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:
- Adobe Q4FY24 Earnings
- Broadcom Q4FY24 Earnings
- Oracle Q2FY25 Earnings
- China Antitrust Probe Into NVIDIA
- Broadcom & Apple Chip?
- Marvell Industry Analyst Day
- Lattice Semi Developer Conference
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:
Patrick Moorhead: Welcome back to The Six Five. We are live, not fake live, not recorded, but live, live. Dan and I are both here in Austin. It’s great to be back. You would think that we would all have been on our last business trip. That’s true for me, but not for Dan as he has a-
Daniel Newman: Almost.
Patrick Moorhead: … one day in and out. So how are you doing bestie?
Daniel Newman: Can I even do that? I can’t do that online. I’m doing well, man. I think I can smell the end. I know when we got home and you’re de-playing, you kind of smiled and said, “That’s it. Last one…” Although there is a informal strategy trip that might be taking place somewhere in an undisclosed bunker in the mountains over the next several weeks where there might be an old fat guy that hasn’t been on pair of skis or a snowboard in 25 years tumbling down the mountain. And if you see that, it’s probably me.
Patrick Moorhead: How about that? Dan, we are staying in different rooms though, right?
Daniel Newman: The internet seems to have its opinion about that. So let’s let them continue to wonder. Because I mean, sometimes I get cold and lonely at night and you and I have that same love language. I joke.
Patrick Moorhead: Yeah, you are like a little teddy bear; big teddy bear.
Daniel Newman: Big teddy bear. Very warm, very warm.
Patrick Moorhead: So hey, we’re moving into the holidays here, but there is a lot of stuff going on. I think we’ve got a good show for you. We’re talking about Adobe earnings, Broadcom, Oracle, and also a couple of topical things that happened. Rumors of a China antitrust probe into NVIDIA. We’ve got rumors of a Broadcom and Apple chip, and who knows, maybe it’s a Marvell and Apple chip. Dan and I hit the Marvell Industry Analyst Day. We’re going to break that down. A lot of content there. And we popped in before the Lattice Semiconductor Developer day and met with the C-suite there. Shot a couple videos, but before we dive in, just want to do a quick reminder. We do talk about public companies. Don’t take anything we say as investment advice. If you looked at my portfolio, you probably wouldn’t take that advice anyways. Neither Dan or I do direct investment in any company that we’ve signed an NDA with or that is a client. With that said, let’s dive in. Adobe, did we see some AI action here, Daniel? What’s going on?
Daniel Newman: Oh, man. So Adobe, good quarter, Pat, but the problem nowadays is a good quarter is only as good as its guide and the company had a really solid result this quarter. It was on the right side of history on both the earnings and on the revenues. Everything’s moving in the right direction. It has some great revenues in its digital experience, some great growth in its document cloud, multi-double digit percentage growth that – in a really great way. But they guided below expectations, and the market puked on it. I mean literally puked on it. Now Pat, let’s talk about what the… that’s going on here. This week we saw some stocks rip under the idea of a measurable AI increase. Most specifically, this was Broadcom. We’ll talk about that later. People want to know how AI is influencing business. So the challenge is if you listened to Shantanu and the team, they kind of talked about AI as a horizontal that’s influencing and impacting the success of many of their product lines in their portfolio.
So there’s the creative, they built their models. They were early to visual generative AI tools. They also have lots of AI embedded in their CRM and experience platforms, as well as in their document tools. By the way, this was being done long before it was even so trendy. The challenge is: how do you account for that? So when you’re growing at around 11% and that looks like the same amount you were growing before generative AI was a thing or in those same ballpark, people are kind of looking for that incrementalism. And when you can’t really discern the incremental growth and how AI is impacting it, what it looks like is… to the outside lens is sometimes that it’s a little bit AI washing. And by the way, this has been a bit of a theme with software as a whole. The early AI trade was all about infrastructure. It was all about chips, then it was all about sort of the OEM, selling lots and lots of servers.
And I’ve said for a long time these kind of second, third network effects, which is the implementation as well as the applications is where AI is going to be explosive. But to some extent, there’s a pretty good argument to be made that hasn’t totally been able to be discerned in any sort of data, whether it’s been Salesforce, which is showing a little bit of love for service now, was able to show a little… But nobody’s like, “Oh my gosh, the business is doubled because we added some generative AI feature to our tool.” This is a combination of things. You have some companies like Oracle; companies like… them are basically saying, “We’re just baking AI into everything to keep our products sticky.” I’m talking about on the app side, “To keep our products sticky, to keep our products usable.” And this seems to be a bit more of what the leadership at Adobe is reflecting when they talk, is they’re making AI part of the product to keep customers on the product, to keep customers working more closely with the company, and keep them attrition rates down.
But I think what people want to see is, “Hey, we’re doubling our growth rate. Customers are spending X amount more per month because of the AI features that we’re loading in, and this is changing the calculus of the business evaluation. And we’re a pioneer and leader in AI.” Pat, TLDR. I like saying that. I don’t know why, is that AI features inside of SaaS applications have not yet become totally transparent in terms of how much revenue and how much stickiness they’re creating for customers. Because of that, companies like Adobe can have really good quarters when they guide down, people go, “Oh my god, does that mean AI is not… Is this another indicator that the AI capabilities and features aren’t impactful enough?” So my take, we’re going to see more transparency. I think Adobe is going to work to show more how AI is driving growth in the business. Double-digit percent growth in enterprise apps matches our intelligence of how the sector growth is expected just to the top end, if not a little bit above. So overall solid from them. But that’s kind of my external viewpoint on why Adobe, despite having a good quarter from a standpoint of earnings, revenue and growth of their businesses, maybe didn’t quite meet or satisfy the street and that was why they sold pretty hard after resolve.
Patrick Moorhead: So Daniel, I am with you on almost everything that you said. The one thing that is starting to change my mind is what Salesforce did. Salesforce was in the exact same position, single-digit growth, maybe super-duper low double-digit growth, yet they came out and gave a very good guide and attributing it to AI. And they said that they had signed up 200 incremental customers for Agentforce within a month of the product going GA, and then the stock absolutely ripped. So it’s hard for me to chalk this up as the state of the industry when we saw what Salesforce did. I know they do… they have some overlapping products, but they have more products that don’t overlap than do. But I guess, it’s fair to say that we’re going to have to wait until Salesforce’s next earnings and let’s see what service now can do inside of AI as well.
I think the other thing that I know I want to do more research on, Dan, is: what is it about marketing, not only the creative side and also on the document side and also I’ll call it the CRM side, that is necessarily different and unique? One interesting thought here, Dan, is that if you lay off marketing people, you have less seats and a lot of the ways that Adobe charges is on a per seat basis versus, let’s say, an outcome. Agentforce is priced as a dollar per transaction or dollar per chat and list prices is two bucks every time that you use it. One thing though, if I look at the core of where Adobe makes most of its money, and that’s on the creative side, and then I would say overall on the enterprise for this end-to-end content to distribution pipeline, I don’t see enterprises moving very quickly right off of Adobe. It’s the safe choice. Now, does it have threats from the upstarts like an OpenAI and more? 110%, but they just can’t be too far behind. Okay? Sora got a lot of discussion and a lot of, “Oh gee whiz,” and even the early image creation tools, but they never really caught on. Sora kind of, maybe, sort of went GA, but there’s still a sign-up sheet, so there’s no way I can even call that GA and I know-
Daniel Newman: Dude, that was crap.
Patrick Moorhead: Yeah. So yeah, I feel pretty good on the outlook for Adobe, but I’m also not blinded. I really want to understand a little bit more. I’ve got to do more research in the difference between Adobe-Outlook and Salesforce-Outlook. All right, baby, let’s go to the next topic here. Broadcom, boy, did they frickin’ rip it and I think… I don’t know, maybe we can check what percentage are they up, Dan, this morning?
Daniel Newman: I’ll pull it right now, but it’s being 16.
Patrick Moorhead: Yeah, and this is all… again, gets all the way back to the guide, because I think they missed on the revenue number, which I attribute to lumpiness in AI, but a lot of really good things came out of the call. And let’s start with VMware.
Daniel Newman: 21, 26%. They’re up $39 a share right now.
Patrick Moorhead: That is just absolutely incredible. So let me start off with VMware. Growth trajectory operating margin reached 70% exiting 2024. And you that that incremental adjusted EBITDA number that they promised at the beginning, which was $8.5 billion when they did the deal and everybody said, “There’s no way that you’re going to be able to do that”? Well, guess what?
Daniel Newman: They did it.
Patrick Moorhead: They did it and they exceeded it much earlier than the initial target of three years. Crazy. I mean, not that I’m like… I don’t root for companies, but what I love to see is when somebody says you can’t do something and then the company does.
Daniel Newman: Hold my beer, hold my beer.
Patrick Moorhead: Exactly. A little bit more on VCF, they signed up 4,500 out of 10,000 customers on the VCF. This is one of the things that people said wouldn’t happen, and not that there’s not any risk in the future for them. I mean, you have companies like AWS who are moving forward with probably transition services and technologies with Q Developer to do that, but it’s not easy. There are still a lot of chaos and disruption. I think VCF has a lot of benefits there. So let’s move to networking here. And so Q4 networking was up 45% on year. 76% of the $4.5 billion is AI networking, an AI and networking are 158% on an absolute role… sorry, in that also included XPU shipments of the top pre hyperscaler customers. Hock did say that that’s four times growth in AI connectivity revenue, which was from the Tomahawk and Jericho. It was one of the big things that got people really excited, is Hock said in 2027, quote, unquote, “We believe each of them,” meaning their top customers, “plan to deploy 1 million XPU clusters across a single fabric.” Great for XPUs and great for fabric. So I mean now we know that. And the other thing that they said, which we might be talking about a little bit later, is they’ve been selected by, quote, unquote, “two additional hyperscalers,” and are advanced development for their own next generation AI XPUs.
Daniel Newman: Yeah, listen… Yeah, Pat, good stuff. I actually think this entire result only cared about one thing. The numbers this quarter were fine. That overall number, which by the way was not really bigger, much bigger than what they presented in the Investor Day last year is what they thought was possible, running towards $12 million, 12 million… $12 billion of XPU, but Pat, these wins and the ramp on these wins, people are seeing the light. Now your most absurd sort of quick takes are going to be, “Ooh, they’re going to take out NVIDIA.” Listen, it’s not a swap. It’s not like, “Oh my gosh, suddenly this company’s going to build a Trainium,” or in this case they’re doing TPUs. TPUs have been a long time. It’s not all going to be like this. But at the same time, people are starting to see the light that there are other infrastructure companies that could really make big money on AI chips. There’s only two players, Pat. We’re going to talk about one of them later. We were at their event this week, Marvell and Broadcom, that are really doing this at scale. And between the two of them, they basically own every major hyperscaler.
Right now, Broadcom is sort of established in one… this consumer AI chip for the big consumer players. They’ve got Google, they’ve got ByteDance, they’ve got Meta. And now apparently, or likely we’ll talk about later, and maybe Apple, we’ve heard OpenAI. They’re going to do something for them. And there’s some others out there. Pat, this is faster growth market than the GPUs. Our intelligence says it’s going to grow at somewhere between 10 and 20% faster than the GPU market. I actually would say our intelligence is underestimating how fast XPUs can grow. The whisper on the street is everyone wants to have a second option. It’s healthy in a competitive market space to have competition. Competition comes in different ways. It can be like AMD and NVIDIA directly with GPUs that have the same kind of use case and the same kind of architecture, but it can also be different. And in this case, an XPU versus a GPU in these large hyperscalers where they have very specific workloads and they can optimize software, optimize the network, optimize their libraries, train the models, and then build and deploy services that can be leveraged off of these, they can do this at scale and do so in a way that’s going to be economically more viable for their business.
So whether it’s these consumer hyperscalers that Broadcom’s using or it’s these hyperscalers that are being used for cloud services by enterprises, all of them are kind of doing these two parallel work streams. One is the NVIDIA work stream, very important, very established. The other is the homegrown silicon for both compute. We’ll talk a little bit about where it’s heading with memory, storage fabric, and then applications and workloads, Pat. And so my take, this is great, but if you want to know why it’s ripping, even though the fact that they’ve really crushed the early Broadcom-VMware deal and it’s looking really promising, that’s not why it’s ripping. It is ripping because people are like, “Holy cow, this could become 20, 30, $40 billion a year of revenue.” And by the way, you want to see what other stocks ripping, Pat? Marvell. Marvell is ripping today because Broadcom is ripping today because everybody sees this is where it’s heading. Really good results for Hock Tan, Charlie Kawwas, and the Broadcom team.
Patrick Moorhead: That’s right. And we’re going to talk about Marvell a little bit later. So let’s dive into Oracle. Dan, did the AI gods shine?
Daniel Newman: It was not the best day for Oracle, it was not. Now, I think it’s important that we kind of look at this somewhat holistically. Oracle is a company that has found itself at near or floating around all time highs. So perhaps price to perfection heading into earnings. I mean, this thing was near $200. It was up high double digits year to date and had seen its the highest price in a long, long time. So what’s happened to the name? Well, I mean the bottom line is it was very sort of narrow misses on a couple of different places, but Pat, I don’t even think it’s that important to talk about those misses. Broadcom missed two and it ripped. The thing here long is it was up… I think you actually put it in here, 68% to the year. So it dropped a few points. But Pat, here’s some things that are really positive. 52% IF growth. This is the fastest cloud infrastructure growth in the industry right now, just to be clear. Is it because they’re the best and the biggest? No, they’re not the biggest, but they’ve done a really, really good job of seeing their cloud infrastructure business grow.
Their backlog is growing, RPO is what they call it in SaaS and software. They’re still seeing their earnings grow meaningfully on a year-over-year basis. They’re launching these massive mega clusters. They’re partnering with some of the biggest hyper scalers on the planet. So Pat, I think this is the company that’s performed really, really well. When I see these misses so narrow, the last thing I’m going to say on this one, when I see misses this narrow, I’m always sort of floored at how you can miss by a penny, but you and I both know that means they probably pulled every last inch in dollar and penny to try to get there. They got really close. But Pet overall, Fusion good, NetSuite good. GPU growth and AI cloud growth, good. OCI growth, good. It’s a narrow miss. This thing is ripped. It has roared, a little pullback’s probably healthy to take into the next growth. I’ve been really impressed with Oracle’s pivot over the last two years.
Patrick Moorhead: Yeah, Dan, I did say this in my tweet and it has everything to do with that 68% up for the year. What I think it’s important for people to recognize is that EPS, up 24%. RPO’s up nearly 50%. Record IaaS at 52% growth, which is well above market, albeit a smaller number. And these very respectable ERP numbers, I think is just fascinating to me. 19% on Fusion and 20% on NetSuite, that’s absolutely astonishing to me, particularly when I align this up against SAP’s growth. And SAP’s cloud growth is big, but this is all Oracle has, right, which is Fusion and NetSuite ERP. A couple of drill-downs that were interesting on the IaaS business was GPU consumption up 336%. They’ve got 65,000 node, H200 cluster that’s ready and online and being used by Meta. I think it’s really cool that OCI has become in some ways an outlet valve for many companies. I mean, Azure uses OCI, Meta uses OCI, and I think some of the big questions that people should be wondering is: how durable is that in the future? I mean, it’s great to see that, but how long do you plan in for that growth?
Azure said, or Microsoft said in the last conference call that they could see being caught up in the second half for CapEx. And I’m just wondering if that continues or how it continues for other vendors. I mean, that growth as a percentage will go down just naturally the law of large numbers. But if the dollars start really trailing off, then it could be an issue. I don’t see that happening in the next 12 months, but it’s something for people to consider. All right, let’s go to the next topic here. We talk a lot of antitrust on the companies, particularly when it comes to MagSeven. We covered a lot of Qualcomm, a lot of NVIDIA, Google, Meta, Microsoft, and Amazon. And to absolutely nobody’s surprise China is investigating NVIDIA. And here’s the way that I look at this. Whenever you are a 95% plus player in any type of market across something that is so incredibly important and transformational like AI, you are going to get investigated. I think that NVIDIA will be investigated by everybody. They’re already being investigated in the US. I think France is investigating them. We’re going to see Korea, we’re going to see Australia. I mean everybody’s… EU.
I mean, everybody’s going to get a pile on here. But remember folks, it doesn’t really… I don’t know if this even matters. I mean, I think the China fine could be, at worst case, $1.1 billion. Who cares? What is that? A week of profitability for NVIDIA? I mean, so what? I think the thing that matters is some change to the type of business they’re doing. And right now the US is curbing exports for NVIDIA into China, and I don’t even know how much this even matters in the future. On the global antitrust stage, you’re looking for a five to 10 year slog. I think the EU Intel related to AMD was like 15 years, and it’s still going. They’re still debating how much is the check, when does it get written? And the last thing I read was that thing was even overturned, overturned in the EU. And anticompetitive folks are trying to take another run at it. So anyways, I don’t know if any of this stuff actually matters here, Dan.
Daniel Newman: Yeah, this was how my week started. I was supposed to actually hop on Yahoo Finance to talk about Oracle and Broadcom earnings, but instead I spent the whole time talking about this, that I’m so exhausted in talking about… Well, first of all, this is like the fifth time you and I have had to actually cover various antitrust probes into NVIDIA in just the last year. So maybe a primer-
Patrick Moorhead: Hey Dan, I got to put just a quick pause on this. A family thing just came up, so why don’t you keep rolling and I will pop back in. My apologies on that one. Sorry, dude.
Daniel Newman: Yeah, it’s all good. So over the last, I don’t know, three quarters, I think this is probably the third or fourth time that I’ve had to talk about NVIDIA antitrust. The first thing that I would say is that the antitrust situation is going to be prevalent for the next, I don’t know, several years. When you have 90 plus percent of a market like NVIDIA does for GPUs, you’re most certainly going to see a lot of interest from regulators to figure out if the way you’ve accumulated the market share is healthy. Now, I always remind people that it’s not against the law to be a monopoly. It’s against the law to be a monopoly that abuses your power and uses your power asymmetrically. So what could that look like? Well, it could look like a lot of different things. In this particular case as it relates to China, a lot of it had to do with the Mellanox acquisition, and it had to do with the fact is when they acquired, there was a commitment made that NVIDIA would make it as easy for any other chip maker, including Chinese chip makers to use their AI chips with now NVIDIA’s Mellanox switching. And create solutions and build solutions that would be meaningful or highly performant and they wouldn’t give themselves a unique competitive advantage.
Now this is always interesting when you talk about fully vertically integrated stacks of technology because you’re always going to be able to build some level of optimization with your own technologies when you’re building in the rack, the front end, and then of course in the back of the rack in the back backend. And so pull all this together and what you end up getting is a situation where China feels that maybe NVIDIA isn’t as favorable or creating as favorable of conditions for its own chip makers. But this then leads to a second part. And then second part is China wants more access to leading edge NVIDIA chips. But NVIDIA, by the way, as far as I know, is happy to sell its most advanced chips to China. It’s a meaningful market. It falls into that sort of low to mid double-digit percent, and it’s a big opportunity. But right now, due to chip controls, the below seven nanometer isn’t supposed to ship into China at all. And therefore, most even the tuned down GPUs that have been built for the Chinese market are not available.
So I’ll summarize this kind of long diatribe about antitrust to: China is going to continue to weaponize antitrust and different probes into US companies so long as the US continues to limit the rate of shipments and the availability of advanced semiconductors into China. We saw them do it with Micron. We saw them do it with Intel. NVIDIA is a stronger company with more resources. It’s not maybe as favorable, but it does create some FUD around NVIDIA. It does create maybe the opportunity for China to try to slow down NVIDIA’s growth. And in a world where we’re trying to continue to fight in a race to be the leading AI chip makers, designers, builders, implementers, LLM creators, and everything else, that’s a fight that China wants to slow down. So anyways, we’ll move on from this one.
Patrick Moorhead: Yeah. No, I’m back, baby.
Daniel Newman: Okay, are you okay? All good?
Patrick Moorhead: Yeah. Yeah, all is good.
Daniel Newman: Ending my best, whatever it is.
Patrick Moorhead: No, no. Somebody in the family had a car accident and everything is okay, but this is a live show, so this is what we get.
Daniel Newman: I just gave a five-minute histrionics lesson on what antitrust is, monopolistic power, asymmetrics. Gave a little background on the Mellanox acquisition and the China’s desire to have Mellanox function, the different between vertically integrated tech stacks and how homogenous versus heterogeneous work differently. And then of course, I landed the plane talking about whether or not this is actually at all about NVIDIA or Mellanox. And really in my opinion, it’s mostly continued microaggressions between the United States, China. And it will be so long as AI is the front in which the economic war will be won. Pat, that’s it. I got nothing more to say about this one.
Patrick Moorhead: No, I’m glad you… I mean, the state-
Daniel Newman: It’s no-
Patrick Moorhead: The state of reason on there was Mellanox. I’m glad you brought that in. So let’s move to another chip related rumor and that is the potential that Broadcom is doing an Apple chip. Dan, is this just nonsense or is there something here?
Daniel Newman: Well, the cited… It was funny, the site that was cited through the citing of the citing I think was that tech power-up site or something was where at least one of the sites where it was being read on.
Patrick Moorhead: Oh my gosh, you mean the same one that said that the Intel yields were horrible, and then five days later said they were good?
Daniel Newman: Yeah, exactly. That’s exactly the one. So you and I have both had a little fun on that particular topic, but hey, at least they backtracked I mean, most people just never acknowledged their theories.
Patrick Moorhead: Yeah, true.
Daniel Newman: I gave them… You’ll give them a compliment sandwich. I’m like, “Here’s a compliment that you owned it. Now I’m going to punch you in the mouth because you’re morons. And then I’ll give you another prop here by talking more about your stuff that you’re publishing.” But they’re not the only ones publishing this. There’s a fairly significant consensus in the market, and remember… And by the way, this actually gets really interesting fact because you saw a report yesterday I think that Intel’s… Sorry, Apple’s got a project to take its own Wi-Fi chips and other specialty components in-house. I don’t know if you saw this. I saw it come out yesterday and meaning that they’re going to take some of that business back from Broadcom apparently. So this is like this push-pull with Apple because Apple-Broadcom allegedly have a pretty healthy… I mean, I’m sure there’s some hard negotiating that goes on between those companies, but compared to the Apple-Qualcomm relationship, much more cordial. I even think Apple-Broadcom about a year ago announced a fairly significant continued partnership around a number of different DSP, Wi-Fi, other technology that they work on together.
And by the way, the stuff that they work on together is hard. The stuff they work on with Qualcomm is hard. The stuff they work on with Broadcom is hard. There’s a reason Apple hasn’t done it. Because if anything that’s easy, Apple will do itself. They will vertically integrate. So going back to that point is because the relationship has been pretty kosher as far as I’ve been aware, there was a lot of consensus that Apple was going to tap Broadcom. Now kind of a little bit more histrionics here is that… Remember Apple didn’t go the NVIDIA path. Apple has not… I mean, I’m not saying they’re not using any, but I’m saying Apple has been pretty steadfast that it is doing most of its own training, most of its own… It’s using Google TPU. The likelihood of it 100% building a dependency to do its own training and inference on only a Google product is almost as improbable as them doing it on an NVIDIA product.
Apple is not a company that ever wants to be captive. And I think through some of the lessons we’ve learned with RF and with Wi-Fi and these other tech that they’re trying to take in that they haven’t been able to is that eventually if Apple can, Apple will. So everyone else is doing it, Meta is doing it, Google’s doing it, AWS is doing it, ByteDance is doing it, Alibaba is doing it. Of course, Apple is going to do it. So the only real question here is: are they going to do it with Broadcom? And I think the bottom line is, we talked about this earlier, we’ll talk about it again later, there’s only two companies they could do it with. And so the reason I put some merit behind this is there is history here. There is a strong history, there is a good relationship. Apple does sort of fit that- talk about Hock and Charlie, have talked about these more consumer focused companies building specialty AI chips. So the pieces to the puzzle fit. The history, the moving forward, Apple’s vertical integration strategy, Broadcom being the cloud optimized silicon provider for a lot of these consumer applications.
And by the way, Apple for whatever reason, keeps ripping to new highs. I mean, Apple Intelligence sucks. I don’t know what… It’s early. I realize I’m taking a breath here, but I’ve not seen anything compelling about this. The super cycle’s been meh, if it exists at all. Intelligence sucks. I’m not quite sure. And again, I know I’m always just Apple. I just don’t know why this of all the names is ripping so hard. I mean, they’re literally now worth half a trillion dollars more than NVIDIA in the last several weeks. This company just keeps going up, but the ability for it to vertically integrate, build its own AI silicon should be an expectation. It’s going to be part of its profit pool, it’s going to be part of how it eventually builds models and trains. And by the way, Pat, the last thing I’ll say is I don’t believe this OpenAI partnership is for the long term. I just do not buy it. I do not buy that Apple is going to just defer all of its meaningful engagements with its customer off to the OpenAI platform. I think this is a stopgap for the company and this is part of that process to doing more of their own, building more of their own. And I think in the beginning it’ll be very inference centric, but they’ve eventually got to train models because the early use cases for Apple Intelligence doesn’t meet the need.
Patrick Moorhead: Yeah, good stuff there Dan.
Daniel Newman: Thanks.
Patrick Moorhead: I mean, you pretty much drain this one for all it’s worth. I always thought it was kind of hokey when Apple Intelligence came out and they said that they had servers in the cloud that it seemed like it was a private server, not a virtualized server. And that just seemed completely weird to me and unscalable. And what was up there essentially were Apple chips, right? There was likely an AWS, as we learned about the tight coordination between the two of thems. One of my biggest questions is: where would this chip sit? Would this sit at AWS? Is Apple somehow going to get good at doing infrastructure, which they’re not? And let me tell you why they’re not good at infrastructure. Or let me just say that they’re good at infrastructure on stuff like streaming, which is if you look at where is most of the apple compute actually happen, it’s streaming content, it’s music, it’s movies, it’s apps, it’s stuff like games, and that’s just not that hard, folks. That’s really all they’ve had to do. And even backups for your iPhones, that’s pretty first-order stuff and not difficult at all.
So big question is: are they going to start hiring a bunch of know AWS, Azure, Google, Rockstars to actually create and run their own infrastructure in hyperscaler data centers? My guess is that the design of this AI chip, if true, would be very leveraged on all of its… Basically, its client-based AI just scaled pretty high. You can imagine the price that Apple is paying to do all of its training and all of its inference right now. It’s likely being done at… Well, the training, I think they were pretty clear is being done at Google. But it seems to me that the inference and actually running those workloads is coming out of out of AWS. So hey, let’s dive into our last two topics here and that was the Marvell Industry Analyst Day. We were both there for two reasons, first to be analysts and soak it in and give good advice to the C-suite. But also, we were there with The Six Five and we did a bunch of videos. I don’t think any of them have been published yet, but a big TBD on that. And here’s the way that I’ll summarize this stuff. I thought they did a great job, particularly on AI. I mean, that’s pretty much what it was all about. And COO Chris Koopmans came out there and basically set it straight.
So the shift from client, consumer telco to the data center has been phenomenal. 35% in ’21 to 71% in four years. Dan, that is hard thing to do. Now, I know that the edge in telco is down and enterprise is down, so it made it a little bit easier, but there’s nothing easy about this. And then from a TAM perspective, if you look what was presented in 2023, it was a 21. For data center, it was a $21 billion opportunity and now… And I don’t know how this aligns with your data, but for 2028, $75 billion opportunity. Most pretty much everything that was presented was about accelerated infrastructure for AI. It was the compute, it was the switching, the interconnects. And the company has shown they can add a billion dollars each year historically to that between ’24 and ’25 and then ’25 and ’26. For what it’s worth, I think that we are going to absolutely start see… I think those numbers are going to absolutely go up significantly. I think the company’s being super-conservative. I mean, Trainium 2 and likely Maya 2 are just starting to come online. And then when you look at the custom, not just XPU and networking, but also custom DPUs and HSM-
Daniel Newman: And CPU.
Patrick Moorhead: Sorry? Yeah, and CPU, they’ve got a tremendous amount of opportunity. Then you layer on optical onto that whole thing, particularly on the DCI, which is connecting data center to data center, super compelling. And then even optical compute fabric between racks, 500 XPUs within 10 meters. And the company did say, “Hey, we’re going to work on algorithms and the reliability to make optical more similar to powered copper,” but it’s super an exciting opportunity for the company. The customization just blows me away that they’re doing, and they had one killer slide showing 7, 14, 20 different elements of technologies and foundational IP that they’re customizing for the hyperscaler. This is the best description of custom data center infrastructure in the hyperscalers that I have ever seen.
Daniel Newman: You did a great job.
Patrick Moorhead: Yeah, and I really appreciated the distinction too between what are the CSP’s doing and then what’s the customer doing and what’s Marvell doing. I’ve never seen a breakout or an explanation like that. Listen, as analysts, we need to be very careful of what we say, but I absolutely believe this company is going to be a juggernaut in custom hyperscaler. The market is finally waking up even before the numbers are coming out. And that’s not normally the case. Now they are driving a lot of AI revenue today, but when I look at what this could become, it’s going to be much, much bigger than it is today.
Daniel Newman: Yeah. You hit a lot of the good items, Pat, and I think for people out there we have a great video previewing what they’re doing with this custom high bandwidth memory. I think what I liked about the event too, Pat, was they kind of talked about things that maybe others aren’t. We’re very, very focused on logic right now. Everybody’s very focused on compute and processing, but all these other parts, it just doesn’t work without it. Now, NVIDIA has kind of done the all in a box system. It’s like the Apple iOS ecosystem, it’s the phone, it’s the apps, it’s everything. It’s just built into one thing, which by the way is why the OpenAI partnership is so dumb, because it’s the most on Apple thing ever. But going back to this, the other is going to be this Android Open sort of ecosystem that’s going to take XPUs, these companies are going to build their own network infrastructure and data fabric.
They’re going to have to build abstractions like PyTorch and they’re going to need to be able to tune these things to remove CUDA from the mix so that they can program these things. And the hyperscalers are so logical to do this because they do things in such volume that they can put the effort into software to make these things run really efficiently. But then they even talked about the next thing, Pat, which is like power. So this power thing is not inconsequential. We hear about these SNRs, small nuclear reactors, that are being built. Microsoft or Google, they’re talking about building these half gigawatt, 500 megawatt. And I mean, there were some good reminders at the conference. This is a five-year plan to get to half a gigawatt. Well, so we’re already looking at racks that are taking… and clusters that are going to take this kind of power. And so Marvell is talking a lot about fiber and what they’ve been able to do in terms of building fiber for long range to be able to connect data centers, because you’re going to have to build data centers where there’s power.
So to get to these million XPU clusters, there’s a very high probability it’s not going to all happen in one physical location. You’re going to need to build a number of these things to connect them together with very little latency to do these training and clusters or to build these inference mega factories. And so that’s another part of the Marvell business that’s kind of underappreciated at time, is everything they’re doing with light as well. Because you can’t do all this stuff with copper. It’s just not possible. So they’re addressing some of the short range copper. They’re addressing some of the long range fiber. They’re addressing the networking and connectivity. And then, of course, they’re now also really playing in the – side. Everything, like I said, from a very high probability, they’re playing with Axion. Rumors on things like Maya 2. I think the partnership was very well disclosed with AWS. So they’re playing in all the right spaces, but they’re not just logic. They really are this continuum.
One last thing I’ll say, Pat, is an amazing pivot. This company is basically out of its old business. I’m not saying they completely deprecated it, but the only reason they’re doubling in growth is all this AI business has basically overtaken the fact that this is the second major pivot. They went from consumer to enterprise and now they’ve gone from enterprise and communications and automotive in these networking spaces to being one of the leaders in custom compute. And obviously, like I just mentioned, networking and everything else. And they’ve somehow propped up the business, made the full 180 pivot and now they’re in a growth mode while their entire core business from three years ago is kind of in the toilet. So it’s a pretty fascinating pivot. A lot of credit to Matt Murphy, Chris Koopmans, Raghib, and the team. They’ve just done a really good job. You’re muted, dude. You’re muted.
Patrick Moorhead: Let’s hit our next topic here. We are ripping, sorry, my dogs-
Daniel Newman: We’re ripping. It’s good.
Patrick Moorhead: … were barking. Lattice developer conference. Dan, give us the highlights, baby.
Daniel Newman: So first of all, Pat, you and I had the chance to come in early and spend some time with their new CEO, Ford Tamer. Really good guy. We did mention I think around their earnings that we had the chance, but this was the first in-person sit down. And in spirit of our partnership, this is probably, what? The third we’ve shown up for their developer conference had a sit-down with the executive leadership team, some new leadership in the team. And Pat, let me say that first and foremost, I mean, this is a company that’s really been priding itself on moving up the stack. So they were always one of the leaders. I think they said something on their Nexus, which is their low end FPGA. What did they say? A quarter of a billion units chip every year?
Patrick Moorhead: Yeah. Yeah, exactly.
Daniel Newman: I mean, the volume on this is tremendous. A couple stats about the event, they doubled their number of demos this year. They saw a huge growth in the actual customer base. And so while FPGAs don’t necessarily always get the attribution of being like a cool AI play, in every one of these designs, whether it’s these big servers that we talk about, whether it’s the next generation AI PCs, FPGAs are very commonly used for things like increasing performance, reduction of security risks, and high-volume reboot and optimizing things like sensor technologies like over-the-shoulder that takes place. The chips that are used to sense that stuff is technology like Lattice. So the company has really been focused… They really went from no developer ecosystem to a sizable developer ecosystem inside of three or four years. I think they had 6,000 developers sign up for this conference, Pat.
And like I said, huge number of demos, partners like NVIDIA. I think last year they had Meta and it’s across industries: automotive, data center servers, handsets, and PCs. They’re playing in really all the spaces. And by the way, Pat, this is another company that saw a really hard pivot away from sort of some of its core revenue areas and industrial comms consumer and has had to really make hard pivot into areas like data center servers and AI to stay relevant and growing. They have had a several, what I’d say, slow down quarters. And so the question is: how do they continue to make this pivot continue to turn the corner? But Pat I, my take… and by the way, everybody check out all the videos we posted a few demonstrations. We’ve got the executive leadership conversations, is they continue to be a leader in their space, leader in low power, a leader in efficiency. And really, Pat, as the big FPGA companies have either been kind of swallowed inside of bigger chip companies or they’ve been kind of spun off from bigger chip companies. FPGA has been very… Sorry. Lattice has been very disciplined. And I do think as the market starts to turn, they’re in a really good position for growth. So I’ll pause there and I’ll kick this one to you to see if you have anything to add to it.
Patrick Moorhead: Yeah. So Lattice is the company that makes FPGA sexy and more programmable. Working at AMD as long as I did, I was very familiar with FPGAs. In fact, we had a high-performance FPGA version of Opteron so we could do testing before we hardened the silicon. And a lot of customers use that today, but FPGAs have really shifted and one of the biggest challenges used to be their programmability, but Lattice has come along and added a bunch of solution stacks to those which makes it easier. And while they are less power-efficient than an ASIC, you can get time to market probably year, year and a half sooner. And many of Lattice’s FPGAs actually have hardened silicon in there. So they removing a lot of these objections here. And by the way, the stock ripped. Stock was up 7%. Dan, when’s the last time that we’ve seen a stock rip after a developer conference? I don’t remember the last time. 10 or 15 years ago, it’s the way that it worked, but it no longer happens like this. So why did it rip? Lowest… In fact, they were the highest of any semiconductor player on the day after of any company from a percentage standpoint.
Brought out a new low-power Nexus-II line. They expanded Avant with the 30 and 50 line. They updated and improved four solution stacks and had a much larger ecosystem at the event than they had before. And they had onstage appearances from Dell, Microsoft, Teledyne FLIR. And a company, that unless you operate on the edge you’ve never heard of, it’s called Sick. That’s a company that makes sensors and it’s the name of the founder’s last name. The thing that probably got the market interested to me is the reality of Avant, is these customers getting up there and talking about Avant. And I just think that that helped. So looking forward to tracking the company, getting to know CEO Ford Tamer, and of course, loves spending time with Esam, who is the chief strategy officer and also heads marketing. So Dan, we made it. Thanks everybody for coming along for the ride here. This was the seven, five and a half and you hung in there and we appreciate that. Have a great weekend. Dan and I’ll back next Friday, same bath time. Dan, good luck with your trip today buddy.
Daniel Newman: I’m going to shoot for a great trip and by the maybe some point, but we should talk about this inability to pick six topics. Are we going to have to change the whole show? I don’t know. It’s a big struggle of us.
Patrick Moorhead: We changed the time. I don’t know. Maybe we should get back on that. Anyways, hey, thanks for tuning in there. Thanks for being part of the community. Hit that subscribe button, try to find a chill this weekend. Don’t be like Dan and I and work the whole weekend. Take care. Peace out.
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.
