end this just by saying, I think the best is yet to come with the company. I mean, there’s two companies who can do what they do, Broadcom and Marvell, right? That’s it. It’s much better to have two competitors in a gigantic market and a growing market than having 10.
Daniel Newman: No, I think that’s absolutely right. As I said with Synopsys, another really interesting periphery play. But not even peripheral just a really interesting play on AI. We’re down to last one, dude. We did it. Do you remember when this was supposed to be a 30-minute show?
Patrick Moorhead: No. It was boring, it was bad. It’s a better show if we can-
Daniel Newman: A better show if we go longer. Actually, everyone out there let us know, do you want Pat and I to go all Rogan? Do you want us to do three hours on Friday mornings? Because I’m thinking of traveling less and just podding more. I don’t know, what do you think? Three hours everybody?
Patrick Moorhead: Yeah.
Daniel Newman: Text us and let us know. You want three hours of us just rambling on about tech all morning?
Patrick Moorhead: Let’s do this.
Daniel Newman: And if you’re going to be mean just send your tweets to Pat because I don’t care. All right. Let’s wrap up here, buddy. We both got to chat to CEO Antonio Neri, HPE. Are they growing? Are they growing?
Patrick Moorhead: They are growing. Yeah, they are. By the way, in the right places.
Daniel Newman: This one’s you, buddy.
Patrick Moorhead: Oh, my gosh. I am so sorry, folks. Sorry. I mean, HPE-
Daniel Newman: We went to seven so now we’re all confused.
Patrick Moorhead: I know. No, HPE knocked out a … Cranked out a real banger. I mean, record revenue, record operating income, big, big cash with a solid guide in there. This was tough in the sense that they had a total fricking banger last quarter. By the way, for the week HPE’s up 12.5%. Here’s another stock we can’t invest in and we can all cry about. And what are investors looking for? I mean, listen, GreenLake as a service is important, but what investors want to see is what are they doing in AI. So servers related to AI up 32%. To $4.7 billion with increased profitability. Had a conversation with Antonio Neri, as I think you did, and you talked about an improved mix, services, and some pretty good timing. AI systems, which is this catch-all for everything they do in AI including networking, up 16% to 1.5 billion and a $3.5 billion backlog. Hybrid cloud up 18%. A lot of their software, a lot of their storage is embedded in here. Edge off 20%. Listen, that entire market is in the toilet, right? Edge is not doing well. And it will come back. But it’s just not a whole lot going on, there’s not a whole lot of carrier investments. Edge investment has moved to more consolidated AI investment for this round. I always like to say, “Hey, is it market-inflicted or self-inflicted?” GreenLake ARR, 48%. I mean, GreenLake is doing well.
Daniel Newman: Is that it? Are you done?
Patrick Moorhead: Absolutely. I’m done.
Daniel Newman: All right. No, I had two, three things I think that really drove it. And look, this is never going to be a parabolic type company, it’s very much a steady as she goes. It’s been a steady hand by CEO Antonio Neri. He really hasn’t wavered from his strategy. And at times it feels like well, those little memes are poking it with a stick at the stock like do something. But I mean, it’s been nothing but consistent. I mean, two things that he’d really committed to when he came in, one is GreenLake, growing that business and moving everything as a service, and I think that’s been prudent. I mean, multi-billion dollar, bigger, and rapidly growing RPO. 48% growth. And if you look at how that growth compounds over years you’re going to have a really, really healthy subscription business which, by the way, the competitive OEMs have really not done.
Now, I know you can argue that some of the financial service models and the way some of these other companies do their services they do have really big numbers, they’re really big companies. But I’m talking about really an on-prem SaaS offering. They’ve done a really good job of executing that vision and scaling. And Pat, look, these managed full stack, out of the box, AI, data solutions, this is something that they’re really doing well. I think they’re doing well, and they’re making available, and they’re sort of choose your own adventure in terms of size as long as it’s within these four parameters. And so the early indications are the partnership with Nvidia is going well, they only launched it in September. They announced it in June at the Discover but they launched it in September and the momentum’s already good. The backlog is growing really proportionally with other large OEM sellers. And they seem to be a little bit more focused on high margin. So their margin at HPE has been pretty well protected.
The last thing I’ll say from Antonio is, he was really, really reiterating innovation. And that’s really through the services and the way they deliver what they do. And the last thing is the Juniper deal, it’s still on track, still expected to go through in January. That was a big one, it was a long one, it’s a complicated one. I think that one might play out really well. While edge network is somewhat on pause, Pat, enterprise, and branch, and networking, for what I expect to be the future enterprise AI, is going to be a bigger opportunity. It’s not big yet because it-
Patrick Moorhead: It always hits after the core.
Daniel Newman: Right. If you want to say that there was a somewhat thoughtful vision when that deal was made. All these workloads, all this inference at the edge, and at the branch, and lower power, all this stuff we talked about earlier is going to be required to be done on this enterprise infrastructure. It’s got a long tail, it’s got some good opportunities, some good growth. Pat, it was a good week so I’m cooked. We said let’s cook. I’m fully cooked, I don’t know about you. Hope all of you out there enjoyed this episode. We covered a lot of ground. And we’ll be back next week. You and I are off to San Jose, we’re going to be at a Lattice event, Developer Conference. Lattice Semi FPGA’s. Come on, where are my nerds at? And then we’re going to be at Marvell’s Industry Analyst Day. We’re heads down, there’s a lot going on. We’re planning CES, we’re planning for Davos. Six Five is going to be a Davos this year, yeah, baby. But somewhere in between we’re planning a skiing trip, right? We’re planning a skiing trip.
Patrick Moorhead: It’s a bestie trip, folks. If you disliked pictures of Dan and I embracing, and doing things together, and dressing the same, definitely stay off the internet starting December 26th.
Daniel Newman: You and I did dress the same accidentally on two different of the three days this week.
Patrick Moorhead: So weird, dude, it was so weird.
Daniel Newman: It was super weird. The marriage really is starting to play out.
Patrick Moorhead: It is. No divorce this year, folks, we’re going to go into ’25.
Daniel Newman: All right, everybody. And that’s it for this episode. Thanks so much for tuning in. Hit that subscribe button, be part of our community. We appreciate all of you. Time to say goodbye.
Patrick Moorhead: See you next week.
Author Information
Daniel is the CEO of The Futurum Group. Living his life at the intersection of people and technology, Daniel works with the world’s largest technology brands exploring Digital Transformation and how it is influencing the enterprise.
From the leading edge of AI to global technology policy, Daniel makes the connections between business, people and tech that are required for companies to benefit most from their technology investments. Daniel is a top 5 globally ranked industry analyst and his ideas are regularly cited or shared in television appearances by CNBC, Bloomberg, Wall Street Journal and hundreds of other sites around the world.
A 7x Best-Selling Author including his most recent book “Human/Machine.” Daniel is also a Forbes and MarketWatch (Dow Jones) contributor.
An MBA and Former Graduate Adjunct Faculty, Daniel is an Austin Texas transplant after 40 years in Chicago. His speaking takes him around the world each year as he shares his vision of the role technology will play in our future.
