Broadcom Inc (AVGO) — Q3 2023 Earnings Call Transcript

Thank you, operator, and good afternoon, everyone. Joining me on today's call are Hock Tan, President and CEO; Kirsten Spears, Chief Financial Officer; and Charlie Kawwas, President, Semiconductor Solutions Group. Broadcom distributed a press release and financial tables after the market closed, describing our financial performance for the third quarter fiscal year 2023. If you did not receive a copy, you may obtain the information from the Investors section of Broadcom's website. This conference call is being webcast live and an audio replay of the call can be accessed for one year through the Investors section of Broadcom's website. During the prepared comments, Hock and Kirsten will be providing details of our third quarter fiscal year 2023 results, guidance for our fourth quarter, as well as commentary regarding the business environment. We'll take questions after the end of our prepared comments. Please refer to our press release today and our recent filings with the SEC for information on the specific risk factors that could cause our actual results to differ materially from the forward-looking statements made on this call. In addition to US GAAP reporting, Broadcom reports certain financial measures on a non-GAAP basis. A reconciliation between GAAP and non-GAAP measures is included in the tables attached to today's press release. Comments made during today's call will primarily refer to our non-GAAP financial results. I'll now turn the call over to Hock.

Thank you, Ji, and thank you, everyone, for joining us today. In our fiscal Q3 2023 consolidated net revenue, we achieved $8.9 billion, up 5% year-on-year. Semiconductor solutions revenue increased 5% year-on-year to $6.9 billion and infrastructure software grew 5% year-on-year to $1.9 billion. Hyperscale continued to grow double-digits year-on-year, but enterprise and telco spending moderated. Meanwhile, virtually defying gravity, our wireless business has remained stable. Now generative AI investments are driving the continued strength in hyperscale spending for us. As you know, we supply a major hyperscale customer with custom AI compute engines. We are also supplying several hyperscalers with a portfolio of networking technologies as they scale up and scale out their AI clusters within their data centers. Now representing over $1 billion, this represented virtually all the growth in our infrastructure business in Q3 this year-on-year. So without the benefit of generative AI revenue in Q3, our semiconductor business was approximately flat year-on-year. In fact, since the start of the fiscal year, our quarterly semiconductor revenue, excluding AI, has stabilized at around $6 billion. And as we had indicated to you a year ago, we expected a soft landing during fiscal '23, and it appears this is exactly what is happening today. Now let me give you more color on our end markets. As we go through this soft landing, we see though that our broad portfolio of products influencing the puts and takes across revenues within all our end markets except one, and that is networking. And so, in my remarks today, I focus on networking, where generative AI has a significant impact. Q3 networking revenue was $2.8 billion and was up 20% year-on-year in line with guidance, representing 40% of our semiconductor revenue. As we indicated above, our switches and routers, as well as our custom silicon AI engines, drove growth in this end market as they would deploy in scaling out AI clusters among the hyperscale. We've always believed and more than ever now with AI networks that Ethernet is the best networking protocol to scale out AI clusters. Ethernet today already offers the low latency attributes for machine learning and AI, and Broadcom has the best technology today and tomorrow. As a founding member of the Ultra Ethernet Consortium with other industry partners, we are driving Ethernet for scaling deployments in large language model networks. Importantly, we're doing this based on open standards and a broad ecosystem. Over the past quarter, we have already received substantial orders for our next-generation Tomahawk 5 switch and Jericho3-AI routers and plan to begin shipping these products over the next six months to several hyperscale customers. This will replace the existing 400-gigabit networks with 800-gigabit connectivity. And beyond this, for the next-generation 1.6-terabit connectivity, we have already started development on the Tomahawk 6 switch, which has, among other things, 200G SerDes generating throughput capacity of over 100 terabits per second. We are obviously excited that generative AI is pushing our engineers to develop cutting-edge technology in silicon technology that has never been developed before. We know the end of Moore's Law has set limits on computing in silicon technology, but what we are developing today feels very much like a revival. We invest in fundamental technologies to enable our hyperscale customers with the best hardware capabilities to scale generative AI. We invest in industry-leading 200G SerDes that can drive optics and even copper cables. We have differentiating technology that breaks current bottlenecks in high-bandwidth memory access. We also have high-speed and ultra-low power chip-to-chip connectivity to integrate multiple AI compute engines. We have also invested heavily in complex packaging technologies, migrating from today's 2.5D to 3D, which enables large memory to be integrated with the AI compute engines and accelerators. In sum, we have developed an end-to-end platform of plug-and-play silicon IP that enables hyperscalers to develop and deploy their AI clusters in an extremely accelerated time-to-market. Not surprisingly, in Q4, moving on to Q4, continuing to be driven by generative AI deployments, we expect our networking revenue to accelerate in excess of 20% year-on-year. This has been driven by the strength obviously in generative AI where we forecast to grow about 50% sequentially and almost two times year-on-year. Moving to wireless, Q3 wireless revenue of $1.6 billion represented 24% of semiconductor revenue, up 4% sequentially, flat year-on-year. The engagement with our North American customer continues to be deep and multi-year across WiFi, Bluetooth, Touch, RF Front-End, and Inductive Power. So in Q4, consistent with the seasonal launch, we expect wireless revenue to grow over 20% sequentially and down low-single-digit percent year-on-year. On our server storage connectivity revenue, it was $1.1 billion or 17% of semiconductor revenue and flat year-on-year. With a difficult year-on-year compare, we expect server storage connectivity revenue in Q4 to be down mid-teens percent year-on-year. And moving on to broadband, following nine consecutive quarters of double-digit growth, revenue moderated to 1% year-on-year growth to $1.1 billion or 16% of semiconductor revenue. In Q4, despite increasing penetration of deployment of 10G-PON among telcos, we expect broadband revenue to decline high-single digits year-on-year. Finally, Q3 industrial resales of $236 million declined 3% year-on-year, reflecting weak demand in China. And in Q4, though we expect an improvement with industrial resales up low-single-digit percentage year-on-year reflecting largely seasonality. So, in summary, Q3 semiconductor solutions revenue was up 5% year-on-year. And in Q4, we expect semiconductor revenue growth of low-to-mid single-digit percentage year-on-year. Sequentially, if we exclude generative AI, our semiconductor revenue will be flat. Now turning to software. In Q3, infrastructure software revenue of $1.9 billion grew 5% year-on-year and represented 22% of total revenue. For core software, consolidated renewal rates averaged 117% over expiring contracts, and in our strategic accounts, we averaged 127%. Within strategic accounts, annualized bookings of $408 million included $129 million or 32% of cross-selling of other portfolio products to these same core customers and over 90% of the renewal value represented recurring subscription and maintenance. Over the last 12 months, I should add, consolidated renewal rates averaged 115% over expiring contracts, and in our strategic accounts, we averaged 125%. Because of this, our ARR, the indicator of forward revenue, at the end of Q3 was $5.3 billion. In Q4, we expect infrastructure software segment revenue to be up mid-single digit year-on-year. And on a consolidated basis for the company, we are guiding Q4 revenue of $9.27 billion, up 4% year-on-year. Before Kirsten tells you more about our financial performance for the quarter, let me provide a brief update on our pending acquisition of VMware. We have received legal merger clearance in Australia, Brazil, Canada, the European Union, Israel, South Africa, Taiwan, and the United Kingdom and foreign investment control clearance in all necessary jurisdictions. In the US, the Hart-Scott-Rodino pre-merger waiting periods have expired, and there is no legal impediment to closing under US merger regulations. We continue to work constructively with regulators in a few other jurisdictions and are in the advanced stages of the process towards obtaining the remaining required regulatory approvals, which we believe will be received before October 30th. We continue to expect to close on October 30th, 2023. Now Broadcom is confident that the combination with VMware will enhance competition in the cloud and benefit enterprise customers by giving them more choice and control over where they locate their workloads.

Thank you, Hock. Let me now provide additional detail on our financial performance. Consolidated revenue was $8.9 billion for the quarter, up 5% from a year ago. Gross margins were 75.1% of revenue in the quarter, in line with our expectations. Operating expenses were $1.1 billion, down 8% year-on-year. R&D of $913 million was also down 8% year-on-year on lower variable spending. Operating income for the quarter was $5.5 billion and was up 6% from a year ago. Operating margin was 62% of revenue, up approximately 100 basis points year-on-year. Adjusted EBITDA was $5.8 billion or 65% of revenue. This figure excludes $122 million of depreciation. Now a review of the P&L for our two segments. Revenue for our semiconductor solutions segment was $6.9 billion and represented 78% of total revenue in the quarter. This was up 5% year-on-year. Gross margins for our semiconductor solutions segment were approximately 70%, down 160 basis points year-on-year, driven primarily by product mix within our semiconductor end markets. Operating expenses were $792 million in Q3, down 7% year-on-year. R&D was $707 million in the quarter, down 8% year-on-year. Q3 semiconductor operating margins were 59%. Moving to the P&L for our infrastructure software segment. Revenue for the infrastructure software segment was $1.9 billion, up year-on-year and represented 22% of revenue. Gross margins for infrastructure software were 92% in the quarter, and operating expenses were $337 million in the quarter, down 10% year-on-year. Infrastructure software operating margin was 75% in Q3 and operating profit grew 13% year-on-year. Moving on to cash flow. Free cash flow in the quarter was $4.6 billion and represented 52% of revenues in Q3. We spent $122 million on capital expenditures. Days sales outstanding were 30 days in the third quarter compared to 32 days in the second quarter. We ended the third quarter with inventory of $1.8 billion, down 2% sequentially. We continue to remain very disciplined on how we manage inventory across the ecosystem. We exited the quarter with 80 days of inventory on hand, down from 86 days in Q2. We ended the third quarter with $12.1 billion of cash and $39.3 billion of gross debt, of which $1.1 billion is short term. The weighted average coupon rate and years to maturity of our fixed-rate debt is 3.61% and 9.7 years, respectively. Turning to capital allocation. In the quarter, we paid stockholders $1.9 billion in cash dividends. Consistent with our commitment to return excess cash to shareholders, we repurchased $1.7 billion of our common stock and eliminated 460 million of our common stock for taxes due on vesting of employee equity, resulting in the repurchase and elimination of approximately 2.9 million AVGO shares. The non-GAAP diluted share count in Q3 was $436 million. As of the end of Q3, $7.3 billion was remaining under the share repurchase authorization. We suspended our repurchase program in early August in accordance with SEC rules, which do not allow stock buybacks during the period in which VMware shareholders are electing between cash and stock consideration in our pending transaction to acquire VMware. We expect the election period to end shortly before the anticipated closing of the transaction on October 30th, 2023. Excluding the impact of any share repurchases executed prior to the suspension, in Q4, we expect a non-GAAP diluted share count to be $435 million. Based on current business trends and conditions, our guidance for the fourth quarter of fiscal 2023 is for consolidated revenues of $9.27 billion and adjusted EBITDA of approximately 65% of projected revenue. In Q4, we expect gross margins to be down 80 basis points sequentially on product mix. We note that our guidance for Q4 does not include any contribution from VMware. That concludes my prepared remarks. Operator, please open up the call for questions.

Thanks for taking my question. Hock, my question has to do with your large AI ASIC compute offload contract. Is this something you feel you have the visibility to hold on to for the next several years, or does this face some kind of annual competitive situation because you have a range of both domestic and Taiwan-based ASIC competitors, right, who think they can do it for cheaper? So I'm just curious, what is your visibility into maintaining this competitive win and then hopefully growing content in this over the next several years?

I appreciate your question, Vivek, but I won’t be able to answer it directly because we generally do not discuss our dealings, particularly the specifics regarding any particular customer. However, I can say that in a broad sense, we have long-term agreements with our major North American OEM customer in the wireless sector that are quite similar. Our relationship involves a multi-year strategic engagement across multiple leading-edge technologies, which are essential for developing their desired products, whether in wireless or, in this context, generative AI. This engagement is very strategic, multi-faceted, and extends over several years, allowing for depth and breadth in our collaboration.

Thank you, Hock.

Good afternoon. Thanks for taking my question. Great to see the market diversification, market leadership, and supply discipline really sort of allowing the team to drive this sort of stable $6 billion per quarter run rate in a relatively weak macro environment. Looking at your customers' demand profiles, your strong visibility, given your lead times, can the team continue to sustain a stable-ish sort of $6 billion revenue profile ex-AI over the next few quarters before macro trends potentially start to improve or do you anticipate enterprise and service provider trends to continue to soften beyond this quarter?

You're asking me to guide beyond a quarter. I mean, hey, that's beyond my pay grade, Harlan. But I just want to point out to you, we promised you a soft landing, late fiscal '22, that likely '23 will be a soft landing. And as you pointed out and what, to my remarks, that's exactly what we are seeing.

Okay, perfect. Thank you.

Hi, guys. Thanks for letting me ask a question. Hock, I want to stick with the networking segment and just get a little more color on the AI demand that you talked about growing so significantly sequentially in the fourth quarter. Is that mainly on the compute offload side or is the networking side contributing as well? Any color on that would be helpful.

These elements are closely interconnected, Ross. Nowadays, AI engines for generative AI are not deployed individually; they are utilized in large clusters, as described by some hyperscalers. To support this, a robust fabric and networking connectivity among thousands, even tens of thousands, of these AI engines—be they GPUs or other specialized AI compute solutions—is essential. The entire framework made up of these AI engines truly constitutes the AI infrastructure. Therefore, our figures are strongly correlated not just to AI engines, regardless of whether we produce them or if they come from other sources like merchant silicon for GPUs. We also provide a significant portion of the Ethernet networking solutions.

Thank you.

Hi, guys. Thanks for taking my question. If I take that sort of $6 billion non-AI run rate and I calculate what the AI is, I'm actually getting that 15% of semiconductor revenue that you mentioned last quarter. Do you still think it's going to be 25% of revenue next year? And just how do I think about how you get to that number if that? So I guess two questions. One is, is that number still 25% or is it higher or lower? And then how do I get it with the two moving pieces, the AI and the non-AI in order to get there? Because that percentage goes up if the non-AI goes down?

There are a few assumptions to consider, but I'm not going to provide guidance for next year. However, I can share that our AI revenue has been on an accelerating trajectory, which is no surprise given the urgent deployment and strong demand we are experiencing. We expect this acceleration to continue through the end of 2022 and into 2023. For fiscal year 2024, we anticipate a similar accelerating trend. To address your question, we have previously indicated that we believe AI revenue will account for over 25% of our semiconductor revenue in fiscal year 2024.

Got it. Thank you very much.

Hi. Thank you so much for taking the question. I had one quick clarification then a question. On the clarification, Hock, can you talk about the supply environment, if that's a constraining factor for your AI business? And if so, what kind of growth from a capacity perspective do you expect into fiscal '24? And then my question is more on the non-AI side. As you guys talked about, you've done really well in managing your own inventory. But when you look across inventory levels for your customers or at your customers, it seems as though they're sitting on quite a bit of inventory. So what's your confidence level as it pertains to a potential inventory correction in your non-AI business, networking business going forward? Thank you.

On the first question regarding the supply chain, the products for generative AI, whether for networking or customer engines, have long lead times. These are advanced silicon products that involve the entire stack from the chip to packaging and even the HBM memory used in those chips. Consequently, there are significant constraints as we try to meet demand within these lead times. As demand comes in faster than our production capacity allows, it creates ongoing challenges. On your second point, we see what I would call a soft landing, with approximately $6 billion of non-AI related revenue per quarter fluctuating on a plateau. While growth has slowed considerably, it remains stable. Our portfolio is diverse, with multiple products in various end markets, each with its own timing based on customer needs. This leads to variations in performance across products, but overall it averages around $6 billion each quarter, which we are currently observing.

Great. Thank you.

Thank you. Just on gross margins, you had a tough compare year-over-year for your semiconductor gross margins, which, of course, remain some of the best in semis, but is there a way to think or quantify about the headwind to gross margins this year from still elevated logistics costs and substrate costs as we think about the supply chain perhaps freeing up next year that perhaps could be a tailwind? Thank you.

It’s Hock. Let me address your question as it requires a more comprehensive response. The main factor affecting our gross margin isn't significantly tied to transactional supply chain issues. While there may be fluctuations at times, they aren't substantial or sustained enough to impact trends. Instead, what drives our gross margin primarily is our product mix. We have a diverse range of products that we categorize and communicate differently across various end markets. Each product has varying gross margins depending on its application, importance, and other factors. Thus, we have quite a variety. The key factor influencing gross margin trends is the adoption rate of next-generation products in each category. As new generations of products are adopted, we have the chance to increase our gross margin. The rate of adoption is crucial because for some products, changes in gross margin occur every few years, while others have a longer cycle. As a result, we experience different growth profiles for gross margins. When we look back at 2021 and 2022, amid a semiconductor industry upcycle marked by lockdowns, behavioral shifts, and high demand, there was a significant acceleration in product adoption. This rapid adoption contributed to not only revenue but also gross margin expansion since a larger share of our products transitioned to next-generation versions faster. Currently, we are seeing a potential slowdown in adoption rates, which may slow down gross margin expansion, but this will balance out over time. Historically, our model, supported by fundamental economic principles, shows that with our extensive product line and varying product life cycles, we've experienced long-term gross margin expansion ranging from 50 to 150 basis points annually for semiconductors. In some years, like 2022, this has exceeded 200 basis points. Consequently, there are years where gross margin expansion is less, around 50 basis points. Overall, this process will be ongoing.

Thank you.

Congratulations on achieving a textbook soft landing, Hock. I have a question regarding the timing for taking off. Your lead time for most product lines is approximately one year, which suggests you can see visibility a year ahead. Are you beginning to notice growth in your backlog for that one-year timeframe? Can we expect to remain at the bottom for about a year before starting to rebound, or is that resurgence occurring sooner, or even before a year? Any insights you can provide would be appreciated. Additionally, Hock, can you clarify whether China approval is required for VMware?

It's still too early for me to predict when the up cycle will happen. Even though we have a 50-week lead time, I see a lot of bookings tied to generative AI and a decent amount related to wireless as well. So, to answer your question, it's too early for me to say definitively, but we do have a decent amount of orders.

And then on VMware, Hock?

Let me say this. I made those specific notes or remarks on regulatory approval. I ask that you think it through, read it through, and let's stop right there.

Okay. Fair enough. Thank you, Hock.

Thank you.

Yes. Thanks for taking the question and congrats also on the execution. I'm just curious, as I think about the Ethernet opportunity in AI fabric build-outs. Just Hock, any kind of updated thoughts now with the Ethernet Consortium that you're part of thoughts as far as Ethernet relative to InfiniBand, particularly at the East-West layer of these AI fabric build-outs with Tomahawk5, Jericho3 sounding like it's going to start shipping in volume maybe in the next six months or so. Is that an inflection where you actually see Ethernet really start to take hold in the East-West traffic layer of these AI networks? Thank you.

That's an interesting question. In my view, InfiniBand has traditionally been the preferred option for many years in high-performance computing, which is essentially the old term for AI. This was largely due to its focus on dedicated application workloads rather than the scale-out requirements driven by today's large language models. We are witnessing an increase in the adoption of large language models, primarily influenced by hyperscale environments. Ethernet is gaining significant traction and is already being deployed in many hyperscales. It works well alongside InfiniBand, with the effectiveness dependent on the specific workloads and applications involved. Ultimately, the decision also hinges on the desired scale of AI clusters; larger clusters tend to favor Ethernet.

Yeah, thank you.

Thank you very much. Good afternoon. Hock, I wanted to ask a question about your custom silicon business. The large customer is ramping up nicely as you mentioned. However, there are several other large hyperscale customers considering custom silicon, possibly influenced by the recent surge in Gen AI spending. Has this increased interest in Gen AI impacted your willingness to take on substantial projects with other large customers in this space? Additionally, are you open to exploring custom switching routing products for customers, or is your primary focus on merchant products in these areas? Thank you.

Thank you for the question. We have one major customer in AI engines. We're not focused on GPUs or significant computing capabilities beyond offloading computing, which is quite specialized. I want to be clear that any potential engagement in a custom program should not be interpreted as a guaranteed revenue stream. Developing the hardware infrastructure to support large language models for hyperscalers is quite challenging. Even with any engagements, those do not easily convert into revenue. To summarize, we are currently shipping custom AI engines to one hyperscale customer, and I prefer to leave it at that. Regarding customized switching and routing, that certainly exists. Some OEMs have been providing switch or router systems with their proprietary solutions and operating systems for the past 20 to 30 years. Currently, around 70% of the market relies on merchant silicon. While this may not apply to the network operating systems, it definitely pertains to the silicon. The overall message is that pursuing a merchant solution offers certain advantages compared to custom solutions, as historical performance has demonstrated.

Thanks, Hock. Appreciate it.

Thank you for the question. I think there are two significant aspects of the Broadcom story that have surprised me. The first is the potential in AI. The second is the strength of the core business, especially in storage and broadband, despite some of your competitors facing tough challenges and being in clear downturns. I’ve been anticipating a reset in storage and broadband for some time, and it seems like Q4 might be a bit softer for you. Perhaps you're referring to that reset as a soft landing, Hock. Could you elaborate on what you mean by a soft landing? Does that suggest we have indeed reached a turning point? Do you expect those businesses to be bottoming out at this point? I know you’ve mentioned previously that you've managed inventory tightly, but is there possibly more inventory being sold off in these markets or are we seeing a decline in demand here? Thank you.

Thank you. Firstly, as I have mentioned in previous earnings calls and Kirsten reiterated today, we are primarily focused on shipping according to the direct demand from our end customers. We are expanding our view beyond just enterprise and telecom to also consider the end users of those OEM customers. While we strive to get this right, it's not always perfect, but we are improving in our approach. The data you’re seeing, especially regarding broadband and service storage, might appear somewhat flat, and that’s why I emphasized considering the broader picture without factoring in generative AI. My product portfolio is diverse and serves different markets. During times like these, which I refer to as a plateau, we experience a soft landing. In this scenario, we don’t experience complete stagnation; instead, some products may be affected by shipment timing, leading to fluctuations in our results from quarter to quarter. What we're noticing now in Q3 and even Q4 relates to the logistics of customer shipments across various products. My comments regarding revenue refer to the ups and downs around a median, which currently stands at approximately $6 billion and has remained stable since the beginning of fiscal '23. As we approach Q4, it's still around that mark. While some segments may rise and others may fall, this fluctuation contributes to the overall picture. I hope this clarifies your inquiry regarding whether this is a trend or just a temporary situation. To put it simply, we are experiencing some consistent patterns around that $6 billion median over the last three quarters.

Thanks, Hock.

Thank you very much. Hock, I want to shift gears maybe a little bit here and talk about your expectations and actually indications from your customers about the integrated optics solutions that will start shipping next year. It seems to me by looking at what you're offering and the significant improvement you get over performance and size. This would be something of great interest. Is it limited by inertia or architectural inertia by the existing solutions? Or what kind of feedback are you getting? And why should we expect to see maybe because it's rather a new market for you overall? You've not been in it before. So I'm just trying to get a feel for what your expectations are and why maybe we should start looking at this more closely.

You should consider it. I made my investment, and I think it's worth your attention. We've invested in silicon photonics, which involves integrating multiple functionalities into a single solution. For instance, our next-generation Tomahawk5 switch, set to begin shipping in the middle of next year, is part of a program we call Bailly. This is a fully integrated silicon photonic switch that operates at very low power. By consolidating optical and mechanical features into an integrated silicon photonic solution, we can eliminate the reliability issues associated with traditional methods, effectively translating to improved silicon yield rates. This approach is believed to be more reliable than conventional techniques. So, why haven't more companies adopted it? The truth is, we are pioneering this silicon photonic architecture. We've conducted proof of concept with our Tomahawk 4 in a couple of hyperscale environments, though not yet in production volume. We have now gathered sufficient reliability data from those cases, allowing us to move forward with the production launch of Tomahawk5. As they say, the proof will be in the results. We expect that if we can demonstrate its power efficiency in one or two hyperscalers, it will start to gain traction among others. Power efficiency is crucial, especially in data centers, and we're looking at a potential 30% to 40% reduction in power usage. This is particularly significant for generative AI data centers, which could become a major application in the coming years.

Thank you, operator. In closing, we would like to highlight that Broadcom will be attending the Goldman Sachs Communacopia and Technology Conference on Thursday, September 7th. Broadcom currently plans to report its earnings for the fourth quarter of fiscal '23 after the close of market on Thursday, December 7th, 2023. A public webcast of Broadcom's earnings conference call will follow at 2:00 P.M. Pacific. That will conclude our earnings call today. Thank you all for joining. Operator, you may end the call.