From the Golden Gate to Fisherman's Wharf, San Francisco is buzzing for AI. This stock must stay ahead of the trend. From cloud to custom AI tech, is Broadcom poised to outrun the competition into the future?
Boy, do we have a treat tonight. Broadcom, the $766 billion semiconductor conductor and AI infrastructure company, has been a real wild trader the last couple weeks. When Broadcom reported September 5th, even though their results I thought were terrific, their AI revenues came in a tiny bit light versus expectations, so the stock ridiculously plunged more than 10% the next day. Everyone pointed to this as still one more sign that the AI infrastructure story was dead. Then last week we got some very bullish commentary from Oracle's Larry Ellison and Nvidia's Jensen Huang. Turned out the whole group turned. Broadcom finished the week up 22%, and it's up nicely as people realize how strong the quarter really was. Can it keep running? Let's take a close look with Hock Tan, the legendary president and CEO of Broadcom. Mr. Tan, welcome back to Mad Money.
Thank you, Jim. Great to be here.
Well, Hock, the last time we saw you was before the VMware deal closed. That was a monumental deal, and it is really starting to work out. Maybe you can give us kind of a reintroduction to Broadcom.
Sure. Well, Broadcom has been around for... I've been running the ship for over 18 years. Started as a semiconductor company, and over the last six years, we kind of moved into infrastructure software, and that has gone very well. And the recent acquisition of VMware was essentially another step towards the direction of creating a very balanced mix between semiconductors, particularly focusing on enterprises, and infrastructure software, again, very much for enterprises.
Let's talk about the VMware acquisition. It turned out some people were concerned about it, about how it was performing, the execution, as always, because it's a Hock Tan company, was superb. But this was a breakout quarter. How come?
No, it was a jewel. The technology and the products and the engineering team which created the products, the core technology, has been around over 20 years. It's great, great technology, and it's very, very critical for infrastructure, especially for large enterprises running applications on the data centers on-prem. That's what VMware is all about. And what we've done is improve the products, make it much more usable from just being compute virtualization to basically creating a virtualization of entire data centers on-prem in enterprises, and basically creating the same cloud experience privately on-prem.
Well, you're the only one, which is one of the reasons why I love the business so much. So let's talk about the ever-growing AI opportunity. At the beginning of your year, Hock, you were talking about AI revenue at $7.5 billion. Then you revised that to $10 billion in March. This time you raised it to $11 billion in June, and now it looks like $12 billion. How can this be growing so quickly?
It is because we are in a segment of the market in AI where we are addressing several hyperscalers. And these hyperscalers have very strong incentive, ambition towards building, continually investing in large language models to basically create models that are smarter and smarter, and eventually embed into their huge, large subscriber-based platform. And they have to keep investing, and they are really investing. They're not just proving it out. And one of the things that drives them to invest is they want to create their own, what they call accelerators, they call it XPUs. We are right in the forefront of enabling these few hyperscalers to invest and essentially start building up these large AI clusters of XPUs. And this is a roadmap over the next three to five years. This is not a one-year phenomenon. As we see it, it's a journey for these guys, but it's a very urgent journey, and it's a huge opportunity.
I'm glad you talk about it being a longer term, because some of the analysts were caught up with the idea that maybe it's slowing, perhaps because they didn't realize that one or two customers may have asked for a little more customization, because this stuff is really dependent upon you. They can't just go, it's not a commodity what you're doing. But the idea that it could be slowing seemed to be completely fanciful.
Far from it, actually. As I said, this is a very focused and it's a long-term, it's a midterm plan, three to five years. Consider midterm could be long. We see that roadmap over the next, for these hyperscalers over the next three to five years, and their intention is to build out these large clusters because what is needed to drive those large language models, those foundation models owned by these hyperscalers, into, you know, into improvement, to higher performance, in what they call higher IQ of their models, seriously over the next five years, is to keep coming out with a new generation. Each new generation LLM requires multiple Xs, two, three Xs, maybe more, of compute each time, each year. So you can imagine that's a driver towards a larger and larger compute opportunity, which is going to be taken up largely by XPUs.
I'm so glad you put it that way. So many people are saying one and done. That's the false narrative. The narrative you just laid out is the one that's going to keep Broadcom going. Now, I had been concerned about the non-AI hardware, wireless revenue, some, what I guess you could call legacy. On this quarter, you said not only would there be green shoots, but you actually called a bottom. Now, you hadn't called a bottom before, so I know this is a serious thing that you did. Do you really believe that the so-called legacy is turning?
