📄
AVGO Q2 2024 Earnings Call
June 12, 2024 at 12:00 AM
🔍
Company
AVGO
Quarter
Q2 2024
Date
June 12, 2024 at 12:00 AM
Speakers
2
Word Count
~6,059
Transcript Content
Operator
Thank you. Our first question will come from Vivek Arya with Bank of America. Your line is open.
Speaker
Thanks for taking my question. Hock, I would appreciate your perspective on the emerging competition between Broadcom and NVIDIA across both Accelerators and Ethernet switching. So on the Accelerator side, they are going to launch their Blackwell product that many of the same customers that you have a very large position in the custom compute. So I'm curious how you think customers are going to do that allocation decision, just broadly what the visibility is. And then I think Part B of that is as they launch their Spectrum-X Ethernet switch, do you think that poses increasing competition for Broadcom and the Ethernet switching side in AI for next year? Thank you. Hock Tan (CEO): That's an interesting question, Vivek. Regarding AI accelerators, we are following a different model. The GPUs, which are the preferred AI accelerators in the market, are extremely effective and are an area where NVIDIA excels. We don’t see ourselves as competitors to them in this space, as they have their strengths and we know our position in relation to that. However, for selected hyperscalers, if there's scale and the ability to create silicon solutions, we are open to using our intellectual property to develop custom ASIC AI accelerators tailored for specific complex AI workloads. I do not view these as direct competitors. I am not positioning myself to compete with them in the GPU market. We are not competitors there, nor do we aim to be. In terms of networking, the situation may differ, as people may approach it from various perspectives. We have a long history with Ethernet, having been involved for over 25 years, and we have navigated many market transitions, successfully adapting from cloud-scale networking to routing and now AI, making our entry into AI a natural progression. We acknowledge that the ecosystem associates GPUs with being the primary AI compute engine, and they strive to create a highly integrated end-to-end platform. Our approach doesn’t involve manufacturing those GPUs; instead, we focus on facilitating their operation efficiently. Ultimately, we aim to complement GPUs for customers who are expanding their GPU clusters.
Speaker
Thank you.
Operator
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 guys. Thanks for taking my question. I want to stick on the AI theme, Hock. The strong growth that you had in the quarter, the 280% year-over-year, could you delineate a little bit between if that's the compute offload side versus the connectivity side? And then as you think about the growth for the full year, how are those split in that realm as well? Are they kind of going hand-in-hand? Or is one side growing significantly faster than the other, especially with the I guess, you said the next-generation accelerators are now going to be Broadcom as well? Hock Tan (CEO): Well, to answer your question on the mix, you are right. It's something we don’t really predict very well, not understand completely except in hindsight. Because it's like, to some extent, to the cadence of deployment of when they put in the AI accelerators versus when they put in the infrastructure that puts it together, the networking. And we don't really quite understand it 100%. All we know, it used to be 80% accelerators, 20% networking. It's now running closer to two-thirds accelerators, one-third networking and we'll probably head towards 60%-40% by the close of the year.
Speaker
Thank you.
Operator
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, guys. Thanks for taking my question. I wanted to ask about the $11 billion AI guide. You'd be at $11.6 billion even if you didn't grow AI from the current level in the second half. And it feels to me like you're not suggesting that. It feels to me like you think you could be guided. So why wouldn't that AI number be a lot more than $11.6 billion? It feels like it ought to be. Or am I missing something? Hock Tan (CEO): Because I guided just over $11 billion, Stacy could be what you think it is. Quarterly shipments get sometimes very lumpy. And it depends on rate of deployment, depends on a lot of things. So you may be right. You may estimate it better than I do, but the general trajectory is getting better.
Speaker
Okay. So are you suggesting that more than $11 billion could be considered the worst-case scenario since that would just be flat at the current levels? But you're also indicating that conditions are improving in the latter half of the year? Hock Tan (CEO): Correct.
Speaker
Okay. So I guess that's a very conservative number? Hock Tan (CEO): That's the best forecast I have at this point, Stacy.
Speaker
All right. Okay, Hock, thank you. I appreciate it. Hock Tan (CEO): Thank you.
Operator
One moment for our next question, and that will come from the line of Harlan Sur with JPMorgan. Your line is open.
Speaker
Yeah, good afternoon. Thanks for taking my question. Hock, on cloud and AI networking silicon, good to see that the networking mix is steadily increasing. Like clockwork, the Broadcom team has been driving a consistent two-year cadence of new product introductions, right of new generations of switching and routing products for the past seven generations. You layer on top of that your GPU - TPU customers are accelerating their cadence of new product introductions and deployments of their products. So is this also driving faster adoption curve for your latest products? And then maybe just as importantly, like clockwork, it has been two years since you've introduced your product, right, which if I look back historically, means you have silicon and are getting ready to introduce your next-generation products, which would put you two to three years ahead of your competitors. Can you just give us an update there? Hock Tan (CEO): Harlan, you're pretty insightful there. Yes, we launched Tomahawk 5 in 2023. So you're right, by late 2025, the time we should be coming out with Tomahawk 6, which is the 100 terabit switch, yes.
Speaker
Is the acceleration of cadence by your GPU and TPU partners also driving the strong growth in the networking products? Hock Tan (CEO): Well, you know what, sometimes you have to let things take their time. But it's a two-year cadence so we're right on. Late 2023, once when we shoot it out to a Tomahawk 5 and adoption. You're correct with AI has been tremendous because it ties in with the need for a very large bandwidth in the networking, in the fabric for AI clusters, AI data centers. But regardless, we have always targeted Tomahawk 6 to be out two years after that, which should put it into late '25.
Speaker
Okay, thank you Hock.
Operator
Thank you. One moment for our next question, and that will come from the line of Ben Reitzes with Melius. Your line is open.
Speaker
Hi, thanks a lot. And congrats on the quarter and guide. Hock, I wanted to talk a little bit more about VMware. Just wanted to clarify if it is indeed going better than expectations. And how would you characterize the customer willingness to move to subscription? And also just a little more color on Cloud Foundation. You've cut the price there, and are you seeing that beat expectations? Thanks a lot. Hock Tan (CEO): Thanks, and thanks for your kind regards on the quarter. But as far as VMware is concerned, we're making good progress. The journey is not over by any means, but it is pretty much very much to expectation. Moving to subscription, well, in VMware we are very slow compared to a lot of other guys, Microsoft, Salesforce, Oracle, who have already been pretty much in subscription. So VMware is late in that process. But we're trying to make up for it by offering it in a very compelling manner because subscription is the right thing to do, right? It's a situation where you put out your product offering, and you update it, patch it, but update it feature-wise, everything as capabilities on a continual basis, almost like getting your news on an ongoing basis, subscription online versus getting it in a printed manner once a week. That's how I compare perpetual to subscription. So it is very interesting for a lot of people to get on board. And so to no surprise, they are getting on very well. The big selling point we have, as I indicated, is the fact that we are not just trying to keep customers kind of stuck on just server or compute virtualization. That's a great product, great technology, but it's been out for 20 years. Based on what we are offering now at a very compelling price point, the whole software stack to use vSphere and its basic fundamental technology to virtualize networking, storage, operation and management, the entire data center and create this self-service private cloud. And thanks for saying it, you're right, and we have priced it down to the point where it is comparable with just compute virtualization. So yes, that is getting a lot of interest, a lot of attention from the customers. We have signed up who would like to deploy private cloud, their own private cloud on-prem. As a nice complement, maybe even alternative or hybrid to public clouds, that's the selling point, and we are getting a lot of interest from our customers in doing that.
Speaker
Great. And it's on track for $4 billion by the fourth quarter still, which is reiterated? Hock Tan (CEO): Well, I didn't give a specific time frame, did I? But it's on track as we see this process growing towards a $4 billion quarter.
Speaker
Okay, thanks a lot Hock. Hock Tan (CEO): Thanks.
Operator
Thank you. 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, thank you so much for taking my question. I guess kind of a follow-up to the previous question on your software business. Hock, you seem to have pretty good visibility into hitting that $4 billion run rate over the medium term, perhaps. You also talked about your operating margin in that business converging to classic Broadcom levels. I know the integration is not done and you're still kind of in debt paydown mode. But how should we think about your growth strategy beyond VMware? Do you think you have enough drivers, both on the semiconductor side and the software side to continue to drive growth or is M&A still an option beyond VMware? Thank you. Hock Tan (CEO): Interesting question. And you're right. As I indicated in my remarks, even with the contribution from VMware this past quarter where we have AI helping us, but we have non-AI semiconductor sort of bottoming out. We're able to show 12% organic growth year-on-year. So almost have to say, do we need to rush to buy another company? The answer is no. But all options are always open because we are trying to create the best value for our shareholders who have entrusted us with the capital to do that. So I would not discount that alternative because our strategy, our long-term model has always been to grow through a combination of acquisition, but also on the assets we acquire to really improve, invest, and operate them better to show organic growth as well. But again, organic growth often enough is determined very much by how fast your market would grow. So we do look towards acquisitions now and then.
Speaker
All right. Thank you.
Operator
Thank you. One moment for our next question, and that will come from the line of Blayne Curtis with Jefferies. Your line is open.
Speaker
Hi, thank you for taking my question. Hock, I wanted to ask about the networking business outside of AI. It seems like there is an inventory correction that the entire industry is experiencing. However, I didn't hear you mention if it has reached a bottom. From my perspective, it seems to be down about 60% year-on-year. Are you seeing that business stabilize? I know you indicated that the overall semiconductor sector, excluding AI, should see a recovery. Do you have any insights on customer inventory levels in that area? Hock Tan (CEO): We see it behaving. I didn't particularly call it out, obviously because more than anything else, I kind of link it very much to server storage, non-AI that is. And we call server storage as at a bottom Q2, and we call it to recover modestly second half of the year. We see the same thing in networking, which is a combination of enterprise networking, as well as the hyperscalers who run their traditional workloads on those, though it's hard to figure out sometimes. But it is. So we see the same trajectory as we are calling out on server storage.
Speaker
Okay, thank you.
Operator
Thank you. One moment for our next question, and that will come from the line of Timothy Arcuri with UBS. Your line is open.
Speaker
Hi, sorry. Thanks. Hock, is there a way to sort of map GPU demand back to your AI networking opportunity? I think I've heard you say in the past that if you spent $10 billion on GPU compute, you need to spend another $10 billion on other infrastructure, most of which is networking. So I'm just kind of wondering if when you see these big GPU numbers, is there sort of a rule of thumb that you use to map it back to what the opportunity will be for you? Thanks. Hock Tan (CEO): There is a connection, but it’s quite complex, so I stopped trying to create a model for it. For every $1 billion spent on GPUs, one might estimate spending an additional amount on networking. While we aren't fully engaged in that market, we do supply components like DSPs, lasers, and PIN diodes for high-bandwidth optical connections. When considering networking as a whole, including switching and all relevant components attached to clustering GPUs, approximately 25% of the GPU's value can be attributed to networking. However, not all of this is part of my available market, as I don’t handle the optical connections directly, but I do supply some related components. A simple way to view it is that around 25% to 30% of all infrastructure components are linked to the value of GPUs. Nonetheless, it’s important to note that deployment timing may vary; there could be early GPU purchases followed by later networking investments, which is why you're seeing a mix within my AI revenue. Generally, over time, you would trend toward that range.
Speaker
Perfect Hock, thank you so much.
Operator
Thank you. One moment for our next question, and that will come from the line of Thomas O'Malley with Barclays. Your line is open.
Speaker
Hi, guys. Thanks for taking my question. And nice results. My question in regards to the custom ASIC AI. Hock, you had a long run here of a very successful business, particularly with one customer. If you look in the market today, you have a new entrant who's playing with different customers. And I know that you said historically, that's not really a direct customer to you. But could you talk about what differentiates you from the new entrant in the market as of late? And then there's been profitability questions around the sustainability of gross margins longer term. Can you talk about if you see any increased competition? And if there's really areas that you would deem more or less defensible in your profile today? And if you would see kind of that additional entrant maybe attack any of those in the future? Hock Tan (CEO): Let me address the second part first, which is our custom AI accelerator business. It's a highly profitable sector. To put it into perspective, each of these AI accelerators functions similarly to a GPU. Running large language models requires multiple accelerators, regardless of their power. Additionally, these models demand significant memory access. Each accelerator comes with considerable cache memory, commonly known as high-bandwidth memory designed for AI accelerators or GPUs. We provide both components in our custom business. The compute function related to the chips has margins comparable to most semiconductor businesses, but when paired with a large amount of third-party memory, it doesn't contribute to margin stacking. Basic math indicates that this affects the margins for AI accelerators that include memory, which we do. While this arrangement increases revenue, it does dilute the margin. However, the R&D and operating expenses relative to this higher revenue are significantly lower. So, on an operating margin basis, this segment is just as profitable, if not more so, given the scalability of the custom AI accelerators. Their performance is even better than our typical operating margin. This return on investment is what drives us to excel in this challenging field. To answer your first question, there is only one Broadcom.
Speaker
Thanks Hock.
Operator
Thank you. One moment for our next question, and that will come from the line of Karl Ackerman with BNP. Your line is open.
Speaker
Hi, thank you. Good afternoon. Hock, your networking switch portfolio with Tomahawk and Jericho chipsets allows hyperscalers to build AI clusters using either a switch-scheduled or endpoint-scheduled network. And that, of course is unique among competitors. But as hyperscalers seek to deploy their own unique AI clusters, are you seeing a growing mix of white-box networking switch deployments? I ask because while your custom sales and business continues to broaden, it will be helpful to better understand the growing mix of your $11 billion AI networking portfolio combined this year. Thank you. Hock Tan (CEO): Let me have Charlie address this question. He's the expert.
Speaker
Yes. Thank you, Hock. So two quick things on this. One is the – you are exactly right that the portfolio we have is quite unique in providing that flexibility. And by the way, this is exactly why Hock, in his statements earlier on, mentioned that seven out of the top eight hyperscalers use our portfolio. And they use it specifically because it provides that flexibility. So whether you have an architecture that's based on an endpoint and you want to actually build your platform that way or you want that switching to happen in the fabric itself, that's why we have the full end-to-end portfolio. So that actually has been a proven differentiator for us. And then on top of that, we've been working, as you know, to provide a complete network operating system that's open on top of that using SONiC and Psi, which has been deployed in many of the hyperscalers. And so the combination of the portfolio plus the stack really differentiates the solution that we can offer to these hyperscalers. And if they decide to build their own NICs, their own accelerators are custom or use standard products, whether it is from Broadcom or other, that platform, that portfolio of infrastructure switching gives you that full flexibility.
Speaker
Thank you.
Operator
Thank you. One moment for our next question, and that will come from the line of C.J. Muse with Cantor Fitzgerald. Your line is open.
Speaker
Yeah. Good afternoon. Thank you for taking my question. I was hoping to ask two part software question. So excluding VMware, your Brocade, CA, and Symantec business now running $500 million higher for the last two quarters. So curious, is that the new sustainable run rate or were there one-time events in both January and April that we should be considering? And then the second question is as you think about VMware Cloud Foundation adoption, are you seeing any sort of crowding out of spending like other software guys are seeing as they repurpose their budgets to IT? Or is that business so less discretionary that it's just not an impact to you? Thanks so much. Hock Tan (CEO): I'm not really sure about any crowding out. What we're offering isn't something that companies want to do on their own since they're already investing in building their own on-prem data centers. Historically, many enterprises follow a best-of-breed approach, which means they create data centers with the best compute, often using vSphere for compute virtualization, alongside top networking and storage solutions, all managed by a common operations layer, which is frequently VMware. The challenge is that this mixed approach does not create a highly resilient data center. With a diverse setup, it's difficult to quickly identify root causes when issues arise, leading to blame being assigned in different directions. Many CIOs we speak to express a desire for a unified platform instead of the traditional best of breed. This creates opportunities, as those starting from scratch in a greenfield scenario can implement solutions more easily, while those in brownfield situations, trying to upgrade existing data centers, often face more challenges. I don't think we're witnessing a significant level of crowding out; there is competition in greenfield scenarios where budgets can be allocated to an entire platform instead of just best-of-breed options. However, upgrading existing data centers presents a more complex situation for us. Regarding our revenue mix, Brocade is performing very well this year, but we know that it won't last forever as Brocade, like many enterprise purchases, goes through cycles. We're taking advantage of this while we can.
Speaker
Thank you. Hock Tan (CEO): Thanks.
Operator
Thank you. And we do have time for one final question, and that will come from the line of William Stein with Truist Securities. Your line is open.
Speaker
Great. Thanks for squeezing me in. Hock, congrats on the yet another great quarter and a strong outlook in AI. I also want to ask about something you mentioned with VMware. In your prepared remarks, you highlighted that you've eliminated a tremendous amount of channel conflict. I'm hoping you can linger on this a little bit and clarify maybe what you did. And specifically also what you did in the heritage Broadcom software business, where I think historically, you've shied away from the channel. And there was an idea that perhaps you'd reintroduce those products to the channel through a more unified approach using VMware's channel partners or resources. So any sort of clarification here, I think, would be helpful. Hock Tan (CEO): Thank you for the question. VMware has taught me a lot, especially with their impressive 300,000 customers. We've chosen to invest significantly in a resale network for VMware, focusing on building strong partnerships with distributors and about 15,000 value-added resellers. This strategy is showing promise six months in as we are seeing increased activity. The resellers tend to focus on the extensive long tail of VMware's customers. The largest enterprises among the top 10,000 customers, such as major banks and healthcare companies, demand tailored service and support from us. To meet this need, we've adopted a direct approach while still leveraging our chosen resellers where necessary. The resellers provide valuable services to the broader customer base. We've strengthened our reseller network to enable them to manage and support customers effectively, without directly competing with them unless the customer prefers that option. Customers can choose the level of support they want, whether through direct sales from VMware or through resellers, streamlining the process for them. In contrast to our previous approach where there was significant chaos in the channel, we now ensure clarity in our relationships. Customers can purchase directly from VMware or through resellers, and we also offer them the option of running their applications efficiently in a managed hosted environment through a network of global managed service providers. This approach provides our enterprise customers with VMware as a service as a distinct and attractive alternative.
Speaker
Great. Thank you.
Operator
Thank you. I would now like to hand the call over to Ji Yoo, Head of Investor Relations, for any closing remarks. Ji Yoo (Head of Investor Relations): Thank you, Cherie. Broadcom currently plans to report its earnings for the third quarter of fiscal '24 after close of market on Thursday, September 5, 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
Thank you all for participating. This concludes today's program. You may now disconnect.
Need more earnings data?
Explore more earnings information and financial data for AVGO
Transcript sourced from Financial Modeling Prep (FMP)