Marvell Technologies has been one of the top performers this year, and the company split post and had an analyst meeting. They end all analyst meetings, and Matt, I've been with you when you had analyst meetings and no one believed in you, and they were wrong, and you because they thought you were too aggressive. They were wrong. So I want to ask you, Matt, what gives you the conviction to make projections that frankly were jaw-dropping, including that one by your CFO, 30 and 30?
30 for 30. That's right. Well, look, Jim, I've been CEO for 10 years. I've been in this industry for 32 years.
And now the silver, you have gray hair running. The picture doesn't match the one you guys flash up.
I probably need to redo the photo. But, look, I think we always put out what we think is pragmatic, judged, and achievable. And so, if you look, actually two years ago, we were in New York, same hotel, same venue, talking about our AI and data center ramp. And we were talking at that time about a $15 billion data center 2028 Marvell revenue. If you look at sell-side analyst consensus today after our meeting yesterday, the implied data center revenue in 2028 for us is over $30 billion. So basically it's fundamentally 2x'd in that period. Now along the way we've updated investors. We've been taking people on the journey. But yeah, we've achieved the targets along the way and the business continues to perform really well.
On June 2nd, a very famous prediction by Jensen Huang: you could be the next trillion-dollar company. It seemed absurd at 42 times the earnings you made. You will be at a trillion dollars. So it is not absurd, and it could be four years from now.
Yeah, it was not rehearsed and shocking when Jensen said that, but at the same time, if you actually take the range of 70 to 90 billion and you put a multiple on it four years out, you know, who knows what's possible.
But he saw it. And I need to know whether you saw it first that it wasn't — it was not the usual kind of GPU versus TPU. It was connectivity. You discovered connectivity. Explain to people what connectivity, which may be bigger for you than actual XPU, TPU.
Yeah. No, no, we're unique actually in this AI semiconductor cycle in that you had the three big compute companies, right, drive massive growth when generative AI first rolled out. That was followed by a huge memory wave, right, which we've all followed, right? Three large trillion-dollar memory companies, that's what they do really well. And what we're seeing now is there's a gap because there's not enough in the connectivity layer to connect all of those things. So you got a lot of installed compute, a lot of installed memory, but guess what? They all need to talk to each other through high-speed optical interconnections, switching networks, and what you don't want is any GPUs or memory sitting idle. And actually what happens when you start talking about agentic AI and reasoning models, there's more and more diverse kinds of compute. So the connectivity layer even gets more complex. So that was my June 2nd sort of thesis at Computex was, look, we see this as the next big unlock, and our business has just been ripping since. And that's when Jensen came on stage and basically said, yeah, I agree with Matt, his business is doing really well.
I mean, it was only April of '24, I think, Matt, when you were talking about your XPU custom total addressable market of $40 billion. Now I think you're saying it's $235 billion.
That's two and a half years.
Yeah. Yeah. I mean, we know the numbers, we share them every day with everybody in terms of the spending, but is it astounding to you, or what gives you the confidence that number is going to be right?
Yeah, the custom number, it really ran on us for sure. I mean, and it was interesting, if you go back two years ago, the debate was actually can custom even materialize as a portion of the market. Now you fast forward and it's not just us, right? That whole segment has clearly been solidified as a real thing. And so you roll forward just a few years into 2030 and yeah, you're looking at a $235 billion market just for custom silicon. We participate in that in two areas, by the way.
Well, you're 13% or so, aren't you? Something like that.
Well, it's pretty reasonable. It's actually the crazy thing is we said, 'Hey, our target is $30 billion in revenues, right, in 2030 in custom.' And then somebody pointed out in the meeting, look, if you just divide that by the 235, it's only 13% market share. And we've got customers there today. I mean, our business in 2028 is going to be $12 billion plus in custom already. So that drive from '28 to '30 is not a stretch and it's not a huge portion of the market. So I think that's why we gave some headroom too on our ranges. We said, look, it's not just an $80 billion target. There's a range of outcomes. We actually gave ranges on all of our businesses. So, you know, investors can take a look and they can make their own judgments, but we think the midpoint of $80 billion is very achievable at the company level.
That's interesting because the street's elated pretty much at all the targets, but I think it was Morgan Stanley that said they're happy, but we don't know whether Marvell needed to raise the bar this high. Was there any discussion about hedging bets?
Yeah, you know, it's interesting. There's always this debate: are you conservative? Are you aggressive? What do you like? And I actually never like to use those words. I always look at the most probable outcome, judged and pragmatic, and we call it like we see it and we give a range, and that has served me and the company, I think, extremely well for the last 10 years. You know, we've grown the revenue from basically low $2 billion in revenue 10 years ago. We're looking at $20 billion next year. We've taken investors on the journey, and yeah, I think you could say I've made some moves that would be aggressive — maybe forecasts we gave, or acquisitions we did, or divestitures we did — but in my mind all of them were very rational, thoughtful and pragmatic, and that's how we think about it. So no change in how we're running the place, no change in me. So when you look out to our targets in 2030, it's very consistent with how we've been doing it.
But you have a particular edge that people have to understand, which is that you figured out and know how to break through the copper wall.
Copper wall, right? I mean, if you think about it, if you go back like 15, 20 years inside the data center today, you look at a data center, there's optical cables everywhere, right? You see all these fiber optic cables plugged in and up. Those all used to be passive copper at one point. And because of the data bandwidth, the data rates required and the bandwidth required, more and more of the data center links have been moving from copper or passive to optical. So that's driven our business quite nicely to date. But there's another wave. Basically what's happening now is there's two terms. One's called scale-out networking and then there's scale-up networking. Scale-out is connecting all of the racks together, all the networking inside the data centers. But now people want to hook up and coherently connect GPUs and XPUs together.
Talking like in the rack, you mean?
