I'm reading the MRVL presentation at Goldman Sachs, and they discuss your topic.....
seekingalpha.com
Toshiya Hari
Got it. And I guess on that point on AI, I think this was the first time you disclosed AI revenue and sort of framed it for us. Just to level set everyone here and on the webcast, what is included in AI in terms of products? I think you sized it. I think last year was $200 million, this year more than doubling. So $400 million plus this year, what’s included in that? What’s – how did you define what’s included and what’s not included? And to the extent there is overlap with cloud optimized silicon, which was the previous and still the current definition that you guys speak to, what are kind of the...
Willem Meintjes
Sure. Sure. A couple of pieces there. Let me unpack that. So we decided to go through a very robust process to classify what’s AI and really looking at it in two main buckets. The one is on the networking side and then the other one on the fund custom cloud optimized. And so when you looked at networking, I won’t go through everything, but sort of two of the main things is really on the optical interconnect. That’s the PEM for 800-gig today going to 1.60. And really today, the primary application for that product is in these AI clusters. It’s not being rolled out widely. And so it’s a fairly straightforward classification. And so we do it in our system by part, by customer. And then something like DCI, the 400 ZR product, we look at – we sort of had a run rate business there. And then that’s just upsided very significantly. And so we had a conversation with the customer saying, what’s the end application? It was very clear there was being driven by the AI rollout to their different data centers. And so able to very cleanly sort of classify what percentage of that is going to AI.
And so in our system, we have tagged these things and have gone through many, many reviews and actually had our audit team review our process and all of our rule sets just to get like external validation on it. And so that we had a [indiscernible] into and so today, I can just log in and I can see that breakdown at any time of all of our revenue of all of our forecasts. So it’s very systematic. It’s not just sort of – and then the key point we make is there is some areas where – we know that there is some impact from AI, but it’s really early and it’s not very – so for example, storage, we just decided not to include in this bucket. So I think that was the first part of your question. Remind me what was the second question?
Toshiya Hari
I guess to the – well, I guess you touched on this, but to the extent there is overlap of the silicon, how should – because I think it led to some confusion.
Willem Meintjes
Yes, absolutely, yes. And so we are not giving exact numbers. But I think broadly, the way we framed it is originally, when we went back to our Investor Day we scoped the size of the cloud optimized opportunity is like 400 going to 800. And so when we looked at the lifetime revenue driving that revenue, we said originally when we scoped at about 20% of the lifetime revenue had AI application. And that’s grown so significantly that now 50% or more than 50% of it is actually from a lifetime revenue standpoint related to AI. And so that’s sort of a way to think about when you look at cloud optimized, what’s that overlap. And to be clear, on the cloud optimized side, it doesn’t include the optical interconnect. It’s the custom product. |
