Tesla Eliminates Miami Safety Drivers, Nvidia Teams Up With Rival, SK Hynix to List on NASDAQ

0:13Kevin McLaughlin:Welcome, everyone, to The Information's TI TV. My name is Akash Pasricha. It is Wednesday, July 8th. Today on the show, we have got inside reporting on how Tesla has changed its rollout strategy for robo-taxis in Miami as it tries to expedite launches. We also have exclusive reporting on NVIDIA's new growth strategy, taking a collaborative approach with fast-growing chip startups. We'll then unpack HubSpot's reversal of a controversial decision involving customer data. And finally, we're going to close out the show with a conversation with UBS about what is ahead for the AI IPO pipeline. It's going to be a great show, so let's get right on into it.

0:52Kevin McLaughlin:Tesla is quickly expanding its Robotaxi rollout, but it is making some changes on the gravel with respect to how it plans to start in new cities. Our Elon Musk reporter Grace Kay reported exclusive details on those changes. I want to bring her on to share with us what she knows. Grace, welcome back to the show. It's great to have you here. Yeah. So the RoboTaxi launched in Miami on Friday, and it's looked kind of different from other launches. So tell me, I mean, it's sort of interesting. I mean, they have rolled it out before. Texas has been the main state, I guess, for the rollout, right? Yeah, they initially rolled it out a little over a year ago in Austin, and then they've kind of expanded into Dallas and Houston.

1:36Kevin McLaughlin:Okay. And so how is the Miami rollout being different than from the Texas rollouts? Yeah, so the Miami rollout was the first time they've rolled out in a state without starting out with a safety monitor, which is someone who would sit in the passenger seat and kind of oversee the vehicle, you know, during the early release. And in Miami, they just released it. There was no one in the car, completely driverless. They also kind of in the lead up to it, I found from talking to some sources at the company that like there was a smaller test driver team in the city leading up to the launch. So a year ago when Tesla launched in Austin, they had more than 300 drivers.

2:14In the weeks leading up, they brought in people from all across the country that they were working with to test out the area for the robotaxi. But that was much smaller in Miami, which was an interesting development and kind of like an evolution of their work with the robotaxi.

2:30Kevin McLaughlin:And so the fleet of drivers, I guess, what exactly was the role for these drivers? When they launched a new city, they were meant to, what, supervise the robotaxis from inside? or you know was it meant to just be there as a safety precaution like what was the role of these people initially yeah they kind of have a dual role so before it launches they have test drivers who go to these cities and they you know validate they test out the software in different scenarios you know how does it react to construction zones or you know maybe this really difficult turn on this one street you know in the city um they do a lot of that work and then after they launch initially in Austin, they had people sitting in the passenger seat.

3:18And a lot of them had their hand on this button on the door that opens the door, but they had re-engineered that button so that it was an emergency stop button for the vehicle. So they were just kind of an extra safety measure when they first launched in Austin. And when they launched in Miami, they launched without that and with a little bit less testing leading up to the launch.

3:39Kevin McLaughlin:okay can we just talk about how funny it is have uh for a driver to have their button on the door i mean it's it's it sounds like a like an emergency like uh exit you know in some cases like which is kind of a funny sight it's like get me out of here now pronto um but okay so they so they've they've cut the the number of people then that are doing uh these testing rollouts. What's the reason here for this decision? Yeah, I mean, I think it's just kind of we're seeing how Tesla's process for deploying the robotaxi is evolving over time. Obviously, Austin was the first place that they rolled out.

4:19So maybe the strategy has changed a little bit there. It's also kind of reflective of Elon Musk's kind of view for the robotaxi. He has said that like he wants the software to be able to deploy, you know, without having to test everywhere it goes. So he wants the cameras and the AI system to be able to figure out roads that it's never driven before, you know, on the first go. And, you know, obviously they did have test drivers in Miami in the lead up, but like maybe having less is kind of another step towards that.

4:48Kevin McLaughlin:Do we know if there have been any issues with the rollout, given that they have sort of taken a more streamlined approach to this launch? I think it's too early to tell because we're only a few days into this. I think we'll know more and a few weeks or a few months. It's also important to know that it's like a very small area in Miami. It's about 10 to 14 miles. So much smaller, actually, than when Waymo launched in the city. And Waymo has since expanded to over 100 miles. So, you know, it's a little bit easier to control a robotaxi in a small geofence like that. Have you sat in both a Waymo and in any of the robotaxis at all before?

5:30Yeah, I've ridden in Waymo's, I think in Phoenix. And then also Tesla has a ride hailing service in California, which is kind of separate from its robotaxi service. And I've ridden in that as well.

5:42Kevin McLaughlin:And what are the initial reviews, not just from you, but from other customers on how the product performance differs between the two companies? I mean, put aside the fact that you need a driver in the car with you in California for those robotaxis that are autonomous. Do they live up to the Waymo reviews so far? Yeah, I think so far the reviews have been good. I think the main complaint people have had with Tesla service is some pretty long wait times, especially in California where they have to have someone behind the driver's wheel. And part of that is just like there's this huge operational effort to have these, they call them AI operators, but like the people who, you know, sit in the front seat in California, you know, they have to hire those people.

6:31They currently have almost 2 ,000 people registered in California for that, and they're still having relatively long wait times.

6:39Kevin McLaughlin:Grace, as you think about the reporting questions that you would like to pursue in this story going forward, what are the big ones that come to mind? Yeah, I mean, obviously, I'm very interested in how they continue to scale. I'm interested in how quickly they're going to be able to deploy in other cities. And there's also this role that I think is very interesting. As they go driverless, they have these remote operators who are kind of like the safety line. So even though there's no one in the vehicle, they have workers, you know, who can remotely take over the vehicle in these scenarios. And I think it'll be interesting how that comes into play and also to see how Tesla's strategy, which is so different from Waymo, plays out.

7:18So Waymo has, you know, cameras, LiDAR and radar, and they do very extensive mapping before they launch in cities. And Elon Musk, you know, for Tesla has focused its cameras only. And, you know, it's this AI system training on the cameras. And, you know, they want to do substantially less mapping. So it'll be interesting to see, like, how those two strategies play out and which one wins in the end. Great.

7:43Kevin McLaughlin:Well, Grace, I want to thank you for coming on. That is Grace K., our Elon Musk reporter, here at The Information. NVIDIA has a new growth strategy that involves partnering with fast-growing chip startups. That is according to exclusive reporting from The Information. I want to bring on my colleague Phoebe Liu, our NVIDIA correspondent, to walk us through what she knows. Phoebe, welcome back to the show. It's great to have you here. Great to be here. Tell me about this new growth strategy that NVIDIA is pursuing. Yeah, so basically, I think a lot of the narrative around NVIDIA's market positioning, I guess over the past few years and how they've been able to really dominate the market for chips that train and run AI has been around how they want to kill competition and kind of talking about aggressive tactics to stay dominant.

8:30A few years ago, at least in NVIDIA's perspective, their attitude towards specifically smaller chip startups has shifted. I think starting with NVIDIA's acquisition of networking company Mellanox in 2019 and then NVIDIA's Blackwell architecture that expanded beyond just selling GPUs, Nvidia has started to basically signal that they're a platform company and not just a chip company. And I talked to Nvidia Senior Director Dion Harris for this, and he basically said just that in an interview yesterday. I was like, you shouldn't think of Nvidia as a chip company. We sell so much more than that. And that shift has kind of coincided with a rise in specialized AI chips made by other companies, including two of NVIDIA's biggest customers, Google and Amazon.

9:23And I guess the TLDR of this strategy and my story is that NVIDIA appears to be accepting a future where there are lots of different types of chips in play in AI, not just NVIDIA GPUs, even though they're still dominant and increasing market share right now, according to some estimates I did a few weeks ago. And basically, NVIDIA CEO Jensen Huang has increasingly drummed up NVIDIA's other offerings, so like networking and things like that. I think on one earnings call earlier this year, he called NVIDIA the biggest networking company in the world. Presumably that's just for AI, but that gives us a sense of kind of how he's thinking.

10:05And I guess if there are different chips and increasingly people are talking about the possibility of disaggregating or running the same AI query across different types of chips for maximum efficiency, those chips would still likely run across a GPU because those are still the best for very compute-intensive AI uses. And NVIDIA would, of course, love for that to be an NVIDIA chip. So that's kind of the background. Right.

10:31Kevin McLaughlin:And so, I mean, the gist of the reporting, the most interesting part is they've actually pursued partnerships then with really, not really small companies. They're not tiny companies, but Dmatrix, I mean, this is the company that covered the information for a while, company that probably fell into the bucket of the NVIDIA killers, maybe at one point, you know, the rivals. So they're partnering with Dmatrix now. What's going on there? Yeah, so funny story. I think according to a couple of people I was talking to, NVIDIA actually reached out to Dmatrix first, even though, so Dmatrix is a chip startup founded in 2019.

11:10I think they last raised at a$2 billion valuation in November, which is big, but nothing compared to NVIDIA, which is the biggest company in the world, even though their Dmatrix is raising again right now. But they're essentially pre-significant scale or revenue. I think Dmatrix's CEO, Sid Sheth, said they're about single-digit revenue right now, and it's mostly kind of proof-of-concept customers. This partnership with NVIDIA kind of takes that into a real customer, which is Parasail, an AI cloud startup that aims to serve just using AI, not training it, for kind of other startup customers. And basically, in this partnership, a D-matrix chip tray can be hooked up to an NVIDIA GPU.

12:02So in this case, from the Hopper and soon Blackwell Generations.

12:07Kevin McLaughlin:And it's pretty simple, right? It's just an Ethernet cable. It's not that hard of an integration to do. How does the business model work, though? So if they're selling this jointly, then do they both generate revenue on a sale of this kind of joint product? That's my understanding, yes. So I think, yeah, so they would both generate hardware revenue, and then Paracel will then rent that capacity out to its customers and then get basically rental revenue from that. But the interesting thing is, this is why it works with Paracel's model. they sell based on kind of tokens per second instead of like GPU or D-Matrix Corsair chip dollars per hour.

12:52So it doesn't really matter what hardware the workload is running on. Customers are just paying by token. So that's kind of how the business model works.

13:02Kevin McLaughlin:Tell me, is this common for chip companies to partner with other chip companies? I mean, you know, the way I understood it, and maybe I haven't paid enough attention to it, but customers will take different chips, you know, because you've got GPUs, CPUs, you know, you need memory, you need all these different chips. Customers will be the ones to integrate all these components on their end. It seems to me that it's rare, though, for companies to go to market together with joint products. Am I off base on that? No, not at all. I think this is a new phase of this kind of disaggregation of running AI that's really starting to uptick now.

13:44Like no one's really doing it at scale yet. I think NVIDIA really set this off with the licensing deal slash quasi acquisition of Brock in December. Although that business model is a little different because there's a financial component from NVIDIA. But I think that kind of proved to the market that these types of partnerships where you have an NVIDIA GPU running alongside something else that's specialized, usually specifically for very fast AI inference, that could potentially work. There are a couple others. NVIDIA is also working alongside Samba Nova in a partnership that also involves Intel, which I think we reported on a few weeks ago.

14:30And there's also a partnership between Amazon's Tranium chips and Cerebris' chips. All of these are still kind of in the proof of concept testing stage, though I don't think anyone's really using it at scale yet.

14:42Kevin McLaughlin:Let me ask you a question, Phoebe, here. Do you think that this type of deal is good for chip innovation, broadly speaking, or does it just mean that the big are getting bigger? Yeah, so I guess obviously just my opinion here. I think because the bigs are so big, it's hard for the small chips to kind of break through without support for the bigs. So I think this is kind of a helpful type of partnership to help startups actually get more market exposure, at least in my ideal world. I think a few sources of mine have used the phrase coopetition, and they're basically saying that that helps stimulate the market because making a chip is so expensive and relies so many kind of supply chain strings to be pulled.

15:33It's definitely helpful to have someone with kind of a lot of punching power in your corner.

15:39Kevin McLaughlin:Great. Well, Phoebe, I want to thank you for coming on. That is Phoebe Liu, our NVIDIA reporter here at The Information. HubSpot reversed course on a controversial decision on how it plans to use customer data for its own AI tools. My colleague Kevin McLaughlin, our enterprise software reporter, has the latest on that story. I want to bring him on to share with us what he knows. Teva, welcome back to the show. It's great to have you here. Thanks, Akash. Okay, so tell me about this debacle at HubSpot. What's going on? Yeah, so this sort of flew under the radar a little bit, but on July 1st, HubSpot sent customers an email basically saying that it planned to use some of the data that they store in their CRM records to power a new AI feature that would make it easier for them to find sales leads.

16:26What made this sort of seismic for customers was that CRM data is very proprietary to them, and it takes a lot of work to find it and verify it and clean it up. And so that didn't sit well with customers, and four days later, HubSpot reversed course.

16:45Kevin McLaughlin:What exactly is the tool that they were trying to train? It was like an intelligence feature that took advantage of the customer data here, or what was it? Yeah, so just for some context, for the last couple of years, HubSpot has had AI powered features that automatically update customer records. Like if their customer changes companies or changes roles, it would update their titles and emails and things like that. And so customers were fine with that because HubSpot told them that it gathered the data to power these features from public sources, third party vendors, sources on the internet. And so it wasn't a case where it was a threat to customers.

17:26What changed and what sort of fueled the controversy here was that, as I mentioned, this time around HubSpot basically said, well, we're going to be using some of the details of your CRM data. Now, HubSpot did say that this was only business contact, business card type info, like company names and roles and stuff like that. But it didn't matter to customers because, as I mentioned before, they're very protective of this data.

17:52Kevin McLaughlin:Right. And the data, I mean, you outlined it in the story, but this data here, I mean, this is kind of, these can be years of notes in some cases on customers and, you know, some of the secret sauce, I guess, that companies hold, right, with respect to all their customer information. I think it was more than the actual nature of the data that was being used by HubSpot was HubSpot actually said that we're going to basically use this information and we may share it with other customers. And so if you're a customer and you spent five years sort of pruning your CRM data to the point where, you know, it's really, it's perfect and it's working for you, why would you want to then have your competitor have access to that data through HubSpot's product, which is of course available to everyone?

18:43Kevin McLaughlin:Got it. So it sounds to me that it wasn't, the training part actually was just sort of level one of the concern. It was the potential sharing that was the real concern here. Yeah, this is a little different. We've seen cases in the past with companies like Slack and Zoom where they had to update their and clarify their privacy policies because sort of internet sleuths looked at their terms and said, oh, wait a minute, they're actually reserving the right to train AI models, or at least it appeared that way. What HubSpot is doing is sort of different. They actually did change their terms of service, their privacy policies and other customer agreements.

19:20This was a comprehensive across the board change. But HubSpot's view was like, hey, customers, we're actually like going to make it easier for you to find sales leads. So I think HubSpot sort of misjudged, I think what the reaction would be. And I think that, you know, as they said in their communications HubSpot, we're trying to help you customers, but customers didn't want to be helped in this case.

19:45Kevin McLaughlin:Talk to me about what you think this tells us about the customer data war story, broadly speaking. It seems like we have these flare-ups, I guess, every couple months where an enterprise software company tries something, it gets some pushback, in some cases bigger than others. Was this sort of a – it feels like this was actually a step further than we'd seen in terms of what a reversal could look like, in terms of what a customer pushback could look like. What do you make of it? Yeah, it's pretty rare to see a company backtrack so forcefully just four days after implementing a new change. That's one thing.

20:27I will say that we've seen many examples. We know that many customers are very sensitive about the way that vendors use their data. So that's nothing new. And this is definitely, there's an element of that to this situation. I think on another level, HubSpot is one of the hardest hit companies by the SaaSpocalypse. And one reason for that is they sell CRM software to small and medium businesses. One of the things about CRM generally is that it's very hard to switch, especially for larger companies. But for smaller companies, it's much easier to switch. And so HubSpot is a threat on that front.

21:04And I think that, you know, given the stock decline, 75 % decline over the last 18 months or so, I think they are under pressure to show that they are able to leverage AI to create new features that will bring joy and delight to customers and make them want to buy more. And so I think that, you know, there's multiple dimensions, but definitely a lot of pressure involved in this decision, I think.

21:28Kevin McLaughlin:Okay, and what do you think the future prospects hold for the company? Do you think it ultimately just gets gobbled up by another CRM company? Do you think maybe it, you know, I could certainly see it trying to acquire AI companies, but it's not that big a company. You could see an AI company acquire HubSpot too, right? Yeah, I don't know. I'm just brainstorming. Like CRM, it's like, it's, I mean, you know, you've got giants, you've got the AI companies touting their own products. Like, what do you think the future of this business is? Well, there's no question that HubSpot is under pressure and under fire right now.

22:07The stock decline and this situation. And, you know, we spoke with a couple of customers who said we were already looking to get off HubSpot just because of the costs and the company's practice of sort of gating off features and requiring customers to buy higher level versions. of the software to get these features. And so, you know, the company is definitely struggling. I think that at some level, like if they are able to develop a hit AI feature, you know, that could maybe change the perception of investors. I wouldn't like rule out their ability to actually reverse the tide. Having said all that, this was not good.