Broadcom Inc. (AVGO) Earnings Call Transcript & Summary

Good afternoon, everyone. We're ready for our next keynote speaker. We're very honored to have Hock Tan, the President and CEO of Broadcom, here with us today. So Hock, thank you very much for participating, overall. Sorry that it's still virtual, but makes you productive today that you don't have to have any drive time.

Overall, I just want to talk about some macro factors before we get into Broadcom-specific dynamics. You had a very interesting view on the semi industry for many years about the maturity of the industry, the cyclicality, the consolidation that you've led into -- the consolidation you've been a leader in, I should say. As you look at the business right now, do you believe some of the cyclicality is actually changing and then some of the demand that we've seen during the COVID times is actually creating a sustainable step upward? Or are you still adherent to the market being pretty much as cyclical as it always has been?

Ross, thank you for having me. I appreciate it. And to answer your question point blank, I do not see any specific drivers or reasons why the strength we see today is really nothing more than of an exaggerated up-cycle. I think we are at a stage, and I think there's nothing that would tell me otherwise, that we will not, over time, revert back over the next 10 years that is as we had the last 10 years, revert to -- into an industry, the semi industry, that's really growing at GDP plus. And that's why I talked before in the past at about around mid-single digit, 5% global GDP plus a certain amount. So the...

You mentioned over the next 10 years, and I know that's your GDP plus statement. But if you think about the strength you're seeing now, how do I reconcile your cyclical view with the view that, I think, as of your last earnings call last week, you talked about being booked pretty much solid through almost the next year, if not all, of 2022. Is that just a continuation of the cycle dynamic you're talking about and just the lead times are so long that, that's what happens? Or is there some volatility that you're worried about even in 2022 from that cyclical framework?

No. If I can put in a bit of context maybe for your audience here, here's one, I think, we have seen what happened. 2017 was, my view, the last up-cycle for the semiconductor industry, 2017. And in 2017, we pretty much have the full portfolio of products we have today in semiconductors. And in that up-cycle peaking very nicely in 2017 and the first half of 2018, we grew year-on-year about 14%, 15% at that point. And that was a strong year, peak cycle. Then late 2018, things rolled over. 2019 was a down year for semiconductors. And that wasn't very long before -- and by the end of 2019, we started to see signs that maybe later on 2020, recovery could start coming in. COVID-19 hit early 2020 as we recall, and that sort of stopped that recovery. And that recovery was particularly driven -- as we see the recovery, a lot driven by perhaps enterprise. But what happened with COVID-19, as we moved to the second half of 2020 when is -- this lockdown of the economy, the work from home, learn from home and even live from home created an, as I call it, accelerated adoption of these certain technologies in this lifestyle out there. This is things like basically, we go on social media mall, we do a lot of activities online, even online banking, but online food ordering, retailing, everything a lot more and more interestingly, we need better PCs. So we see a huge demand for PCs around that time particularly, and we still continue to residual effect right now. And that draw -- so there's that accelerated adoption and that drove the strength of a lot of the semiconductor industry because the hypercloud guys have to scale up, have to expand their data centers to accommodate social media increases, things -- more cloud-adopted applications. Hypercloud drove that. Then by late 2020, we start to see a strong demand from broadband, telco service providers. And that's because they need to connect to the homes or you can do all this online thing, and we see that strong adoption, politically driven, but also economically lifestyle driven. And that's why we see, in 2021, reflecting that demand of bookings, strong bookings in first hypercloud, then broadband, very strong 2021 performance that ranges for semiconductor until the most recent quarter in July, almost 20% year-on-year growth. And you compare that to 2017 purposely, I mentioned that, yes, a strong up-cycle. But yes, I haven't spoken a lot on enterprise yet. But now over the last 3 months, we start to see enterprise as the economy starts to return and recover and loosen up, we start seeing a lot of demand from enterprise through OEMs. And by that, I mean OEMs like Cisco, IBM, Dell EMC and HP, those guys. They are the guys who are the system integrators and suppliers of technology and boxes to enterprises, large enterprises, small enterprises. We start to see a surge in demand now. Now that's not to say they won delivery nor can they get delivery immediately, but you're talking about out into the coming future. And I see the enterprise continuing their recovery simply because, if you look back, in 2019, they didn't spend; 2020, they didn't spend, so there's pent-up demand. And that's what perhaps is accounting for the fact that we're seeing 2022 is virtually booked up and it's driven by hypercloud, broadband and now enterprise.

Do you think there's a substitution effect where the first 2 of those would be slowing in 2022 and the latter one would be picking up? From what you just said, it sounds like all 3 are strong, but I would assume that if enterprise is going to spend because they haven't, cloud or hypercloud, as you call it in broadband, would be the opposite. They would slow down because they've already spent. What's wrong with that kind of symmetry?

Nothing. It may very well happen. Nothing. It may well happen. And in the sense that the growth we have seen in the last 12 months in hypercloud may stop growing as much or at all. But it's fairly elevated at this level, but the growth won't be there as much in '22 is my guess as it has been in '21 or '20. And in broadband, the same might apply that they have reached a level -- see my broadband has grown over -- 20% year-on-year over in '21 quarter -- for the last 3 quarters. And this current quarter is hitting almost 30%. They can keep growing at that speed. So it's elevated, it might not decline. By a doubt, it will grow much more. So if you want to look at year-on-year growth, '22, I think it will all be largely enterprise, but it'd be all fairly elevated. And the question, coming back to your initial question, what did all stay up there for a long time. And my view is no, come on. Things will be -- we always go through a period of digestion. There's no way we can consume on all that forever. And that's what is called a cycle, particularly when we expect supply to come into play out of this -- out of the current tightness, but dated back to 2020 to start coming in 2023. And the massive investment and CapEx will start deploying capacity in '23 earliest. Then I see '23 where we have supply. And I think digestion of demand might just start to occur.

That's the ultimate cycle. When supply is up and demand is down and opposite of what we have right now. So you mentioned the supply side of the equation. Do you think it takes till 2023 to actually get to some sort of equilibrium because of the time to capacity duration?

You don't see that much change in supply of some of the big heavy ticket items until '23. So now you have back-end assembly test and maybe even substrates coming in '22 as we are seeing ourselves, in late '22 as more substrates coming in. But on big heavy ticket item on foundries, I think you'll take that long.

Is that -- a year ago, people were debating about this and a lot of the foundry guys I talked to weren't really that confident in growth. So they seemed to delay some of their investments in the back half of last year. Do you think that delay is why it is going to take a few more years that they also are viewing the world as a bit more cyclical and so therefore, hesitant to put in supply? Or does it just take this long to put on the supply?

I think it just take this long. In this equipment that -- especially at the leading edge nodes, it just takes that long. If you have a shell in place, you may take faster. But if you don't, you're building from ground up, it'll take you 2, 3 years, 3 years more likely.

Got you. So let's move beyond the shortage topic but hit on one other macro dynamic, and it kind of goes into this adding supply to a certain extent, and that is the government's involvements in the industry. How do you view whether you want to call them subsidies, incentives, those sorts of things for the industry? Is that an aspect that's important to Broadcom, whether it be for R&D tax credits, investment credits? I know you don't have fabs. Or is it more just broadband spending if that's incentivized by government? So whether it be the demand side or supply side, how do you, given your history in the industry, view the government involvement that seems to be rising?

I think on the demand side, especially areas like broadband, I'm all for it. I think it's very, very -- it makes a lot of sense to expand spending infrastructure to connect every home. And the easiest -- one of the easiest way to connect home that is very cost effective is expand our network that exists -- a wireline network that exists in broadband already. And one of the cheapest way is to put fiber, put fiber to as many homes as you can, and we're seeing that. We're seeing that here in America, we're seeing that in Europe in a big way. And we have always seen it, we all know that, for years out of China. And we will also start to see that in India, as Jio Reliance embarked on a massive investment program which we are participating in a big way. So yes, we're seeing government operators. If a lot of them not pushed by government, some of them are making the additional investment, multiyear, multibillion dollars to connect homes with fiber optics because it's more effective. You do it once and you have huge bandwidth to put through to future proof for the next 10 years as opposed to trying to push copper. And most of Europe to now has been copper and they're pushing that. And when they do that, they're also pushing the cable guys to upgrade their coaxial. And that's also happening, and we are benefiting from that as well. So no, broadband, so from the demand side, we're all for doing that because then you literally future-proof connectivity to the homes for whatever comes -- may come in the next 10 years. It's the supply side that I would question, how wise it is to do that as opposed to letting capitalism, letting economics -- global economics work it's way to the system. Because in the in semiconductor industry, global economics has worked its way through for the last 40 years in a very, very efficient, effective manner and what makes governments think they can do it better.

Do you believe that geographic near-shoring is going to be a dynamic when you consider who to use as a wafer supplier? And I know you multisource already, but is geography part of that consideration for you?

We are all economic animals in the industry, especially in the American semiconductor industry, particularly. And look at it this way. Today, there's tightness of supply, so prices remain tight, hard to change, high in some cases. And we've seen supply chain materials costs go up over the last 12 months fairly dramatically, and they continue to hold firm. If we believe in the cycle, and that will happen, you'll see those material costs go down pretty dramatically, especially with new capacity coming in. And guess one, fabless guys like us will go to places where wafers have a price advantage over places where they produce at a premium. And then what will those foundries, manufacturing facilities sitting in high-cost location going to do? They're going to ask for government tariffs to protect themselves because if not, they'll be out of business very soon, which is why there's a danger of having political or government intervention into what would otherwise be a free -- a very efficient free market.

Got it. Let's move back to the revenue side of the equation and get a little more specific to Broadcom. As I think about the comments you said earlier about how different parts of your revenues are growing at different points in time, you mentioned about the enterprise side coming back up. I think on last week's call, you said enterprise in aggregate was still down year-over-year and might have even been down double digits. Can you remind us what percentage of your semiconductor business in its entirety would you describe as enterprise? I know half of your networking business is, but when you just take $20 billion in semi revenues, which is 75% of the company, how much of that is enterprise?

I would probably guess on our enterprise side, I would say, almost 40%, 35% to -- about say 35% is a fair way to describe it because there are little pockets here and there, say about 35% if you take the entire -- just the semiconductor industry. And by the way, if you add in my software, then it would be 50% because all of its software -- infrastructure software goes to enterprise. But just on semis, I would say about 35%, 35% is enterprise.

So a 30-year business, a little more, is still down year-over-year, up until, I think, your guided quarter, the October quarter, and yet you're still growing the better part of 20% year-over-year this year. So the math is pretty compelling behind the growth rates of your non-enterprise business to get there. You very thankfully are breaking out your business by segments, by end markets. Thank you very much for doing that again. It helps a lot with transparency.

I'm going to regret it. I know, but go ahead.

I'm not teeing you up for a [indiscernible] question. But if I was to look, regardless of your end market segments, about what's driving the non-enterprise growth, I can see what the wireless segment by itself is doing. We can talk about that later. But in your other segments, your networking, your broadband and your server storage connectivity businesses, those 3 big ones, what are the growth drivers that are delivering such significant growth year-over-year? And how sustainable do you believe them to be?

Okay. And I mentioned it earlier, one of the biggest -- yes, I think 2 very strong growth drivers that's happening, that's non-enterprise, what we consider non-enterprise. And by the way, we consider service providers as part of non-enterprise as opposed to enterprise, which is what some companies may do. And so broadband to us has been growing right now 25% to 30% year-on-year. That's very, very strong. And in networking, you're right, about half our revenues in networking are non-enterprise related. They are both service providers on one side and hypercloud on the other side. And those 2 areas, especially in the July quarter just past, we're growing very, very strongly. No surprise. As I indicated, we started seeing the growth in 2020 and they continue up to today. Broadband -- in fact, broadband, in particular, especially in '21, started getting their deliveries and started seeing that being strong. So networking as it pertains to telcos and service providers, which is upgrading, investing, upgrading their backhaul to 5G networks as well as call edge routing, metro routing is just booming and -- just as hypercloud is. And that's what's offsetting the growth -- offsetting the weakness even year-to-year on enterprise.

How much of that growth is the pie getting bigger as the hypercloud guys are spending so much to support the kind of the work-from-home dynamic versus Broadcom benefiting from share gains might not be the way to think about it, but product cycles? Whether it be generations of Tomahawk, generations of Jericho or Qumran, how much is kind of the rising tide versus Broadcom-specific?

Well, it's hard to figure that out -- the mix because, by the way, you're seeing enterprise, which means includes enterprise data centers, enterprise spending down, as I indicated year-on-year, and hypercloud and telcos strong. Whether it's broadband or backhaul in their networks, they're up. So we are definitely seeing a swing. If I have to split up my revenues instead of by end markets, if I have to structure it by breaking it between 3 groups, which is enterprise, traditional enterprise that is, service provider and cloud as one group and consumer as a third group, you're probably seeing a swing out of enterprise over the last few months -- last few quarters more into consumer and more into hypercloud and service providers, and that's what happened. And -- but equally, just to make sure it doesn't flip the other way. It's -- next year, it might rebalance out more as enterprise will probably outgrow the other guys, outgrow the other 2 segments.

So within the cloud business, you've talked about the switching and routing being a really strong driver for you. And I think everybody is well aware of the leadership position you have in that. How are you viewing the competitive landscape there? There's been some recent acquisitions about some people entering the space. There's been newcomers for many, many years trying to get into it. Are you seeing any meaningful change?

No. No meaningful change.

We'll leave that part as it is. ASICs is another part of the cloud business. I know it's always very, very sensitive just by its nature and what you can say in your ASIC business. But are you seeing different demand patterns on that side than on your traditional merchant, silicon switch and routing business as some of those cloud vendors go more to doing their own ASICs?

It has been going on for -- and I've indicated that before 6 years as far as I recall. So as far back as 6 years ago, we have been engaged in a fairly active matter with various hypercloud guys on trying to create ASICs, silicon ASICs. And let me put it in the context, why that happens. It really comes down to this. I mean all hypercloud guys, if they have a choice, would want to just architect -- design, architect and build their data centers, their infrastructure as simply as possible. They're not here to prove to the world they can -- they are part of an analyst frontier, pushing in an envelope on the best bright shiny object out there in technology. They want to run their data centers. They want to run their cloud infrastructure as effectively, simply as possible. Because at the end of the day, you're an operator, you're an operator of a utility almost as a high-tech utility. And so they do that. But see, what is happening is data centers tend to be used to be until -- and I use the word used to be very compute-centric. You use powerful and every generation gets more and more powerful general-purpose CPU, the CPU-centric. And then you load this CPU with all kinds of software, you write all kinds of software and enable the CPU to drive it, it just compute. And in many ways, it does compute, but it also does its all kinds of ancillary functions that is not compute and use software to make functions work. Like example, YouTube provides a lot of video. Social media -- Facebook does that too. Most of all that is created through software running on general-purpose CPUs they have in their data centers. Now these CPUs are starting to hit the wall of Moore's law in performance. So guess what? So the cloud guys are frustrated and say, I want to then maybe improve on the quality, resolution, performance on my video, and it's the function called video transcoding. So guess what they're doing. They're now asking us to create specialized accelerators, engines. And part of it is compute engines or it's smaller compute engine but it is to be able to enable them to perform more efficiently and effectively on video transcoding. So they create that ASIC. That is for a purpose, very technic, but it's because general-purpose CPUs slowing down, not because they love to show the world they can do it. Then the same applies for what they think about doing in trying to save orchestration or virtualization of their CPUs and they create specialized engines to do more of those specialized workloads. And most obvious is machine learning, right? Technically, you can run software on CPUs and do machine learning. Believe me, you can do that now as you want to push that envelope on real-time, quick response, high volume -- high regression analysis. So you start -- they start to create specialized engines more than just using general-purpose CPU. So they move to GPU. All, as Google does, create their own TensorFlow specialized engine, which we are very much a part of. These are very, very high-end specialized engine, special purpose to do machine learning. But you could still do CPU. So the message learned is if you supply to the cloud guys and you can supply what they need pretty much or mostly what they need, they are not out there to create their own silicon. They do it because you cannot supply. So we are all over these guys for years. And the problem in this model of creating specialized silicon for these cloud guys is half of them are science projects because, number one, scale, big as they are. Many of them are not big enough to handle these leading-edge, highly specialized silicon with lots of IP embedded. They don't have the revenue to buy a lot from you, so it becomes expensive. And then they look at it and say, what other choices do I have? And so a lot of these projects become like science project also because some of the technologies doesn't get far enough. So there are successes as I'm happy to tell you that we told you they are in this compute offloading phenomenon but there are also a lot of failures. And lately, some people get very excited. There's a lot of hype and market hubris about all these cloud guys are doing their own ASICs and all that. No, no, no. It hasn't changed for 6 years. They're just looking for specialized silicon engines for specific workloads that they need to have. It's very practical. It's not any big special phenomenon that has changed for the last 6 years we've been engaged. And I can tell you there are successes, but there are also a lot of failures, but we'll keep investing.

So we have 5 minutes left, and I want to hit on 2 topics, and so it will have to be -- these topics could be a half hour conversation each, but I want to just make sure to hit these last 2. Your software business is 25% of the company. You're holding an analyst meeting or a deep dive into it in New York on November 9. So I look forward to hearing more about that. But do you believe that -- well, since you've entered that market, you've grown that mid-single digits exactly like you had planned. Do you believe that is sustainable? And what do you believe people are missing about your software strategy? That's the first question.

That's a long answer. So I will give you an overview, put in a few very interesting observations I've seen. We have a framework in mind when we decide to go into infrastructure software, and they have all proven out in space. It's a contradiction to many of these snowflakes running around, by the way, and -- but it also depends on the model we chose to have. And as you know Broadcom, as in semis, same in software, we manage our business -- our business model is to manage for sustainability and earnings, earnings growth. That's we think at the end of the day what sustains. And so we run our company that way, but enough for me to say the way we run our software and is proven out is simply that when you sell software and run the business software, you can -- if the software, the millions and millions of lines of codes you create to develop a tool, whether it's a DevOps tool, ValueOps tool, any of those infrastructure mission-critical tools, decent-size software, people would like to use it. It's all about cost of adoption and the cost of doing it. But you can always get it in the hands of people to use. More software models today are all about let's grow it, let's put it in the hands of as many new enterprises out there as we can because that's how you grow. You show easy growth is putting in. There's a direct correlation with spending money resources on sellings to growing your business. The only thing is what is meant by selling? And this is an interesting part of software because they already send the whole thing over, the whole stake, people use it. It's not enough, you sign up a new company and you acquire a new customer or enterprise to either sign up a subscription or licensing, you now have to teach these guys how to use it, right? That calls for solution engineering, application engineering spending. So you have to do that. Otherwise, it just get thrown away. After you teach them, but as part of teaching them, they will start to know, no matter how good your software is, it's never universal. Everyone will have their own special peculiarities, their own idiosyncrasies, they want special workloads. So they want tweaks in your software. They call it -- we call that feature requests. In the software industry, I really use it. And half of what you spend on R&D is actually feature requests based on customer requests. And then, of course, you still have to support it based on how many customers. So what I'm trying to tell you is when you sell and lend new software in a new customer, that's a whole trail of OpEx you have to spend. It's not just it. I tell you that for many of these new companies growing, $100 of revenue, you spend $90 just to acquire a customer. Now our approach is let's get a footprint of the largest customers, 1,000 or so, who has big capability of buying and just grab their wallet share in a bunch of software we have that they already have big footprint and expand and renew and keep expanding and supporting, if you like. You spend out of that $100 of revenue, as we show you in our data, $30. That's all you need to do, not $90, if you're trying to grow. So take a choice. Do you want to make money and the old fashion rate, so the growth -- to run your business, but you grow 5% instead of trying to grow 30%. That's the difference.

So the 5% is very sustainable?

It is very sustainable.

Perfect. Well, Hock, we are out of time. So there's many, many more questions I could ask you, and I will leave those for a later date. But thank you so much for joining us at the conference, participating and sharing your years of wisdom with us. So look forward to seeing you soon, and thanks again for your attendance.

Ross, glad to be here. Thank you.

Thank you. Bye-bye.