Health care made just for you. Start today at Fortt.com.
Look at the stock of Broadcom go this morning. This big time maker of chips, networking equipment, and infrastructure software announced a major deal with OpenAI. And the stock shot up nearly 10% in response. This is just the latest in a string of big data center deals that make me real glad we own Broadcom for the charitable trust. Luckily, we're here in San Francisco, so I got a chance to speak with the one and only Hock Tan, the president and CEO of Broadcom earlier today. Take a look. Hock, it's an honor to have you on the show on a very big day for your company.
Jim, thank you for saying that. It's a privilege to be here.
Oh, thank you. Now, one of the things that people want to dispel something immediately. There are a lot of people who make these criticisms about what's really going on here. I know you, as you say, I am not a cheerleader. As you said in the Goldman Sachs conference recently, you're not a cheerleader. You're a hard-nosed businessman. You're not doing this because you hope something will happen. You're doing this out of necessity. It's needed.
Absolutely. That's a real hard business case for doing what we're doing, which is investing and enabling those select few customers that we chose to work with in enabling them to run their LLMs. And the biggest thing they're short of is compute capacity.
Now, one thing that did concern me is, to put in perspective how much power you need, you're talking about nearly 18 Hoover Dams worth of power. Who's doing the back end for you? Who are you making sure that the power is available? I know you wouldn't just dream it up.
No. There is a lot of power out there. Even in America. There still is. It's just that you need time to get there. There's available power. We hear a lot about it, but what's needed at the end of the day is usable power. And that comes. There's a lot of available power. We've seen that. We've seen the announcements that Oracle made on finding power and making deals with OpenAI, and more quietly, I know for a fact people like Google, Meta, and Microsoft have access to available power. The key now is to make these sites usable. And a lot of it comes not just one big. We hear a lot of news about one gigawatt, two gigawatt. Sure, it's great to have that, but you don't need that for generative AI, especially in the case of inference. You want it distributed so you can find lots of locations on the power grid for 50MW, 100MW, 200MW. And they're very available now, less and less so, no doubt. But with time, this becomes available and becomes usable power, right?
I know as always, you are thinking ahead about what will happen. I don't think people realize you. People say, oh, they play this guessing game, oh, is it going to be this company? Is it going to be that company? Maybe it's Anthropic. Maybe people should be looking at it differently. How many companies can do customized silicon at scale? How many would you work with? And always, you don't do anything where you sacrifice the margin for the shareholder.
Absolutely, Jim, and we've been thinking about it for a while now, and we've been learning, especially the learning experience, for the last 7 or 8 years with Google, who is the first player we have engaged with on creating custom AI accelerators. And that was back eight years now, and we're still engaged very deeply. And both of us learn as we progress during that eight-year journey. And what we have chosen to do is we look at the market, the players in it, and we chose to really address and select a group of players who are creating superintelligence. I call it all generative AI. Another way of putting it is these are the guys, the source of intelligence, creating a better and better generation of models. Foundation models called LLMs. You don't do LLMs, you're not in this game.
Right now, no one else is in this game at this scale. You are. I am concerned. You've always been a purchase order outfit. In other words, no work until you get a purchase order because you are a business person, not a dreamer. With this deal, we don't really know the economics with OpenAI, but I know you wouldn't go into a deal unless you're satisfied that they will work. OpenAI is a private company. Can you give us some assurance of what you see that makes it so that they're definitely worth working with?
Well, for one, Jim, I mean, they are one of those few players in the forefront of creating foundation models. And best I mean, they are private, but I think there's a sense of their choice of being private. But they are valued. The last valuation, even as a private, over $500 billion, and their revenue is growing at a very strong rate. And so, as I said, the first step you want to look at this whole game is how real is generative AI. I think that's where it begins. Then you look at the key players in it who are going to be there over the next three, five years, and these are the people we select, and they like us, too, to be fair. And we're privileged to be that way where we invest in technology with those guys. And you say we're very hard-nosed businessmen. Yes, we are. We look at it very hard. We calculate it. But that doesn't stop us from looking out 3 to 5 years and looking at this phenomenon, this wave called generative AI. And what it is, is trying to create intelligence through foundation models, which, by the way, in a lot of cases is a bunch of equations and algorithms on a whiteboard, which until more recently, a few years ago, back when the ChatGPT moment happened, where you can translate what is in theory on whiteboards into actual intelligence through models, by creating compute capacity that can actually drive, train, and create basically intelligence through this whole process. So we believe then we picked about seven players today that we are working very closely with. And four of them are now real customers. And we define customers in a manner you indicate, Jim, as those guys who have given us production purchase orders at scale.
Now, I have to believe that the others had the ability to give you a purchase order. It's just that I can't ask you to ask customers who haven't been finished yet. That's not fair. But your confidence in this narrow group of customers is incredibly high.
Yes, we feel very good about it because each of these guys need a lot of compute capacity for them to basically play in this game and eventually win this game of creating the best foundation model in the world.
Right? I know you say from outside, on hardware compute capacity, to some extent the rubber hits the road on this. It's really incumbent on us to keep optimizing, pushing the envelope on leading edge technology. So you have to stay ahead of the other guys.
Well, to be honest with you, the main competitor we have is merchant silicon because they were there first, right? Outside of Google, they were there earlier.
I don't want to get a cage match between you and Jensen. That's not paying off.
