How AI “Thinks”, Amazon vs Trade Desk AdTech, xAI's Talent Pool, and Robotics | Aug 15, 2025

0:12Welcome everyone to the information's TITV. My name is Akash Basricha. It is Friday, August 15th. We have got some fun guests for you today. We are talking about robotics and AI with the CEO of one of the hottest companies in that space. We're also going to talk about the wonderful world of AI inference providers and hopefully get to the bottom of what that term actually means. We've got a scoop about XAI that I am excited to get to with our Elon Musk reporter. But we're going to start with a market moving scoop from the information that we published just yesterday. Now, one of the most exciting rivalries in tech right now is how many companies are competing in the ad tech space.

0:52That's a category that Google and Amazon are playing in. And it's also where lesser known companies like the Trade Desk play. Now, you might never have heard of the Trade Desk or paid attention to them, but it is a$24 billion public company. And the information had some news yesterday about its relationship with Walmart that sent shares down in trading. Joining me now to discuss that is Kathy Perloff and Theo Waite, two reporters of The Information, who wrote that story. And we also want to bring on Shweta Kajuria, who follows the company at Wolf Research. And it's great to have you all here. Kathy, I want to start with you.

1:26Look, most people probably have never heard of the Trade Desk. And so talk to us about why the company is important and why we should actually care about it. Yeah, the Trade Desk. Yeah, maybe not a household name, but really well known in the ad world. It is one of advertisers' favorite place to buy ads across the internet. So the company automates the buying of ads that run on websites, on mobile apps, and particularly on streaming television. So they, and they're really easy to use intuitive platform that has sort of become popular as a challenger to both Google and Amazon, which are also in this business.

2:08And because of their popularity, they've been able to kind of carve out relationships with, you know, premier media partners, including Walmart, which we can, you know, get more to. But they've used their sort of good technology to sort of leverage it into having partnerships with media that advertisers want to be on. Got it. So you've got a company, you know, a beverage company, let's say they want to buy ads on the internet. They use Trade Desk to help them find the different places where they can buy ads, what the right place is to place those ads. What did you find about the company's relationship with Walmart?

2:42So the Trade Desk and Walmart have had this exclusive partnership since 2021. And that means that if advertisers want to use Walmart's shopper data to inform their ad buys across the Internet. So if they want to target, you know, Walmart shoppers who are in the market for shampoo on, you know, Bustle.com or on Hulu, they can use the Trade Desk. And the Trade Desk is the only place they can do it. But what we found out is that the exclusivity of that relationship was Walmart renegotiated the deal so it would no longer be exclusive last year. So that means that Walmart can work with other ad tech companies to facilitate this buying.

3:30Also, you know, we've heard that Walmart might want to buy a company. Maybe they'd build, you know, their own company. And one of the reasons that precipitated this change was there was, Walmart was somewhat unhappy with the high fees that the trade desk charges. Right. So, I mean, to be clear, they still work together. You know, the companies have said they're still committed to the partnership, but, you know, there's some sort of friction there. Right. And so, Shaita, I want to get to you to talk a little bit more about how the market responded to this. But, Kathy, just a few really simple questions here just to make sure because I'm sure people are wondering this.

4:07So here we're talking about Walmart as an ad buyer here that is using Trade Desk to buy the ads? No, we're talking about a good question because it is very confusing. But is Walmart as an ad seller? So Walmart has a business where because so many people like are familiar with the Walmart brand and they have such a big audience, they can sort of sell that audience to advertisers who want to target it either on Walmart's website or across the Internet. And so Walmart is sort of selling. It's weird, but in the advertising ecosystem, if you have people to look at ads, you're a media company, which is how, yeah.

4:46So Walmart is selling ads and Walmart had a relationship with Trade Desk that represented the buyers basically coming in. And that relationship is no longer exclusive. So now Walmart can work with other companies that represent the buyers. And so that is obviously a hit to Trade Desk's business. Shweta, I want to come to you. What was your reaction when you read the story? Reaction to the stock move or to the story? Well, the story and the stock move shortly after that. I think Kathy did a great job. I think that in terms of explaining the relationship between the trade desk and Walmart and what it really could mean for other players in the market.

5:29In terms of the stock reaction, however, I think that maybe it was a little bit more exaggerated. Here's the reason why. First of all, it is a negative for the trade desk to lose exclusivity. That just brings in more competition. However, the way the stock reacted implied as if that the trade desk was going to lose Walmart partnership altogether imminently. That may not happen. First of all, Walmart is not as meaningful of a revenue contributor to the trade desk. We think it's a very modest impact, even if all of Walmart is still there. It's a very modest impact for the 2026 estimates. Second is that it's probably going to be a slower bleed instead of all of Walmart just transitioning right away.

6:17Perhaps they will start increasing competition between Yahoo or Criterion. I know Kathy mentioned a couple of those. So anyway, that's my immediate reaction. I do want to take a bit of a step back here. I mean, just last week, we saw an even bigger price movement for shares of TradeDesk. They dropped something like 35%, 40 % almost in a single day. Talk about what that was about, and could this have really just been more fear? Exactly. That was where I was going next. So you basically read my mind. The stock reaction post the print was the lack, it was a reflection of perhaps investors' way of saying that there are two issues here.

7:04The management team's credibility is now being questioned after so many years of consistently producing results. And now this being the second quarter out of three quarters where perhaps the results were not as expected. And that's very unusual for a company that has been outperforming consistently for several years since their IPO. That's point one. And the second is that it was also a reflection that even if we don't know really what is the issue that the trade desk may be facing, whether it is weakness in brand spend or competitive dynamics, the reality is there is some issue and that creates a structural headwind and puts a cap to the multiple for the stock.

7:50Theo, how does Amazon fit into this ad tech world? Amazon has their own DSP that competes with the Trade Desk and with the other companies we've talked about. And they've been trying to steal a lot of business from the Trade Desk, as we've written over the past few months. I wouldn't expect Walmart to ever use Amazon because they have obviously a massive rivalry and Walmart goes to great lengths to make sure it never touches any part of AWS, for example. But, you know, that being said, they are putting pricing pressure on the whole industry. And, you know, I think there are a lot of people at the trade desk that wish they could offer pricing like Amazon is right now.

8:29And I want to unpack that a little bit. And, you know, again, just to translate. So Amazon has an ad tech product that competes with the ad tech product that the trade desk has that helps buyers buy ads on the Internet. You know, one of the things you just mentioned was pricing pressure. talk a little bit about what Amazon has been doing with its own ad tech product to compete with Trade Desk. So Amazon has this huge ad business, the vast majority of which is still its e-commerce site, right? You pay to boost the visibility of your items when people search on amazon.com. And that's still the vast majority of their business.

9:03But they've been expanding into Prime Video. They've launched, you know, and put more emphasis on this DSP that does compete with the trade desk. But, you know, ultimately it is a tiny side quest for a very profitable company. And if, you know, people in Amazon's S team decide we want to grow the, you know, our DSP, we want to, you know, basically shut down the trade desk or, you know, compete with them, you know, more strongly than we have been, like, that's really not a big deal for them to do, because they just have so many resources. And, you know, they're probably not going to be able to poach Walmart, like I said, but they're, you know, definitely putting pressure on them.

9:40Right. Very quickly before we go, Kathy and Trit, I want to ask both of you, you know, we have these two Google antitrust rulings. One was to do with Google's ad tech business. Is that an opportunity for the trade just right now, Kathy? Yeah, it's a huge opportunity. I think that Google, you know, has been the dominant company in advertising technology, which is why the Department of Justice brought a case against them. But since they've been under scrutiny in a lot of the key ways the digital advertising market is growing, like streaming television and e-commerce advertising, they've not been as aggressive.

10:17And that's allowed companies like the Trade Desk to kind of get a footing in the market. And if, you know, the Department of Justice wants to break this business off of Google and, you know, that could help the Trade Desk even more and Amazon even more. On the other hand, because it's been part of such a, because this unit has been part of Google, some might say that that's been like halting its ability to innovate. And maybe if it was a standalone company, you know, with the kind of regors as they could be as strong as they've been in the past couple of years. So, yeah. Shreita, is this opportunity for Trade Desk, is that priced into the stock right now or can we expect more movement there?

10:58No, I don't think it is priced in. And I don't think that investors may be expecting a meaningful impact from the ruling post the ad tech case. My view also is that it may impact some of the supply side platforms like a Magnite or a Pubmatic perhaps meaningfully more than it may on the demand side, which is where Trade Desk fits. But don't get me wrong. I do agree with Cathy. It should be a positive for the Trade Desk. Great. Well, thank you to the three of you for coming on. It is a very, it's an interesting world that has a lot of terms and complicated concepts and, you know, value chains attached to it.

11:37So we appreciate the three of you coming on to discuss all that. That is Kathy and Theo from The Information and Shweta from Wolf Research. Okay. XAI is one of the closest watch AI companies right now. And Talent is one of the closest watch priorities for every company in this space. And this morning, Theo, who was just on the show, was coming back. He gave us an inside look about exactly where XAI's talent is coming from. And I want to bring Theo back to talk about that story. Theo, good to see you again. Good to see you. So look, it feels like talent at XAI, we know less about that compared to other companies like OpenAI and Google.

12:17We hear a lot about the people that those companies are going after. What do we actually know about who worked at XAI? So XAI, you know, was, was, it's less than three years old. It was formed, you know, with Elon making these, you know, grand statements about open AI and, and Google, you know, being evil and needing a competitor that was different and controlled by him. They hired a bunch of, you know, high profile researchers initially. And then, you know, it's, it's been a little quiet and we've seen all these headlines recently, like you said, about, you know, meta throwing hundreds of millions of dollars at people.

12:50So we decided to do a story that was basically just, okay, who actually works at XAI now? And what we found is, you know, there are quite a few people from Google. That's the biggest source of talent. There are about 19 that have joined this year at least, and 40 since the creation of XAI. There are also a lot from Meta, a lot from ByteDance and Microsoft. And it kind of just underscores the extent to which there are basically a handful of important companies in this space and talent is kind of always circulating between them. And what is it like to work at XAI? I mean, is this comparable to the other AI companies?

13:32You know, I think Elon has this reputation as someone who, you know, drives his people really hard. And that's definitely the case at XAI, especially because it's been where he's kind of focusing his attention largely in the past, you know, couple of years, outside of politics, that is. And, you know, we understand that there are some pretty, you know, incredible demands that are made on people there. I understand that, you know, some teams can work seven-day weeks pretty frequently. And, you know, that can be demoralizing for some people, but others are, you know, very invested in the mission.

14:10And, you know, if it's 2025 and you sign up to work for an Elon Musk company, you do kind of know what you're getting into. Well, and one of the things, my perception at least of XAI is that it's actually a much smaller company, I think, from what I understand than some of the other companies, and yet their models are pretty good, right? Yeah, and a lot of that comes from the tremendous amount of resources that XAI has poured into building their initial data center, Colossus and Colossus 2, which is about to go online later this month, potentially. and you know just having this this incredible amount of fundraising and money that you can throw at compute does does go a long way now one of the most interesting details about your story was this idea that you know elon likes to talk about people you're talking about the research here and he basically you know the company likes to call the technical talent just members of the technical staff is this common in the industry or is this something that is unique to xai You know, I think it's a phrase that does exist in other places, but I think is more emphasized at XAI.

15:19You know, Musk has kind of attacked the title of researcher as this, you know, extravagant academic idea. And, you know, if you're really trying to build serious products that people will use on tight deadlines, you're an engineer, you're not a researcher. Researchers, you know, just sit in their offices and, you know, think extravagant thoughts or whatever. Right. And last question for you, Theo, we've heard about these big pay packages that are being put out in front of researchers to try to get them to come. Have you heard anything like any grand sort of gestures in terms of pay packages that XAI has made at all or to attract researchers or anything like that?

15:59I mean, the people that are there are definitely, you know, making decent money. I haven't heard anything on the scale of what Meta has been, you know, offering people. But, you know, what I would say also is that it's kind of notable that there really haven't been that many, like, super high profile hires by XAI. It's more been like, you know, assembling this large group of people that, you know, were not like superstars at Google or Microsoft, but very smart and, you know, kind of are attracted to the hardcore culture and have been able to catch up with those companies despite being less famous largely.

16:37Right. Well, great. Theo, I know there's a lot left to uncover. And so we look forward to having you back on the show. That is Theo Waite, who covers all things Elon Musk related at the information. Okay, I have been dying to cover robotics on the show, and I am so excited to bring on our next guest. Jake Luserarian is the co-founder of Gecko Robotics. He has been building in this space for 12 years now. His robots are used across several different sectors, including oil and gas, defense, pulp and paper, and mining. And I want to bring him on to talk about where this sector is headed. Jake, welcome to TITV.

17:13It's great to have you. Yeah, great to be here. Thanks so much. So, look, I've seen the videos of the robots crawling on Navy ships and these little mini drones flying around. I mean, explain to us what your robots actually do. Great. Well, yeah, a little bit about the company. What the robots do is they inform operational data platforms that we sell to those different sectors and industries. And so what they're doing is they're going out and gathering all this information and data about the health and also the performance of all those assets that we're climbing all over. So the robots specifically that you probably see are gathering corrosion, erosion, cracking data on these physical structures.

17:54So steel, concrete, composites, and trying to assess where all the damages are, what's going to fail. And then also how, with that information, how do you help the system actually work way better with this new information and data set to how our AI models to be the best in the world. And so the way that I understand is, you know, you have physical infrastructure like a bridge, right? And, you know, the government or, you know, whoever is operating the bridge will hire Gecko and its robots to come. Robot will sort of crawl on the bridge. And then, you know, the robot sort of sends the data back to the software called Cantilever.

18:32And then the software says, hey, the bridge has fault lines here, here, and here. You've got to fix it right now. Yeah, that's exactly right. And I'd just like to put this in perspective for folks. I mean, we got our start in the energy sector. I started the company out of a college dorm room in 2013, bootstrapped it, and then was in YC for the 2016 batch after turning down an acquisition offer. So, you know, truly a hundred bucks to the name kind of story. And, you know, what I learned was that one, most technology companies do not like step foot into these sectors and industries. They don't understand the problems at their core.

19:08And they don't understand that there's data sets that just simply have never existed before that you need to be able to help these sectors be able to digitize and be able to enter a world where software and AI can truly be impactful in unlocking these kinds of big game-changing solutions and decisions. But you're right. You're gathering information data with the robots. They're feeding it into the software so that they can make, the software can help to supercharge the kinds of decisions that wouldn't have otherwise been impossible to make. And it's not just on the health, but it's also looking from a system-to-system standpoint.

19:44How do you actually make the system perform better? So it's a combination of a bunch of different hardware solutions like that that are gathering data that never existed before. So I want to talk broadly about robotics, but I have a couple of rapid-fire questions here about the business. It's a very fascinating business. and I just want to make sure I understand it. So first of all, you know, you're working with mining companies, oil and gas companies, you know, all these industrial equipment companies. Is each company getting their own bespoke robot or how many different types of robots do you have?

20:15Great question. So we have a core architecture on the hardware side and that allows for us to be able to adjust and augment the robots to be able to meet whatever the conditions are. We have some robots that can go underwater. We have some robots that can climb up walls or go in pipes. We have some ones that are just, we just use a common architecture as well to make fixed sensors. And so you want like a modular platform. But to answer your question, we actually don't believe in this concept of building robots and selling them to these different companies. We believe in a model where you build the technology stack and the advantage to build unfair software and operational platforms comes from that data set.

20:54Why would I give that away? And why would I believe that a software company would be able to understand the complexities of the data sets and the new kinds of sensors or robots and payloads that need to be built at these facilities that you just don't have any other way of identifying understanding. So you're not actually selling the robots to the companies. You sell them like, you know, one to two year contract or something like that with the software and you say, we'll give you the robot to use for that time. Yeah, that's right. We compete on a software to software competition. So what we're looking at is we will come in with one of those companies that you're referring to, they'll buy software licenses, they'll buy cantilever credits, which is a combination of different kinds of sensors and robots on total amounts of equipment that we're looking at.

21:38And all of that is able to, you know, fuel the ability to make sorts of decisions around how to extend the useful life of the infrastructure that was supposed to only last a year or two longer, or how to get better heat rates at power plants to make them, you know, make more kilowatts for less BTUs. There's just like a comp there's so much that you can actually unlock that was never possible before because of these extra information and data layers that never existed and i'm here you know one of the things i'm here to say too is like look we're talking about the who's going to win the ai race and energy so associated with that what's very clear to me and then and what uh you know people are going to start to talk about way more is you know those that can win the robotics race and and the data race are going to be the ones they're going to be right you know most advantaged in that in that pursuit.

22:23And also that can help unlock more energy creation for less investment, less assets. We're seeing that with our customers now, whether it's on the energy side, but also same kind of concepts in mining and on the Navy side too, internationally and in the US. Right, right. A couple more quick questions for you. So, I mean, you make these robots, this is kind of like a neat business model. This is like robots as a service, kind of the way I'm thinking about it. I mean, how much does it cost to make one of these robots? Yeah, the cost could be like millions of dollars. It doesn't matter the cost, I guess.

22:59Like we're not trying to optimize for the bomb. We're trying to optimize for the impact of the information we're collecting. And the difference between the robots as a service and what we kind of do is like, it's kind of unfair, right? And that's kind of the whole thing. Like I want to, you know, the software we're building is like no one else can actually deliver that because we're the ones to have all the robotics stack and the technology that supports us offering a solution that no one else can bring to bear. So you actually don't think of it as robotics as a, as a, as a, as a, you don't think of it as like that.

23:27We sell software solutions. So, you know, we just happen to use tools that other people don't have. Got it. Got it. Well, I mean, I think that's interesting because, you know, we say robotics companies and we sort of, we think about the robots. And I guess what you're saying is, yeah, we built the robot, but the robot can't do a whole lot unless you have a software platform that actually can make sense of all the data. Yeah, and gosh, the reason I like to keep the name robotics in the name of the company is simply because the position is great. It's like, yeah, we come in very differentiated because we can offer a software.

24:02And also, I want to redefine it too. While people think about robotics in general, whenever you speak on these panels, I'm always talking about how, look, I would love to buy as many humanoid robots as Elon can build. And I'm going to work in these power plants that I run and operate with the software. And if they can do great, incredible things and get different information and data sets that I couldn't have gotten before and make jobs safer and help to enable humans to make smarter decisions and make these, you know, truly help the U.S. win a race on energy or manufacturing or whatever it is, you know, I want to put these robots to work.

24:35So I'll build a bunch of robots, but I'm also not going to reject Optus when it becomes a tool that's reliable. Do you think you're going to get into humanoid robots and building that at all? If folks aren't delivering products that are good enough for me to use, then I definitely will, and maybe I am now. As in, maybe you are working on humanoid robots as we speak? Mobile robots are an important part of the future, for sure. okay well i'm going to take that as a gecko robotics is looking seriously at humanoid robots unless you tell me otherwise that's very that's that's very that's very exciting uh jake thank you for coming on the show i really appreciate it and you know next time you come on we'll have to get a demo of one of these of one of these robots and i'm sure you guys have lots of new features coming out so please come back on the show and show us very soon again that is jake lucerarian from Gecko Robotics.

25:31Okay, for our final guest today, inference has been a term that you have probably heard about, loads about in AI, and you might also have heard of this term, inference provider. Admittedly, this is a type of startup that I have had a little bit of trouble understanding myself, even though the information has broken a ton of news on them. And so today, we're going to get to the bottom of what an inference provider really is. Joining me now is Lin Cao, the CEO of Fireworks AI. They are one of these inference providers. Lin, welcome to TI TV. It's great to have you. LIN-HUN CHOI - Hey, thanks for having me.

26:05So let's talk inference here. I mean, what is the difference between an inference provider and then a cloud provider like AWS and Azure? It feels like they're the same thing to me. LIN-HUN CHOI - There's many layers of model serving, all the way from a fundamental infrastructure, from data center power and the stack and racks of deploying compute the gpus the cpus the storage and and so on interconnect uh install that into the center and light it up and then there is a lower level software stack to power those there's a higher level um you know cloud level infrastructure like Kubernetes, part of those.