NVIDIA's High-Stakes Bet, Nuro Cofounder on Uber's Robotaxi Fleet & Google's AdTech Future | Aug 28, 2025

0:13Welcome everyone to the information's TI TV. My name is Akash Pasricha. It is Thursday August 28th. We have got a great show planned for you today. We are talking robo taxis and autonomous vehicles with the co-founder of one of the companies that Uber is leaning on to expand into that space. We've also got a great story for you about Google's ad tech business, but I want to start with the earnings that are moving markets today. NVIDIA put out its second quarter results last night. There was a lot to it between the guidance, the state of its China business, and even some disclosures about its cloud business.

0:47I want to bring on three people to help us make sense of all of that. Doug O 'Glocklin is the president at Semi Analysis. Anissa Gardizi covers the cloud here at The Information. And Jay Goldberg is an analyst covering NVIDIA at Seaport Research Partners. Welcome to the three of you. It is great to have you here. Jay, I want to start with you. You put out this really interesting report last night. I'm just going to quote from it because it really sums it up. You said, NVIDIA's results increasingly depend on sales of complete systems, entire racks, and sometimes entire data centers of NVIDIA configured systems.

1:24This holds appeal for some customers, but is anathema to many others, especially the hyperscalers. This was a really interesting point that we hadn't thought about too much. Tell us a little bit about what you meant there. So NVIDIA is an immensely technically capable company. They're doing incredible things. They have all this incredible innovation. And one big part of that is the way in which they are able to tie together not just a chip, and not just a server, but entire racks and then entire data centers. And Doug actually wrote a great piece about this last year where he's saying NVIDIA is basically increasing the Moore's Law growth of compute, but by pulling all those other levers.

2:04So technically, it's incredible. But commercially, they really want to sell you the whole system. They want to sell you a rack. They want to sell you a complete system with all the networking and all the cabling. And for some customers, that's just a non-starter, especially the hyperscalers. They want to do it themselves. They want to be able to customize to their own needs. And I just wonder if it's getting so complicated that that is creating some friction here. I know it's something that the hyperscalers are reluctant to adopt, but may not have a choice in some cases. And to be clear, is this coming from conversations that you're having with customers, or is this just a concern that you have?

2:38It's a little of both. Like, I think, I mean, my thesis, I'm, you know, I'm fairly downbeat on NVIDIA right now. The conversations I'm having with customers about their adoption of Blackwell, the latest GPUs from NVIDIA, it's just, it's hard. It's really hard to do, right? It takes a lot of work to configure. It's a lot of new skills for people that they're not comfortable with. And it just feels like there is a lot of friction that's slowing adoption in one way or another. Doug, let's go to you. So what do you make of Jay's thoughts here on NVIDIA trying to sell the whole system and some concerns he has with that?

3:14Yeah, so I had a really good piece about this last year called, I can't even remember the piece, but the fact that you think about it, you scale Moore's Law outside of the chip into the rack. And the point of it is that it takes the least amount of power to do the computation necessary. And so by using a copper backplane, doing a scale up, having a bigger world size, they They are solving the problems that AI needs to solve tomorrow, which is specifically having the biggest scale up world possible for inference. And this also helps for MOE training. I think while it's very complicated, it is the single best performance.

3:51And, you know, Jensen and Colette will tell you this on a token per watt basis. No one beats NVIDIA. And even though it's very hard, I do believe that the system scale up is the future. There is definitely quite a bit of problems with scaling the GB200 backplane and the GB300 backplane. But AMD is going this way. The TPU pods are systems as well. It might be a hard problem in system to digest, but everyone can see the writing on the wall. I think it's inevitable that data center systems is going to be the next unit of compute, not just chips in a rack, but the rack itself. So to be clear, you're saying, whereas Jay is saying this is a challenge for NVIDIA, that might be dragging them down.

4:32You're saying it's hard to do, but this is what everyone's doing, and this is the way they're going to have to do it. That's the point you're making. It's inevitable, in my opinion. Okay, so Anissa, we've got Jay saying this is dragging the company down. We've got Doug saying this is propelling the company forward. This is the future. Where do you land on this? Both are really good points. And I agree with both of them in different ways. I think the worry here that people might have about NVIDIA is that, you know, even though today these are the best systems, like Jay mentioned, they're very, very challenging.

5:07And customers typically, hyperscalers typically don't want to buy everything in a data center from one company. So I do think it's also inevitable that this will cause the hyperscalers to put a lot more resources into working on their own in-house data center equipment and also quickly looking for alternatives as well. Doug, I want to go back to you because from my view, the way I see it is this might be the way all these companies are going, but we hear all these conversations of customers saying, we are looking for alternatives to NVIDIA, right? That's why we're talking to these startup chip companies, these hyperscalers developing their own chips.

5:41And so you've got these customers looking for more variety there. Why is it that you think NVIDIA would be able to sell them increasingly on the whole system when customers are looking for more variety? Because they're the best. I mean, that's really what it comes down to. We can talk about all this, but there really hasn't ever been a single successful accelerator startup company. Name one that has any kind of volume. And then the custom silicon chips themselves are worse on a performance per power TCO basis, total cost of ownership. We think NVIDIA is going to continue to dominate the market.

6:16Now, I don't think we expect their level of market share dominance to continue for that same reason that you talked about. I don't think anyone is really interested in having one player have 95 % of the market. Now, that's where the custom silicon projects that most of the other hyperscalers are adopting or using or already have will come into the equation. But in the long run, I wouldn't be surprised to see the market share of the industry look something like 70-30. That's oftentimes how market shares kind of like end up. Think about the AMD versus Intel CPU of old, or even some of the transceiver markets today.

6:53Oftentimes you have one extremely dominant leader that has the best performance, and then you have a second source that is often not as performant, but cheaper. That's how I view the ASIC programs, and I think that that's probably where it ends up. Now, to be clear, 95 % to 70 % is a big gap, but I do expect NVIDIA to be the leader for as far as we can see. Jay, are you still the only analyst that you know of to have a sell rating on the stock? Yes, I am. Okay. And tell us about some more of the questions that you have about the company and specifically the ones that were actually bolstered from the earnings last night.

7:30So I think my my concern with Nvidia is very much near term cyclical. I feel like this AI, I mean, let's face it, we're in a bubble right now. I absolutely believe in a long term power of AI and to do incredible things. We'll get robots and autonomous autonomous cars and agentic AI. But I think it's going to take longer than the market is currently pricing in. So I agree with with Doug that like Nvidia is immensely capable, they're well positioned for this. I just think that it's this is all getting a little long in the tooth and it's getting harder and harder for NVIDIA to keep the momentum going.

8:04And it's going to take the broader industry a while to digest this massive wave of CapEx we're seeing. Longer term, the changes, but right now, it's, you know, we're up against some limits. Anissa, I want to go to an area that you've written about NVIDIA is looking to grow in, which is their cloud business. You noticed something last night in their SEC filings about how they're spending their money on cloud. Tell us about what you found. Yeah, sure. So a lot of big tech companies will disclose in their SEC filings how much they're spending on cloud computing, which is renting chips back from big cloud providers.

8:41Even though NVIDIA sells chips, it also needs them as well. And last year, I guess it was in 2023, NVIDIA started talking about launching its own cloud business, which is where it would rent GPUs back from companies like Amazon, Microsoft, and Oracle, and then resell them to its own customers. And last year, it started mentioning in its SEC filings that part of its cloud spend on the big hyperscalers was to support this nascent cloud business. It's called DGX Cloud. And yesterday was the first time we saw that they no longer said that their cloud spend was supporting the cloud business. They went back to just saying their cloud spend supports R &D.

9:22And so that's definitely something that I'm going to be digging into. it seems like the cloud business has sort of dropped in terms of its priorities, and the majority of NVIDIA's cloud spend is, like it said, going to R &D. Doug, how much do you pay attention to NVIDIA's cloud business? Quite a bit, but I think you can look through the public. You can look at what they've done recently, and the answer is pretty available to you. So, for example, they recently purchased Leptin. Leptin is effectively, we're going to use platform as a service, specifically as a moniker here. platform is a service provider that kind of brings up the lower level of neoclouds like the essentially the almost like commoditized layer of infrastructure so that they can have that dgx level service by providing software to commodity infrastructure players to then sell gpus into the market so i think that that's what we're really seeing here the pull away from dgx cloud is because they recently bought leptin to be their platform to push uh you know it's from the top down versus the bottom up.

10:23And so it's a bottom up enablement of the NeoClouds to then sell GPUs to customers. Jay, one of the questions I had for you, we haven't talked about the China angle to the earnings last night, which was honestly, I don't know, a lot of people, they go on Twitter, they say, you know, I control left how many times the word AI came up. I feel like we should be doing that with China and NVIDIA's earnings because it keeps coming up. You know, they were open last night, they said, look, we haven't included China in any of our forecasts. We've basically taken that business line or that region out of our estimates.

10:57And it got me thinking, you know, if I was a CEO and there was a side of my business that had basically stalled, a lot of other CEOs might try to not draw attention to that. And here you have NVIDIA talking very openly about what's going on with the China business. Why do you think they're talking about it so much when the business doesn't even exist right now. I think they're negotiating in public with the United States government, right? They're trying to put pressure on the government to make something happen. And they've actually been doing this for quite some time. Right. But the problem isn't the U.S.

11:30government. I mean, the U.S. government said, we'll give you the licenses, right? I mean, they don't know. But the problem is they haven't they've I think they've reached an agreement with the president. What they said on the call was that they need further documentation. And I think it sort of implied that they might even need legislation in order order to get the approval, right? Because the deal they cut was very much sort of high-level terms. Now somebody actually has to put it all in writing, and they're waiting for that. And that has slowed things down. And so they're trying to put public pressure on the U.S.

11:59government to get things moving. Doug, what do you think? I think that's correct. They're definitely negotiating in public. A lot of this, we're starting to see this happen a lot more often with the chip companies, especially with the Trump administration's deal-driven government, right? It's all about a deal. So I think that that's probably correct. The H20 clearly has demand in China. The ecosystem is a little bit more nascent. We've written quite a bit about it here at Semi Analysis, but we are, you know, I think if that deal is, you know, is inked, we will see some kind of China revenue. So one of the reasons why it's become such a big part of the earnings conversation specifically, I think, is because of how incremental it is to the estimates.

12:43One of the reasons why they slightly missed DC this quarter was because of H20. And the big, you know, X factor for a slight revenue beat next quarter is the 3 billion of H20 revenue, or I think up to 3 billion of H20 revenue that is possible. But what about the China government, which is the big question here? I mean, there's national security concerns from them as well. And you still have Jensen talking about this question. I don't know if he mentioned the government of China explicitly. I don't think he did on the call last night. But what do you make there about that issue that exists right now?

13:20Banning. Oh, sorry. Just repeat what you said. With regards to the security of the chips and the U.S.-China relationship? Yeah. Or with regards to the fact that Chinese officials and the Chinese government themselves are saying, don't use H-20s. we're trying to the latter the latter okay okay um yeah i think that that's a that's an interesting point and i think that this is in public because while they say that the reality is almost all of the all of the inference that's done in china is done on nvidia effectively and so i think they would more than happily take the compute but there is a level of you have to plausibly deny almost uh what you say in public versus what you do in private so if i had to guess they're going to say don't use it but in reality subsidiaries and affiliates of many chinese companies will be purchasing h20s great anisa last question for you talk to us about what came up on the call last night with respect to sovereign ai yeah so so jensen said that um sovereign ai it's something he's talked about for a long time it's essentially um different governments across the world um building their own ai supercomputers and ai data centers based on nvidia so that um their artificial intelligence can be based and owned in the country.

14:36He said that he expects that business to generate$20 billion in revenue this year. And I think I was kind of wondering how they define sovereign AI, but, you know, that's 20 billion isn't a small number. And he's been talking about sovereign AI growing for a while. So it seems like this year he does have high expectations for the business. Great. Well, I want to thank the three of you for coming on. We could talk for probably a whole lot longer about all of this fascinating perspective just before we go. And people listening to the podcast won't be able to see this, but I want you to give us a thumbs up, a thumbs down, or, you know, comme si, comme ça, as they say in French, of the earnings last night.

15:15All right. So, Jay, Anissa, Doug, just give it to me. You're on, I guess.

15:24Okay. Yeah. All right. Well, thank you so much to the three of you for coming on. That is Doug, Anissa, and Jay, and we will be talking more about that in the days to come. Well, earlier this summer, Uber said that it would test 20 ,000 autonomous vehicles over the next six years in partnership with two other companies. The first of those was Lucid, the car company. The second was Nuro, which builds autonomous vehicles software. And on the heels of that announcement, Nuro announced a few days ago that it has raised$203 million at a$6 billion valuation. I want to bring on Dave Ferguson, the co-founder of Neuro, to talk more about robo-taxis in this current moment.

16:05Dave, welcome to TITV. It's great to have you. Thanks, Akash. Good to be here. So when I think of self-driving cars right now, you know, one of the questions that I think every company is asking themselves is whether a platform play is better or whether a partnership play is the future. And just to explain to listeners what I mean by that, I mean, if you look at a company like Waymo, they're offering rides through their own platform and through their own app. You've also got companies going the partnership route with Uber. Tesla has basically said they're going to run their own platform. Nuro, though, is partnering with Uber.

16:41And so I want to ask you about how you got to that decision in your mind. For us, I think it was relatively simple. I think we wanted to focus on what we believed we could do very well. And so the partnership that we have with Uber and Lucid is very much each party doing what it does best. Lucid is going to manufacture amazing vehicles with self-driving hardware on them. Uber is going to manage them and operate them and then obviously provide the service to customers, which it does better than anyone else in the world. And Neuro is going to provide the AV tech. And so we felt like it was a really lovely marriage of bringing together all of the strengths across partners in the ecosystem.

17:21Right. And one thing that I've been thinking about is partnerships kind of seems to be the way Uber has struck partnerships with a lot of these companies. Do you think that Waymo, for example, will someday sort of get rid of its own app and sort of lean on partnerships more heavily? I think everyone's honestly trying to figure out what is the future of this ecosystem look like. It is such an enormous opportunity. And you know the space and the industry and the TAM associated with it. So it's definitely going to have multiple very strong players. And I think each of us, including Waymo and Uber, are really trying to navigate through the next few years to figure out what makes the most sense.

18:02So you mentioned Waymo and Uber. Waymo does run its own app and does a great job of that. in San Francisco where I live, I can catch a Waymo through their own app. But then they also partner with Uber in certain cities where Waymos will be on Uber. And in other cities, Uber actually manages the fleet for them. So already we're seeing even within the same two partners, multiple different possibilities. And I think what I personally think long term is that we're going to converge to a state where we do have incredible experts that are focused on pieces of that puzzle, Like we'll have fleet management companies that are the best in the world at managing fleets.

18:39We'll have obviously the ride share networks doing their part and we'll have AV companies doing their part. But we're going to see a lot of experimentation and also figuring out how it's all going to shake out over the next few years. What do you think of the Airbnb model that Tesla has talked a little bit about? I think that the personal vehicle ownership of L4 vehicles, so people personally owning cars that can then go operate driverlessly for mobility services, I think it's going to be incredibly powerful and it's going to be a massive, massive component of the future of mobility fleets. I think right now it's a little bit early and we're seeing some promising initial testing, but I strongly believe from an AV technology perspective that we do need LiDAR to be able to do this really safely and to get the performance that you want for full driverless vehicles.

19:32And so I do think we're going to see perhaps more fully equipped vehicles in terms of sensing technology than what we have today on the Tesla fleet. But I mean, just going back to where you started that answer, I mean, so you actually think that the Airbnb model that companies like Tesla have talked about where people are renting out their own robo-taxi or their own vehicles to be robo-taxis, you actually think that's a sustainable and a smart business model? I think long term, we're going to have mobility services like Uber that have some vehicles that are a dedicated fleet that they manage or another party, another partner of theirs manages.

20:09And we're going to have a significant portion of their fleet be personally owned vehicles. And so Nuro as a company is very focused on licensing our tech across both of those different markets. So we're spending a lot of time with OEMs to ensure that our tech is on future versions of personally owned vehicles, because we do see that as a massive part of the market in the future. Got it. Got it. Well, and I was going to ask you actually about this fleet management question. I was going to ask you what role you saw Nero playing in that. And so you guys are actually working with OEMs to sort of develop those fleets.

20:39You know, one of the questions I had for you is take me in the future. I mean, you have these fleet management companies. what are some of the big questions around that in terms of like business model, you know, you have to own these vehicles, right? Like, you know, maintenance, stuff like that. What are the questions that you think are unanswered right now about fleet management? I mean, I think we do have a couple examples of companies like this that are very successful. It's not all that dissimilar from today's rental car companies, right? It's massive capex. It's a lot of vehicles. You need to have depots for them.

21:14You need to maintain them, keep them nice and clean, and they need to be turned over very, very quickly. And so I think we already have a few pretty clear successful examples of industry leaders that are able to do that. I think that's what we're going to see in the future. And I haven't paid too much attention to the rental car industry, but rental cars, to me, I mean, there are a lot of them around, but it strikes me as a really difficult business model, though. It's tough. I mean, it's very CapEx and OPEX intensive, but, you know, there are a few businesses like that that are still very strong businesses and they're critical to the industry and what we all collectively want as future products.

21:54So I think they're going to play a very, very strong role. And, you know, we will, as in all other parts of the ecosystem, I think we'll figure out who are the best players that can do that most cost effectively and still able to make very strong margins or at least reasonable margins to build strong businesses behind it. Great. Last question for you, Dave. You raised all this money. What are you going to do with it in the next 18 months? Well, we have plans over the next 18 months to launch a robo taxi with Uber and Lucid. That's going to require quite a lot of effort. But 20 ,000 of them, actually, yeah.

22:26A lot of it will be going to that. But really building this technology and keeping it at sort of the highest possible bar from a safety and performance perspective requires a lot of resources. So that's where we'll be putting it. We'll be focused again on the autonomy tech and making sure that we nail that for our partners. Great. Well, Dave, thank you so much for coming on the show. I look forward to talking to you more as this space evolves. That is Dave Ferguson, a co-founder at Neuro. Okay, Google is gearing up for a court ruling on whether or not it will need to spin off part of its ad tech business.

22:59As it waits, the company has actually been taking steps already to make that vision more of a viable standalone business. That is according to a story that we published this morning in the information. I want to bring on Kathy Perloff, who wrote that story. Kathy, welcome back to the show. It is great to have you here. Hi, nice to see you again, Akash. Okay, so look, I'm going to do my best to sort of get the lay of the land here on ad tech. And just for context, Kathy and I were in the newsroom this morning. We were whiteboarding out how the ad tech industry works. And first of all, the thing that people need to know is that buying ads is an action, an auction process on the internet.

23:37Okay. And the way I understand it is you've got this auction. Okay. And on the buying side of that auction, you've got brands, you know, beverage companies, for example. They have to use companies with technology to help them get the lowest possible price for those ads that they're buying. On the flip side, you've got publishers, companies like news publishers, for example, that are selling the ads. And you've got another group of companies with technology that are helping all these websites get the highest possible price for those ads. Google plays on both sides of that. Okay? But the part that we are talking about today is the side that is helping websites like the New York Times, like news publishers, sell ads at the highest possible price.

24:22And you had a story today about how they're looking to change that side of the business. Talk to us about what you found. Yes. Well, thanks for explaining that all, Kash. It's funny. The Justice Department is spending so much time on this part of the market that it is a little complicated for the everyday person to understand. But yes. So basically, Google internally has discussed, you know, because of these antitrust concerns, like this business might get spun out. And how can we make it more independent? How can we make it more not dependent on the rest of the businesses that Google owns? And one thing they could do is have more relationships with brands because that's something that the sell side, this publisher set of technology, which is being threatened to spin off, doesn't really have those relationships.

25:10Right. They only have relationships with the publishers, the websites selling the ads. Exactly. So they have been trying to court advertisers to forge closer relationships with them. They had a dinner in New York. It's important to know this, you know, antitrust is one reason for this. Another reason is like this is kind of a trend in the industry and Google is trying to kind of honestly because they're so big one could say they maybe haven't had to move as quickly as some of their smaller competitors and this has been something some of their smaller competitors have done as well. But regardless they've been having dinners with agencies.