π
AVGO Q4 2024 Earnings Call
December 12, 2024 at 12:00 AM
π
Company
AVGO
Quarter
Q4 2024
Date
December 12, 2024 at 12:00 AM
Speakers
2
Word Count
~5,918
Transcript Content
Operator
Thank you. Our first question will come from Blayne Curtis with Jefferies. Your line is open.
Speaker
Hi, thanks so much for taking my question. It's kind of a clarification question. I thought I heard you say that AI networking revenue was 76% of networking. I just couldn't get that math right, but maybe the broader question is you've seen growth off that low point in April in AI. Can you just talk about ASIC strength versus networking, the trends you're seeing in kind of October into January? Hock Tan (CEO): Well, that's a very interesting question. Both were growing, not at the same rate, but we've been shipping, I believe a lot more of network AI connectivity, networking components in the back half of this year, compared to the first half of this fiscal year. And we suspect a lot of that will continue in the first half of next fiscal year before more XPUs, as I indicated, more of the new generation of 3-nanometer XPUs will start ramping very much in the back half of '25.
Operator
Thank you. One moment for our next question, and that will come from the line of CJ Muse with Cantor Fitzgerald. Your line is open.
Speaker
Yeah, good afternoon. Thank you for taking my question. I guess, Hock I wanted to hit on the $60 billion to $90 billion revenue range for fiscal '27 for AI. I was hoping you could speak to the mix you see there between XPU and networking. And within that construct, are you including all kinds of the customers that you see out there in hyperscale and vertically integrated consumers? Or any sort of help in terms of what you're including in that potential mix would be very helpful. Thank you. Hock Tan (CEO): Thank you. Well, thanks for the question. Give me an opportunity here to clarify and be very specific. First, on the total dollars, the revenue opportunity for us is what I call Serviceable Addressable Market, as we all term SAM. Not TAM, SAM, and it is Serviceable Addressable Market for three of our hyperscale customers. That's it. It's a very narrow Serviceable Addressable Market we're talking about. And we're talking about XPUs and AI connectivity at that scale, AI connectivity could probably β we estimate to run approximately close to 15% to 20% of the dollar content.
Operator
Thank you. One moment for our next question. And that will come from the line of Joe Moore with Morgan Stanley. Your line is open.
Speaker
Great. Thank you. I wonder if you could talk to the XPU market. How are your customers sort of reacting to some of the rack scale products from your merchant competitor from NVIDIA, how do they sort of get the connectivity to multiple XPUs inside the rack, just how does that present a competitive dynamic for you? Thanks. Hock Tan (CEO): Everyone is trying to understand how to connect a cluster using a single fabric of 10,000 XPUs or GPUs and scale that up to 100,000, 500,000, and eventually 1 million. This presents a new challenge in terms of architecture. When you set up these racks, you engage in what's known as scale-up, and as you connect racks together, which is necessary to reach numbers like 1 million or even 100,000, you refer to it as scale-out. This architecture is continuously evolving. However, I believe that our hyperscale customers have, at this stage, figured out how to achieve this. As a result, we have a roadmap that aims to expand from a 100,000 to a 1 million XPU cluster based on a similar architecture over the next three to four years.
Operator
Thank you. One moment for our next question. And that will come from the line of Harlan Sur with JPMorgan. Your line is open.
Speaker
Hi, good afternoon. Thanks for taking my question. Hock, I know the team has been putting out an AI guide for fiscal '25. And I appreciate the multi-year sort of SAM opportunity outlook. But for this year, can we look at what your customers on the networking and custom accelerators are thinking about from a data center CapEx spending perspective? So for example, our latest roll up is that the top four cloud and hyperscalers are going to grow their CapEx, 35%, 40% in fiscal '25. I would expect that your AI business would sort of closely mirror this trend, maybe even think about it as a base case when we think about Ethernet taking share from InfiniBand, ASICs growing faster than merchant GPUs, maybe the profile of your AI business can go even faster than the CapEx trend, either way, plus or minus, is that how we should think about the growth in the AI business roughly in-line with CapEx growth trends of your large cloud and hyperscale customers? Hock Tan (CEO): Harlan. No, it doesn't. I mean, I think the hyperscalers tend to give you an overall CapEx numbers. I'm not sure they really break out between what's AI and what's non-AI out there. And clearly, the spend in AI outstrips the spend in non-AI even on the CapEx. And so no, I won't necessarily stop at that.
Operator
Okay. Thank you. One moment for our next question. And that will come from the line of Stacy Rasgon with Bernstein. Your line is open.
Speaker
Hi, everyone. Thank you for taking my question. I have a tactical question regarding the software pushouts into Q1. Should we expect those pushouts to roll off as we move into Q2 and the latter half of the year? What are the implications for the software shape for the year and gross margins? I'm speculating that in the latter half, software might be slightly weaker compared to Q2, while XPU could be stronger. Should we anticipate the gross margin trajectory from the current elevated base to ease as we reach the latter half of the year? Any insights you can provide on the software shape concerning the pushouts and their impact on revenues and gross margins would be appreciated. Hock Tan (CEO): It's simply a minor setback. I believe you're overanalyzing this project. It's just a small issue; when you look closely, you can see the differences in growth between Q4 and the reacceleration in Q1. That's all there is to it.
Speaker
Should Q2 be lower because you had push out into Q1 is what I'm asking. Hock Tan (CEO): No, it doesn't. Not Q2. No. I don't think it will have a material impact on the rest of fiscal '25.
Speaker
Do you think software can kind of hold at these levels or even grow off of these like $6.5 billion levels as we go through the year? Hock Tan (CEO): I'm not giving guidance. I might remind you for the rest of the year. I'm just giving you guidance for Q1. But I'm just telling you your analysis is kind of defective.
Speaker
I've been told that before. Okay, thank you Hock. Hock Tan (CEO): Thank you.
Operator
Thank you. And one moment for our next question. And that will come from Benjamin Reitzes with Melius. Your line is open.
Speaker
Hi, thanks a lot. Great quarter, Hock. I wanted to ask you about the $60 billion to $90 billion with a bit more clarity. Previously, you mentioned a cumulative total addressable market from your customers. Is this a run rate total addressable market or a cumulative one? Should we take the $12.2 million, add some growth for the next two years, and consider it in that way? Or should we look at a share of a $75 billion total addressable market and your revenue yield? Also, could you clarify that you're not including those two new customers? Do those customers have the same $20 billion to $30 billion total addressable market each that the current three do, or do you believe they are smaller or larger? Sorry for the multipart question. Hock Tan (CEO): No, you asked a question that was similar to an earlier one, but I'm happy to clarify. The $60 billion to $90 billion figure is not referring to cumulative SAM or TAM anymore. Instead, we are presenting a specific destination or milestone, which is three years away, in 2027, possibly extending slightly into 2028. We anticipate that the deployment of those large-scale AI clusters, each operating on a single fabric to handle large LLM models, will generate between $60 billion and $90 billion during that one-year period across all three clusters. Regarding the second part of your question, itβs possible that at least one of them may fit that criterion. However, I advise against assuming that one plus one equals two in this context. We do not view those figures as validated in our model as customers, so please refrain from adding that to the $60 billion to $90 billion in SAM that I mentioned.
Operator
Thank you. One moment for our next question. And that will come from the line of Ross Seymore with Deutsche Bank. Your line is open.
Speaker
Hi, thanks for letting me ask a question. I want to talk about the cash return side of things, a great job on the dividend increase. The other 50%, is fiscal '25 a year where you're going to still be paying down debt? Share repurchases in the mix? Or Hock, you mentioned that VMware, the integration is largely now behind you? Usually that puts you on the prowl looking for deals. Is that something we should, in general, think about or the regulatory issues that are still a concern? Just trying to figure out what that other 50% is going to go towards this year. Hock Tan (CEO): To begin with, yes, the other half of the cash that will be generated, which is beyond dividends, will have limited uses. We have typically stated that one use is to strengthen the balance sheet for potential acquisitions. However, since we are acquiring large companies, it's fair to say that this 50% cash might not be sufficient. As Kirsten mentioned in her prepared comments, the most likely use of that cash will be to pay down debt. We do plan to allocate part of that 50% of free cash flow, which is not designated for dividends, to reduce our debt, especially given the size of the debt we have assumed since acquiring VMware. Kirsten Spears (CFO): Yes, Ross, it's Kirsten. We want to focus on reducing interest expense. So we'll go after those term loans. So yes, the focus will be on paying down debt.
Operator
And one moment for our next question. That will come from the line of Vivek Arya with Bank of America. Your line is open.
Speaker
Thanks for taking my question. Hock back to AI. What do you think is the SAM in 2024, so we can get a baseline view of what your $12.2 billion in sales represent and is your assumption that you maintain the share, right? You grow it? Or what happens to that share over time? And then sort of related question to that is what happens to your semiconductor gross margins if AI grows, right to such an extent? Because you gave us a mid-single digit for non-AI and I'm wondering if AI gets to be such a big part of semis, what happens to gross margin. So both kind of the baseline of what SAM was this year and what happens to your share and margins over time? Hock Tan (CEO): That's a very insightful question regarding our expectations for the baseline in three years, specifically concerning our three key customers. I estimate that for 2024, the revenue will be around $15 billion to $20 billion, which will help us grow towards the $60 billion to $90 billion range. As for margins, I would advise not to focus too much on gross margin, because while it may decrease in the semiconductor sector, the critical factor is that revenue will leverage significantly against the investments we make to generate it, leading to improved operating margins from our current position.
Operator
Thank you. One moment for our next question. And that will come from the line of Harsh Kumar with Piper Sandler. Your line is open.
Speaker
Yes. Hi. First of all, guys, huge congratulations on successfully integrating VMware so much ahead of your time frame. And Hock, I had a two-part question. Is there a simple dollar metric that we can think of for network attached to XPUs, for example, is $1 of networking to $1 of XPUs? And then for my question, in one of your posts, you talked about sovereign data centers and VMware. I guess my question is, is there a play for Broadcom outside of the software piece? In other words, are you noticing that sovereign guys are wanting to use XPUs or are they strictly sticking to merchant silicon? Hock Tan (CEO): Let me address your question in reverse order, starting with the easier one. The sovereign players are similar to the enterprise market, which primarily consists of merchant solutions. These sovereign entities typically lack the ability to develop not only the hardware but more crucially, the software framework necessary to enable transistors in hardware to interface with high-level languages, which are essential for the operation of large language models and AI applications. This will be accomplished primarily through XPUs. Regarding your first question about the ratio of AI connectivity and networking to XPUs versus compute, this is a dynamic figure due to changes in cluster costs. However, certain ratios are worth noting. A key aspect to consider is the distinction between scaling up and scaling out. As we grow into a larger single fabric cluster of XPUs or GPUs, the importance of scaling up becomes increasingly significant. The ratio we discuss as we scale up tends to increase almost exponentially. Currently, networking accounts for approximately 5% to 10% of AI content in silicon, but this is projected to rise to 15% to 20% as we approach clusters of 500,000 to 1 million XPUs or GPUs.
Operator
One moment for our next question, and that will come from the line of Toshiya Hari with Goldman Sachs. Your line is open.
Speaker
Hi, good afternoon. Thank you so much for taking the question. The $60 billion to $90 billion SAM forecast for fiscal '27, Hock, I'm curious if you guys have a view on the TAM. So just want to know how big the SAM is as a percentage of the total opportunity set? And then my main question is you talked about going for leading market share within your SAM, which makes sense. I assume you're not assuming 100% share. So, the value of that $60 billion to $90 billion that you won't be capturing, is that a function of some of your hyperscale customers wanting to capture value internally? Or is that always having a backup or a second source? Is there a low-margin business that you just simply won't pursue? How should we think about the part of the $60 billion to $90 billion that you won't be going after or won't be capturing? Thank you. Hock Tan (CEO): To address the broader question, I don't have a clear picture of the total addressable market. My focus isn't on macro trends; instead, I concentrate on individual customer interactions. I try to understand their overall plans, including their product development and consumption patterns, which informs our serviceable addressable market. It's a bottom-up and top-down approach. I only have insights into the customers we are closely collaborating with. Regarding market share, I can't provide specific numbers, but it's evident that there's a vast market opportunity with space for many competitors. Our goal is to capture our fair share, and we are well-positioned with leading technology in this field. We boast a superior combination of technology for XPUs and their connectivity. Our extensive silicon technology resources at Broadcom put us in a strong position with our key customers. We have been cultivating deep relationships with their engineering teams for some time now, working on a multiyear roadmap to help them achieve their ambitions. Our strengths lie in areas such as silicon design, packaging design, and optical technology.
Operator
One moment for our next question, and that will come from the line of William Stein with Truist Securities. Your line is open.
Speaker
Great. Thanks for taking my question. I want to add my congrats to all the great results this year and for the quarter and outlook. But Hock, this is sort of a stunning turn of events in the last year with what we've been accustomed to thinking of as a sort of mature, slow growth business at its core with all the M&A tacked on to it. And I wonder with the sudden acceleration of the organic business, given your exposure to these capabilities in ASICs to bring AI technologies to customers, does that change your interest level in M&A? And does it change your focus area of potential M&A going forward? Thank you. Hock Tan (CEO): No, it doesn't. We remain open, as we always have been, which has been a fundamental aspect of our strategy and business model for the last 10 years. We are always interested in adding valuable franchise assets to our portfolio, whether they are in semiconductors or infrastructure software. As long as these assets meet the rigorous criteria we have set, we would always consider acquiring them and incorporating them into our portfolio. So, no, it hasn't altered our perspective at all.
Operator
One moment for our next question, and that will come from the line of Vijay Rakesh with Mizuho. Your line is open.
Speaker
Hi, Hock. Great results here. I have a question regarding the AI custom silicon side. It appears that of the $17.5 billion total addressable market, you have about 70% share. If we maintain that 70% looking towards 2027, your custom silicon AI revenue could reach around $50 billion in fiscal 2027. Do we have a clear view into fiscal 2026, indicating a solid ramp to hit those numbers, potentially with our plans to produce 1 million XPUs? How do you see that? Thank you. Hock Tan (CEO): I don't have a clear enough view to share specific details with you, and we generally don't provide guidance beyond what we do once a year. However, we want to give you an idea of where this journey is leading us, particularly concerning our AI semiconductor revenue trajectory. We have officially decided to focus on AI revenue moving forward, so we want to outline the potential market for AI. While our non-AI revenue has been stable and growing in the mid-single-digit range, AI will not follow that pattern. Therefore, we are taking an unprecedented step now to define a roadmap for the potential AI market. Currently, our market consists of the customers we serve and the end markets we operate in. We are creating this serviceable addressable market, and what we can see extends to 2027. The progression of this journey with each of our customers will vary based on their adoption rates and their own XPUs, which will play a significant role in this. We expect some quarter-to-quarter variability due to having only three customers and the fact that deployments typically happen in large increments. So, my best response is that I can't provide any more clarity beyond what I've already shared.
Operator
And thank you. That is all the time we have for our question-and-answer session. I would now like to turn the call back over to Ji Yoo, Head of Investor Relations for closing remarks. Ji Yoo (Head of Investor Relations): Thank you, Sheri. Broadcom currently plans to report its earnings for the first quarter of fiscal year 2025 after close of market on Thursday, March 6, 2024. A public webcast of Broadcom's earnings conference call will follow at 2:00 p.m. Pacific Time. That will conclude our earnings call today. Thank you all for joining. Operator, you may end the call.
Operator
This concludes today's program. Thank you all for participating. You may now disconnect.
Need more earnings data?
Explore more earnings information and financial data for AVGO
Transcript sourced from Financial Modeling Prep (FMP)