00:00 Speaker A
Andrew, the uncertainty you're talking about there, when it comes to financing this data center, at least for some of the names, you would say, okay, that's uncertainty, no, it's not a real fundamental risk at this point.
00:15 Speaker B
No, I don't think that's the fundamental risk. I think with Coreweave, NeoCloud, and later Oracle, there won't be as much room to issue debt. Core Weese doesn't necessarily have that either, it doesn't really have the free cash flow to keep up with other companies. Of course, I'm lumping Open AI and Anthropic together as not having enough free cash flow to meet their data center ambitions. Well, those are all valid concerns, but I don't think it's going to slow anything down for traditional hyperscalers that have plenty of free cash flow, and they have a lot of room on their balance sheets to issue debt if they choose to go that route. Therefore, scaling laws still apply. We expect AI CapX trends to be around 50% year-on-year growth. Well, if at some point that number starts to go higher, I don't know if we'll come up with a higher number. Um, I think this is a little pothole of uncertainty that we're in, uh, evaporating and then a number of small accelerations, but it's really about revenue growth and network equipment revenue growth. I think the intonation is set high.
01:53 Speaker A
Andrew, before I let you go, uh, let me pick a few. Is it lying there because I was about to ask why Arista is being bought?
02:04 Speaker B
Okay, so, uh, scale up. This is what's happening now: scaling out, building more data centers, installing servers and server racks. Scaling up means adding more compute. This is basically adding it to your XPU cluster, so to speak, and attaching it inside each rack to connect the racks together. That leads to higher attach rates. It's fueling the exponential growth of switch equipment, and that's what Arista does. Arista makes Ethernet-based switching equipment, but when people are looking for GPU replacements, right? They obviously want to diversify away from Nvidia because they don't want to be trapped by price increases and things like that. For example, TPU architecture. More efforts will be made to lower inference costs, but this will only fuel exponential growth in networking and switching equipment. As we move from Nvidia-based Infiniband to Ethernet, we also have $2.5 billion of deferred revenue on Arista and all the Arista land, and I think that's going to happen over the next 18 months, two years.
03:16 Speaker A
Yeah, I'd like Andrew to throw in one more thing here. It made sense. The stock is currently up about 100% in 2025. Andrew, can I still buy it here?
03:35 Speaker B
absolutely. Yeah, I mean, Coherent, Lumentum, Fabrinet, these are optical transceiver companies. These are critical components as we need to gradually move away from copper-based UM equipment. And, you know, I think the transition from 800 gigs to 1.6T to 3.2 is happening much faster than anyone anticipated. And there's no chance after that. After going through this scale-up architecture, scale-up is achieved. That's after that. Therefore, data centers will be interconnected, which will require optical equipment. We expect Coherent to be the closest to being able to meet the surge in demand. Well, that would be our top choice.
