VMware & Custom AI Chips: How One Supports the Other in AVGO

**SPEAKER_1** (0:00)
So, joining us now to break down what it all means, and of course, expectations, is Stephanie Walter, Practice Leader of the AI Stack at HyperFRAME Research. Stephanie, thank you so much for being on the show today. Now, as we do head into Broadcom's earnings, what really are the most important metrics that you'll be watching to determine whether that momentum, say, is still accelerating, not waning?

**Stephanie Walter** (0:22)
Thank you for having me. Always great to be here. So, I think Broadcom's earnings really need to demonstrate sustained AI growth and strong outlook. So, Broadcom has two main business segments. They have semiconductors and infrastructure software.
AI creates opportunities for both of these, but in different ways. And I'm looking for growth in the AI semiconductor revenue. I'm looking for more customer projects moving into production and management expectations for continued demand. So, Broadcom supplies both specialized AI chips and the networking that connects the processors. So, understanding where growth comes from matters.
Expensive processors are less productive if they spend time waiting for data. So, Strong Quarter, of course, is going to tell its customer their spending, and that outlook helps us tell whether that spending is going to continue.

**SPEAKER_3** (1:17)
When you look across the business here and you start thinking about VMware, how important is that to this AI story?

**Stephanie Walter** (1:27)
It's very important. So, this week, and right now, this week VMware Explorer is going on right now in Las Vegas, and the announcements, if you've been following them, show how Broadcom wants to help enterprises put AI to work alongside their existing applications. So, they announced private AI Cloud, they announced an AI factory, extended model choice, agent governance through AgentMinder.
And so, what they're doing is they, Broadcom supplies infrastructure for AI build out, and then it's positioning VMware to help enterprises actually use that infrastructure. The test is going to be whether VMware and the related products make production AI actually easier to deploy, operate, and govern for these enterprises.

**SPEAKER_1** (2:19)
Okay, and I mean, I think that, of course, integrating VMware has been part of its broader AI strategy, so I mean, what really stands out then from some of the recent announcement we've seen that they want to continue to explore with VMware?

**Stephanie Walter** (2:33)
So I really think that the spanned AI spending depends on useful applications, not just more infrastructure, and applications are going to need that infrastructure software to basically exist. So if you think about it, AI demand extends well beyond just training models. Every time someone uses AI, infrastructure has to run it, and there has to be infrastructure software to manage it. So one agent completing a task may call several models and tools. So computing demand can grow faster than the number of users. So strong chip sales demonstrate infrastructure demand, but really sustaining that demand requires applications that deliver useful results, and VMware is a key part of Broadcom strategy for that.

**SPEAKER_3** (3:27)
Stephanie, probably the most important part, I guess, of the growth side of things is this custom AI chips business, and Marvell is a smaller notable competitor in the same space. They just disappointed.
Part of that was not having a lot of visibility on their alphabet partnership. But when you look at the custom AI chips business for Broadcom, that's got a lot of attention. What are we expecting out of that front? Is there enough room for these companies beyond NVIDIA?

**Stephanie Walter** (3:59)
I do think that enterprises want to have multiple technology vendors. It reduces the risk for them enormously.
Broadcom helping these large tech companies design chips around their particular workload is really important to them. A general purpose processor can handle many different kinds of work. A custom chip that's optimized for more specific and more very specific set of needs actually. And that can improve efficiency at a very large scale, though the custom chips do require a substantial investment, both in development and the supporting software. So if you look at them, they're not automatically cheaper than a general purpose processor.
And customers can use NVIDIA for some workloads and custom chips for others, depending on cost and efficiency and prioritization of the workload. So this is not a winner take all market. Customers want the right infrastructure for each workload, but they also need the economics that justify the investment.

**SPEAKER_1** (5:08)
Okay, and we also have seen this massive spending across the board, really into AI infrastructure in the industry. And so at what point then do we need to see real strong evidence that these various AI applications are generating real business value and say, not just driving this overall hardware demand?