Broadcom Inc. (AVGO) Earnings Call Transcript & Summary

Good afternoon, everybody. Welcome to the Goldman Sachs Community pain Technology Conference. Lunch session. My name is Jim Schneider. I'm the semiconductor analyst here at Goldman Sachs. It's my pleasure to welcome Broadcom and CEO, Hock Tan, the stage today. Welcome market. Really thrilled to have you here.

Thank you. Thank you for inviting me.

Hock, before we dive in talking about your AI business, I want to get some of your broader thoughts on AI and its adoption just across the world here. given your deep view into the enterprise and all aspects, I think it's worth asking in how you characterize the state of AI adoption across the average enterprise customer today?

That's a very interesting question, Jim. And I guess, I will start off with Broadcom as a company. We, like most large enterprises are out there today, we have been looking closely reviewing how we can adopt generative AI through use of those frontier models and possibly even those open weight models that are floating around these days, but we've been looking at it. And we've been doing that in earners now for 1 year. And part of the benefit we have is we have quite good full access to Mitas preview and came available. And we look at how does it help our businesses. And where we've been looking at very closely is obviously we're a technology company and how we create products, how we design products, hardware as well as software. And we have engaged 15,000 of our 30,000 engineers in the company in looking at it both in hardware and software and a lot of insights, a lot of learnings. The biggest issue is finding use case that generates what you call, a return on investment that is clearly meaningful and significant. That's not easy to do. And I suspect a lot of companies out there face the same issue. The second thing is 2 frontier models, any model created as applications, it's just a tool. They're not meant to literally replace your people, not at least until you realize it's a tool and how would you apply it. And we find out that different users, depending on who they are, could use the tool very differently, very effectively, very efficiently as well as our others who are having time really getting anywhere. -- is us use GPT. And frankly, depending on how I ask, how I prompt, I get different sets of answers and some more useful than others. And over time, you get better and better at it. But there's a limit. And so what we really find is to find great use cases. My thinking is simply 1 immediate use case, and that's probably from our perspective, is creating new products, creating or help us create new technology and what we're trying to ask for is not for it to create it out of thinner as much as create it through our engineers. And what we need -- what we find available out of it is they created with very high quality. The quality is very critical, and that's what those AI tools enable us to do and on a more accelerated fashion. And for a semiconductor chip, that makes a difference because our engineers can still design the chip, but if they have a bulk and I have to respin that's 6 months and $30 million for leading-edge silicon. Open AI and tropic tools, Gemini tools could help us eliminate the need for that re-spin. So there's a return on investment. So we find that as purposeful. But in terms of just simply productivity improvement, the return on investment is debatable. So it's really about quality accelerated time to deployment or creation of leading-edge technology. You don't go back and redo what's in the past. And the other application, of course, is cybersecurity through metals to scan to basically secure our infrastructure better. And finally, and it's something we have not used it, we have not used that, but we can imagine you can use something like a fall in the minds of critical thinkers in specific functional areas to really replicate the software stack that we're using in specific -- that definitely is big, and that can generate a big return. But that is something that will require a lot more investment. And we haven't reached that part of the thing in. So some total of it is still early stage, but we do find it's important most of all the users. And the good users and good users are defined as critical thinkers. More in simple terms, the architects, senior people who are able to understand the context, understand the environment of where we are, the workflows and been able to prompt the models to do it. And I can imagine the day when they start using agents, which we are starting to last 3 months, that becomes even more important. So to sum it up, it helps smart -- very smart people become smarter. Yes. That's why I won't say the converse.

Now as the models get more efficient, more powerful, you talked about token costs coming down quite dramatically. What's your personal opinion about where the greatest value is going to accrue across the AI SAC, model layer, application layer, someplace else?

Well, I'm a believer in the fact that intelligence and the more there is of it is unique. It's like doing semiconductors. When we design a switch, a network switch. That's the best of its generation at a timely basis. The world did a path to our door. There's no substitute for the best, second best doesn't get anything. So I look at this the same way for AI as we -- as long as there is continual improvement, I will almost say exponential improvement in the LLM that keep coming out from the Frontier model guys. The value is going to accrue over time to the guys who do the best frontier models. And we look at it this way. Take the case of a very good frontier model, 1 gigawatt could generate $30 billion ARR as you know. The cost to run 1 gigawatt, as you all know, is about $10 billion a year. So you have $20 billion that accrue to the model guys and the application that goes with it and $10 billion fought over buying the power guys, the cloud provider guys, even the chip and memory guys. So the value comes, I think, in creating the best models and through it the best products to monetize the applications.

Understand. Okay. Now clearly, the pace and scale of this spending and infrastructure build-out is exceeding a lot of people's expectations. In your custom silicon business, you've consistently talked about working with a very limited number of large customers who have their own models can support very large spending programs over time. What do you think your customers are trying to achieve when they work with you? Is it supply diversification, cost optimization some performance advantage by marrying software with custom silicon. And maybe talk a little bit about or quantify even the advantages they can get by working with you.

James, really very, very direct and straightforward. When we design, we develop, I would say, what I call a custom accelerator. We call it XPU as you all have for a customer. It's really very effective when we actually codesign it work closely with that particular customer who happens -- who has to be, in our view, to be a customer, a frontier model developer. Because what happens is we are closely embedded our engineering teams with with the team of our customers who help create those frontier models. And it starts with the Frontier models. They understand what they're trying to create. They understand what breakthroughs they are making to make their -- for their algorithms to create workloads that run up models much more effectively, more productively, more efficiently. And they put the specs over to us, and we work with them to create a chip all the way through to the physical part of the chip that that enables the chip, the transistors to be able to handle the kind of workloads that model does. So that co-design co-working together, collaboration, tight collaboration really makes the difference. And if you do that that collaboration, you have seen a walking example of what happened recently and its news out there with open AI. We work closely with the open AI team. And within a year, we created an XPU called Helio, which performs as well, if not better than the best state-of-the-art general purpose accelerate out there.

Yes. Fair enough. Now I want to get to the heart of the forecast you provided at your earnings call last week, you basically laid out the path to doubling your AI revenue for each of the next 2 years. $115 billion in , $230 billion in 2028. And you said those are supply-wise estimates. So I want to take a minute to sort of unpack some of that. In terms of the supply constraints or the constrained part of it. You mentioned the ability to deploy physical data centers as 1 of those land power shell. What is the time frame that you think is kind of the pinch point for data center availability on the infrastructure side. What's the biggest limiter inside that as you see it? Grid power, gas turbines, construction? And then how would you sort of throw the sort of wildcard about political backlash into the equation?

What we are seeing now the positive side to this phenomenon happening is you cannot build up easily in AI data center without a lot of lead time. So it gives us pretty good visibility out there, what needs to be done. But equally, that lead time leads to a fair degree of sometimes uncertainty on preciseness of the timing. And 1 of the biggest -- I mean, before I even say that, now, for creating the chips and even that we do, whether it's accelerators, AI accelerators or network chips, connectivity chips. We understand what it takes to the supply chain for chips. We understand the need for wafers, leading-edge wafers and memory as well as even substrates. And that's pretty well locked in. We can pretty well locked that in for sure in '27 in the process of locking in '28 gives us within some limits. Then it's more within our control. What is interestingly is we find out that's not enough when you want to put together AI data centers. You need power. You need a site, you need a facility, the house their chips point to run. And that's the other part of the constraint. I'm not saying chips and memory are not constrained. They are constrained. But at least there's constraints, we can work with a more deterministic. What is less deterministic is a power site that is power-ready, no, that's power available, but not ready. So you want to make it power ready, saying '28 got to start construction on the site now. And that's not us, our customers has to start construction on the site now to make it available and ready by '28. And that's 2 years -- almost 2 years, 1.5 years out. And a lot of it is not just equipment, whether they're transformer gas turbines, that obviously are the pieces that go in it. It's also construction. Construction, including local permitting, though some of it exists, but also the rate of construction, you're talking about those if you're trying to renovate or build your own houses, you know how construction is like that's what at large scale, this is happening. And so there's a lot of other factors that come in that kind of make it I describe it as perhaps constrained because if I turn around give you guys a forecast, which we do -- we want to be sure that's a forecast that is very clear line of sight. And that's what translates to the numbers you mentioned is a judge number based on not just supply of memory wafers, but availability and readiness of power sites.

Fair enough. Okay. Now in terms of the customers, Google is 1 of your long-standing strategic partners for custom silicon. Thomas Kurian was just up here a couple of hours ago, outlining their strategy back in April, you disclosed a long-term agreement with Google to supply TPUs and networking products through 2031. Can you maybe talk about how this long-term agreement deepens the partnership between you and Google over that time frame?

Google and us have a very strong relationship. We've been doing this for over 10 years ever since the first version of the TPU. And we've been doing every version since then. And we will continue to do so under this LTA because if anything else, it expands and strengthen the relationship very technical relationships, strategic relationship we have had with Google now for 10 years, and we see that continuing. And 1 of the things that we offer -- we believe we offer the Google is the strength, the breadth of our technology in -- especially in semiconductor design. Because with every new generation of GPU or XPU in our case, we are pushing. Our engineers are pushing the limits of semiconductor technology. And 1 part of -- 1 example is, as you know, in the past, 10 years ago, you get better performance out of a CPU in chips by moving to a next-generation process node. And each time you move that and process new generation process node every 18 months or so. You double of transistor count, you double performance. Well, that's more slow. Mason has come to an end. Going -- moving down process on not -- you cannot achieve double transistor performance an you barely got 5% improvement. But you get power. So that's 1 positive thing. But building the kind of accelerators we are building today, and these monster multipliers very, very complex chips. One, we are doing with our customers is we're trying to what I call substitute malls law with something else. And a simple way to look at it is Maslani on chip sector chip is limited by the radical law of physics on optics, 800 square millimeters. That's the largest size die you can make. So we can only cram so much transistors into one. And the transistors are obviously mostly multiplies in AI salary. What we're doing now is we're putting 2 dies together in 1 chip. So effectively, you now have 1,600 square millimeters, double. That's a way of getting the double the performance. Well, the current generation we are working on is 4 die can just imagine, and the 1 after that probably will go to maybe no limit 8, 8 dying in 1 chip. So you had chip size is equivalent of, say, 800 time train, 6,000, over 6,000 square meet running as 1 single chip. The amount of technology, the complexity of technology needed to be created to enable that to run. And it's technology that's not just created at our wafer fab or with the memory guys. It's us designing it is interest -- it's very, very challenging, very interesting. And that the relationship we find very exciting with every 1 of our customers because we're able to keep enabling them to create more and more high-performance chip, which enables them to drive as a key part of their model of their stack much but higher, better, exponentially better and better LLMs which drives towards what I said before, more and more value out of super intelligence. And my view of it is with the ability to create those kind of chips and imagination and intelligence of the R&D engineers, the data scientists, writing those algorithms to -- and finding breakthroughs towards super intelligence, we do not see a limit at all yet towards the exponential improvement of those frontier models.