00:00One of the stars of the show has been Neuro, right? We saw this big announcement coming from
00:04we had Lucid on the show, but Uber and indeed Neuro teaming up to provide the future of robo
00:09taxis. But it's a crowded field. How did you decide Neuro was going to be helping win in
00:14that autonomous fight? Sure. So if you rewind time and you look at the history of Uber,
00:19you'll remember 10 years ago, Uber had announced that they were going to invest in self-driving
00:23car technology. Now, a lot of things happened since then. So they originally had planned
00:31it with Travis Kalanick at the helm of the company and then had to jettison that whole business
00:35before going public under Dara. And I think what Dara did is pick that back up and say,
00:41we need to accelerate our strategy. So about two years ago, a lot of those plans were put
00:46into motion. And what was interesting was Neuro was a company that had been, one of the few
00:51companies like Waymo that had been training around Bay Area streets with their LiDAR and their cameras.
00:57And they recently launched a autonomy intelligence platform, which is basically an operating system
01:03to allow any third-party automaker to have self-driving capabilities. And so what you saw here
01:09was a great launch between Neuro, Lucid and Uber to not only fit, retrofit a Lucid car with autonomy,
01:18full level four, not 8S, right? Not level two plus plus assistance, but full self-driving
01:24capability like Waymo and Tesla FSD. And I launched that with Lucid on the Uber network. So you'll be
01:30able to open up your Uber app at the end of this year in the Bay Area and be able to take a self-driving
01:35car.
01:35It's like a capital-like model.
01:37Yes.
01:37Let someone else take on the burden of the hardware and the stack, just do the fleet management.
01:41But for me, the bigger piece of news in that world was Alpamayo, right? Because NVIDIA is coming
01:46with the same thing, saying, hey, we have a full-stack solution. I know that you caught
01:50notice of that.
01:51So AGX Thor, which is inside of Neuro, is part of that...
01:54It's like the inference card, right.
01:56And so what NVIDIA is doing, and it's very smart of NVIDIA, they are selling chips, right?
02:00And so what they want to do is sell you a starter kit. They want to give you an open-source
02:04model. They want to give you open data sets. And they say, go to town. We want you to build
02:09a self-driving car platform to fit these cars, to fit your ride-sharing or delivery network,
02:14your logistics network. And what they're saying is, we want to have many competitors out there
02:20so we can sell you more GPUs. And so Neuro is actually one of the very first partners that's
02:25been using that for years now. NVIDIA is an investor in Neuro. And so Neuro and Uber are
02:31sort of the prime example and hopefully a flagship for them and their business. But NVIDIA on a
02:36larger scale at this show, what they're doing is they're king-making. They're king-making an
02:41entire sector. They're saying, you know, a year ago they were talking about robotics.
02:44This year they transitioned that into physical AI. Now they're not just talking about robotics,
02:49they're talking about autonomy and mobility because it's coming to a crescendo. And so
02:53they're very, very smart in saying, we're sort of like the Fed, right? We're the Fed for AI.
02:59We're going to tell you... That's the headline, right?
03:01Yeah. And we're going to tell you where everyone's going to go. You may be on that wave,
03:06you may not be on that wave, but you should be. And by the way, we have the models, we have the
03:10data, and we have the chips and overall compute for you to go build that. And so, you know, I think
03:16there's been a lot of argument about, you know, is NVIDIA's kingdom, right, at risk right now? And I
03:22would say that they continue to surprise and be several years ahead of their competition. So we're
03:28very excited, but we're watching closely.
03:29I mean, look, you were known for backing Uber, and then ultimately there were two key winners out of
03:33that, Uber and Lyft. When I think about the areas that you're also thinking of, I mean,
03:38how many players can there be in autonomous? How many players can there be in foundational
03:41models? You're in perplexity and everyone's thinking maybe there's only an anthropic and
03:45an open AI and a Gemini space.
03:47Sure. As you're rightfully noting, any innovation curve starts with many competitors,
03:51and it whittles down to two or three. It's just a rule of thumb in Silicon Valley. So right now,
03:58Neuro is entering the space. There's Waymo, there's Tesla, and there's Neuro. In China,
04:03there's several other competitors, right? In Europe, there are none, right? And so on level
04:08four capabilities. So what I think is happening is you're seeing a whittling down because of the
04:14resources, the training data, the amount of years that you need to do to train across all these
04:20streets. It's not a software company that you can just start up out of scratch. So I think one of the
04:26things that you'll see also in robotics is you'll see this whittling down of competition,
04:32and you see Boston Dynamics, which is owned by Hyundai, which is very interesting. They're
04:37doing quite well, and they revealed a humanoid. You know, I've been very careful and skeptical
04:43about the development of humanoids, and I would say that this is the first time I've seen humanoids in
04:47action that look practical. And so I think you'll see that in robotics as well. I think you'll see
04:53that in generative media. I think you'll see that in agents. So a lot of these platforms will start to
04:58whittle down to two, three, maybe four players. And this year will be the year of that, which is why
05:04inference, you know, one of the other things that we're seeing here is that more further evidence
05:07that inference is going to exceed training in terms of compute. And that flip that flippening,
05:13if you will, right, if I can use a Bitcoin term, that flippening is means a lot. So demand for
05:19compute is far exceeding supply, even with the HBM shortage, it's going to increase that. So there's a
05:26battle royale happening here in compute.
Comments