Altimeter的动态

Transcript

Well, Silicon Valley investor Brad Gerstner, known for early investments in Meta and Uber, the founder and CEO of Altimeter Capital, Brad joins us now here in Miami. Brad, always great to get your take, especially after yesterday's sell off. Where do you stand? Exciting times, exciting times, Great to be here. Great event is my annual pilgrimage onto this set and I love it. And it couldn't come at a better time. I mean, you know, NVIDIA today, I think you just said trade 20 times revenue. It's trading at 24 times earnings, right? People talk about the bubble. This isn't what bubbles are made of. And as far as I can tell, the world's going to be, you know, have a compute shortage for the next three or four years. I don't think anything yesterday changed that. In fact, what happened is when you give models like Llama. Very inexpensive, open source for free or a deep seat. I don't think there'll be 400 U.S. companies using an open source Chinese model that you have to send your corporate data to. But let's stipulate that the cost of intelligence is going to come down. That's a good thing. The amount consumed is going to go up. I heard yesterday deep seeks out scrambling to get more GPU's to support the inference because they went to number one on the App Store here, here in Betty is down 17% and the company that's causing it to be down is out there trying to get their hands on more. GPUs to support the inference. So I think there was a lot of hyperventilation yesterday and very few facts. Yeah, Brad, two months ago, about a month ago on your podcast, the BTG podcast, you guys, that's not Satya Nadella on The CW g ^2 podcast. Sorry about that. It's a great podcast, by the way. It's you and Bill Gurley too. Fascinating. Yes, investors. You got Satya Nadella, the CEO of Microsoft on the pod. You asked him directly, are you still chip constrained? Yes, he said no, I am power constrained. That to me. Guy and I were talking about it on our podcast. Followed on the podcast store they're people no problem I know well you get out of your system No, but we were talking about it at the time and like this is one of the largest companies early beneficiary you know investment in open AI two years ago that sort of thing and they're not chipped constrained anymore. Well I I think parse the words yeah he said I'm power constrained and what that meant is that he's more power constrained than he is chip constrained. Trust me all of these major companies are still chip constrained. There is a fight the biggest challenge Jensen. Hasn't he said that on our podcast? Is he? All these guys get angry with them and now Stargate comes along. We forecast it on our pod that if they build $100 billion next year in Abilene, which is the objective and I think, you know some differing opinions on that, we think it will happen. We estimated that's 2,000,000 incremental GPU's. The total number of GPUs produced this year are for forecast is 6,000,000. So you have a new player on the scene who wants to buy 30% of the total supply of GPU when everybody else. In the world also wants more GPUs. So again, what happened yesterday it wasn't a breakthrough on the model. They actually build a model that's kind of where open AI was 10 to 12 months ago the O one or the O1 preview model. But there was this idea that they did it for a fraction of the cost and I heard reported all over CNBC yesterday that it was 6,000,000 versus billions of dollars to train O one or O1 preview. I would say my informed speculation is that the cost of training O one or O. On preview was less than 20 million. This was not a big bright price breakthrough. It's actually what you would expect. The cost of compute comes down by about 50% a year. So if it cost Open AI a year ago to build this model, $15 million, you would expect it to cost 7 1/2. And we know they also use distillation techniques. It's not to take anything away from deep sea. I think it's a great thing for the world that we're going to, you know, make this cheaper for everybody. Everybody's going to have access as a consumer. All these enterprises are going to have access is not just going to come out of places like Deep Sea Llama is going to one up them. I imagine there are gonna be other companies like Open AI and Google and others who are gonna be open sourcing their lower tier models while you know, they still charge a premium for the frontier, right? I mean your words, people should listen when you talk and I'm one of those people. And you mentioned price to earnings is not only reasonable, but given where we've been, it's extraordinarily cheap. But my concern is in terms of NVIDIA, they're sort of out earning. Their revenue, in other words, the margins at 77% suggest that something's gotta give at some point. That price to sales number is still, for me, a bit of a sticking point. So thoughts on that? Yeah. I mean, listen, I think, you know, as an investor, I'm always trying to disprove my conviction and right, Like, what is the thing that would cause me to change my mind on NVIDIA? It's not the fact that we're going to drive down the cost of of intelligence, because I think Jevon's paradox, the actual amount consumed goes up. What actually would change my view is if I thought there was an alternative. Competitor who could produce accelerated compute in the world, you know, anywhere close to total cost of operation what they can do. So there was a lot of fear about custom ASICS and TPUS and Influencia and Graviton, you know coming out of of of Amazon. But what we see is the explosion and demand is so great that there are use cases for those custom chips. But still the number of GPU's ordered by all those hyper scalers are going up, including Google who uses their own custom Tpus. So I don't think. This is either or. I think it could still work for Broadcom, still work for places like Astera Labs who are in that complex. But you know, I'm just looking at the numbers. I don't see the level of competition at the frontier of compute today to Invidia. So if this pulls forward adoption of AI, what else does it pull forward? Presumably it pulls forward peak spending, peak CapEx spending. Maybe it distributes the games to software faster. I mean, what, what sort of changes in this, the things that get us excited, a couple other themes. One is, you know, people look, you know that we wrote the, the letter to media in the fall of of of 22 and people wonder how did, how did Meta double, you know, more than two X their revenues while going from 87,000 employees to 65,000 employees because they were leveraging AI, right? Every market leader can leverage AI to grow top line at faster, accelerate their top line while reducing bottom line and they're doing it across their businesses. It's accelerating today. The other one in software, we have 25% of our book. Software, everybody, last year Bill and I did a pod when everybody declared software dead right and Sacha came on our pod and said maybe application software is a thin layer on top of a CRUD database. Again, got everybody scared about software. Softwares trading below its 10 year average at 5.7 times forward, forward revenues at the peak in 21 it was the 18 times. So I think there's a lot of opportunity in the market today and AI is going to accelerate it. You know you mentioned Facebook levering AI and it's manifested itself in the last two earnings. Releases, Walmarts, the other company in terms which without question in terms of their margins. Any other companies out there that you've sort of noticed? I don't know tangentially you just sort of looking that it's starting to figure it out on the AI front, I was just with a CEO here before I came on set. He said over the last two years we've doubled our revenues and we've held OpEx flat. Leveraging AI, I think it's ubiquitous. There is not going to be a single enterprise in America that's not leveraging AI as Satya said on our pod to drive their top line and to hold the line on their bottom line. Who's not going to benefit, OK. What I worry about is if you're not a market leader and you have to you get your, your margins competed away in a commodity. Industry, even if you benefit right from AI, you're going to compete it away. It's going to go back to the consumer as a consumer surplus. Those market leaders can benefit from it and hold on to that incremental margin. Brad, thanks so much for stopping by. It's great to be here. Hopefully we'll see you before next year. Our annual pilgrimage. By the way, this I have to say real quick, all the shows on CNBC wanted Brad to come on yesterday and he said no, I'm going on fast money with Melissa Lee. So kudos to Brad. A lot of you guys have a great shot. You have a great show. Thank you. Love the show. Thanks. Prisoner altimeter BG squared right this time.