Apple's AI Talent Crisis, Robotaxis, and Stargate’s Slow Progress | July 22, 2025

0:13Welcome everyone to the information's TI TV. My name is Akash Pashrich. It is July 22nd and we have got a busy busy show planned for you today. We are talking about Apple's AI talent struggles with some brand new reporting from our newsroom this morning. We've then got a special conversation with Midas List alumni Steve Jang from Kindred Ventures. We're also going to hear about the latest on OpenAI's new deal with Oracle announced this morning. And we're going to close out the show by talking to a founder who just raised$110 million. That is the CEO of Recca AI. Now, before we get started, a couple headlines for you this morning.

0:49Crypto company BitGo filed to go public confidentially with the SEC. It is yet another crypto company that is marching towards the IPO window. Anthropic has said it will look to raise money from the United Arab Emirates and from Qatar. That is according to Wired. And last but not least, Skydance Media is talking to Oracle about a major cloud deal that could be worth$100 million. That is according to Bloomberg. OK, those are the headlines. Let's get to our first guest. Apple's AI efforts have been quite the roller coaster over the past few years. The company has pushed out its long-awaited Siri update with all of its latest and greatest AI capabilities until 2026.

1:28It was initially due this year. And this morning, my colleagues at The Information reported that the company is also losing ground in the AI talent war. And I want to bring on Aaron Tilley, our Apple reporter, to talk about the story that he published. He joins us here from San Francisco. Aaron, thank you so much for being here. Thanks for having me. So tell us about this story. What did you find? Yeah, so the core team at Apple building their AI models, their models, their large AI models that they're trying to build something akin to a chat GPT, that whole group is sort of in crisis. their leader Ruming Pang has left and that's like a really important key figure in their whole effort because he was such a draw for other talent and he recently left for Meta and it's kind of created this huge earthquake inside that group and a lot of folks are potentially leaving and Apple executives are panicking and trying to keep them there.

2:36And why did he leave? There was, it was a big, it was a big pay bump. But more than that, more importantly, I think it was the mission. Apple is a very sort of, you know, they're doing building basic features for the products. They don't have this crazy sort of ambition for a super intelligence like Meta is pitching or AGI that OpenAI and others hope to accomplish. Apple is very grounded in product and sort of grinding away at specific features. They don't have this huge ambition to achieve super intelligence. So I think that is creating a situation where researchers, environment researchers don't really see very much interest for researchers.

3:33So I think that's a big factor. And, you know, another part of this is a lot of these AI researchers, the best of them, they want to be where the cutting edge is. And Apple is nowhere close to that. They're trying to get there. but there's just so much going on in AI. And I think these, these talents, there's, they want to keep at the cutting edge and that's where Apple is. Tell me a little, you talked about the earthquake that this caused inside Apple and, you know, it having been the subject of a lot of, of, um, you know, lack of better word, you know, uh, chaos developments, you know, Apple's AI group is kind of, um, for lack of better word, flailing right now, you could even say.

4:17I mean, tell me about how much weight Roaming Pang had inside the organization and why he was so important. Yeah, he was just a huge talent draw. Like, he was enormously respected by his team. He was really, I mean, he was very, we heard him describe very kind and generous with his time. And he just had so much skill building large, these large AI models that Apple needs to build alongside the rest of the industry. And so that was for his team, like they were so committed and devoted to him. Like it's without him there, like that presence there, it's kind of hard, like for people to justify staying.

5:00He kind of was seen as protecting them and defending them against sort of the bigger politics at the company. and without them there, they feel kind of defenseless. So do you get the sense that there's going to be more departures, more people leaving? Oh, yes. I mean, it's hard to tell at this stage. Apple has talked to the team. Apple executives have talked to the team about increasing compensation. So they are trying to make efforts to stop the stem, the tide from people departing. But they have, you know, already some folks have left. Mark Lee, Tom Gunter, who left before Rooming, but he's going to Meta with Rooming.

5:46And, you know, one of the things I wanted to ask you, you know, before we let you go here is Apple used to be, in many ways it still is, but, you know, it has always been seen as the company to work for in tech. And it strikes me that in this AI group, I mean, you know, other tech companies like Meta are sort of establishing this brand as the AI company to go work for. Do you think Apple can come back from this or what do you make of this sort of shift in brand? The company is always like, I mean, without a doubt, it's the best sort of like consumer technology company in the world. They build some of the most beloved sort of tech products.

6:24But when it comes to the age of AI, I mean, they have never been on top of it. It has been always a challenge for them to wrap their head around, to get a strategy that attracts really good talent in the space. So I think they've always struggled with AI. And this isn't the first time they've kind of hit some road bumps here. But yeah, I think in the case of AI, they've never been at the bleeding edge. Got it. Well, thank you so much for coming on the show. It was a really fascinating story. That is Aaron Tilley, our Apple reporter on the talent struggles inside Apple's AI group. I want to move on to our next guest.

7:08Steve Jang is the founder and managing partner of Kindred Ventures. He is invested in just about every sector you can think of. He was early Uber. He also backed Coinbase. He's an investor in Perplexity, Poshmark, Postmates. He is a Midas List alumni. Steve, welcome to the show, and thank you so much for being here on TITV. Thanks for having me. So, look, I don't really know where to start because, I mean, it seems like you've really invested in just about every sector. You've really taken a couple turns around the sun. I mean, let's start with Uber. The company partnered with Lucid, an autonomous driving company, Nuro.

7:45We saw that news last week. They're going to put 20 ,000 cars on the road in the next six years. I mean, I get the Lucid angle to this partnership. I hadn't ever heard of Nuro, and they're pretty key to it. So help us understand. I mean, why do you think they went with Nuro? Who is this company? Yeah, I mean, let's take a step back and look at the different companies that are out there today. So first, you have Waymo, right? And if you're in a Waymo-enabled city, it's amazing. And it's been such a game changer to watch them over the last 15 years, train their models, test their cars out. And then you see Tesla out there, and we all know that story.

8:24I mean, it's personally owned vehicles that are being transformed into potentially a robo-taxi network. And so it's two totally different approaches with Waymo and Tesla. One is a robo-taxi network from the start. And then the other one is actually a personally owned vehicle network from the start. And so you have two different origin points, no pun intended, for both of those companies. But then you have a third. And the third is an interesting network because it's not only a network of personally owned vehicles by all of those drivers, but you have the largest ride-sharing network globally. I think at this point, it's millions of drivers in cars.

9:08And so I think everyone's been sort of waiting for what's next for Uber. There's been various partnerships with Uber and Waymo and different autonomous vehicle technology companies around the world. And everyone's been waiting for the shoe to drop. And I think what's interesting about this is that this feels like a native Uber ride sharing, AV ride sharing network that's being put into place. So 20 ,000 vehicles is a huge start. And if you look at just North America, they can get started in many tier one cities. But tell us about Neuro. I mean, why did they go with this company? Sure. So, you know, Neuro was one of the AV companies focused on delivery historically.

9:57And so they had built their own vehicle that sort of looks like a small shuttle bus. And I think they took their technology, from what I understand about the company, they've taken their technology and they've transformed it into a Neuro driver, which basically integrates in from a hardware and software perspective into third-party vehicles. So in this case, it's Lucid, which is an EV company based in the Bay Area. Now, I think, you know, from that perspective, this three-way partnership makes sense. Lucid doesn't have their own AV technology, right? And so, but it is a great EV company. It's actually a very high-priced EV.

10:38I think the model that they're using here is an$80 ,000 at base price for personally owned vehicles. So what's interesting about this is this is the first time that Uber has a native AV network inside of their own network. So this feels like a first party product as opposed to a third party product that they're partnering with. And I think that's really exciting. I think the ride sharing network, obviously distribution, operations, customer service, orchestrating all of that maintenance of these cars, upgrades of technology, all of that are operational and logistics work that I think a lot of these companies, Waymo and Tesla included, have not had to deal with historically.

11:23And so this makes this now a three horse race between these three companies globally. You talked about the timeline here. I mean, they're putting 20 ,000 cars on the road in the next six years. Six years is a long time. We've seen a lot of progress, but still, I mean, six years is a while. So, you know, when are we even going to see a return on investment here or profitability as it relates to these vehicles? Yeah, I mean, this is challenging. So this is probably the first robotics area that will touch consumer experiences for a long time at scale. And so, you know, the hard part about autonomous vehicles and this robo taxi network, as we're calling it, is that the rollout takes time.

12:10You have to get it just right. There's really no margin for error, as we've seen in the past. you know, safety, local government, you know, public opinion, all of these things are super sensitive when it comes to AVs. So when you take self-driving cars and you roll them out as a business and these fleets, it takes time. And so they have to go city by city. They have to expand it out by radius. And really it's a slow expansion as opposed to in pure software technology. For For example, ChatGPT and Perplexity, you just roll it out worldwide from day one. And so it's just going to take a lot more time.

12:52The interesting thing is that the goal or the objective of this is to have such lock-in that the switching costs are incredibly high. And so I think that what you're going to find is that it will take longer. But these three companies seem like the three AV companies that will be operating these robo-taxi networks, these self-driving car networks for the long haul. I want to move on to Perplexity, which is also a company that you're investing in, that you are an investor in. But very quickly before we go, I mean, let's just stay on the robo-taxis for a second. How do you see the business model changing here?

13:26For a company like Uber, I mean, this is a totally different ballgame in terms of how the business works. How do you see them monetizing this? How does the business model change? Where do you go there? You know, I think this is about being a direct provider. You know, my guess is that this is going to be a higher margin product than some of the other partner products. one of the interesting parts about lucid is that this car is going to be able to not only be a personally owned vehicle out there but now with the with the partnership with neuro it'll be interesting to see whether or not this becomes available as a personally owned vehicle as well as part of the robo taxi network you know i think that there is always an opportunity to create sort of an UberX for AV.

14:18That's the Tesla model, right? The Tesla model is to contribute your car long-term to the Tesla network so that other people can receive rides from it using FSD. So I think there's a lot of opportunity there, but I think part of this is offering a direct service is to have control the experience, control the rollout, have a higher margin product as well for Uber riders. So, you know, perhaps this is just as simple as saying this is another Uber option. When you open up the Uber app today, you see Uber Black, you see Uber VIP Premium, Uber X Comfort, and you may see an Uber AV option. And I think that just logically makes a lot of sense.

14:58And it would also economically be a high margin product. Right. Let's move on to perplexity. So I want to talk about the browser wars here. You know, Perplexity, they have introduced their own browser Comet. I believe it's open to it. It's open by invitation only right now. Hasn't yet fully rolled out. We've heard a lot of reporting. We've reported here at the Information about OpenAI's browser. Why do you think Perplexity is going to win here as an investor? I mean, why is it that you think they will beat OpenAI here? Sure, sure. Actually, just to correct one thing. You know, it is available.

15:36The Perplexity Comet browser is available today if you're in the Perplexity Max subscription. And I think the company has indicated that they're going to roll that out to other subscriptions over time. So it is a phased rollout, but it is available to Max subscribers today. So, you know, look, there's sort of an argument about, you know, what can the underlying LLMs, the models, take on in terms of product or experience? And then what can application layer, so-called application layer companies take on? I think perplexity has proven that they're one of the best product teams in AI. They've been able to build a really captivating product for answers, daily search, and also higher order research knowledge.

16:25You know, a lot of people using it for work or for school. And I think that they've proven out that they've been able to really carve out a great lead in that product category, in that application layer. Now, when it comes to the browser, you might ask, you know, why a browser? Is there a need for that? And I think what we've seen is that, you know, the history of browsers has been, you know, this has been a game where speed, stability, little applications, so-called Chrome extensions have really changed the game for Chrome. So before that, it was Firefox, Safari, Explorer, and Chrome really came on the scene about 15 years ago and brought something new.

17:10And what they brought was speed, stability, extensions, also Chromium, right? The underlying open source framework for the browser. And that's really, you know, what Perplexity is doing is saying maybe it's not just about knowledge and answers. maybe it's also about being able to help people take actions through the browser. And I think that's what's really exciting here. You know, in the Perplexity Comet browser today, you know, what they're underlying it is LLMs, right? You can choose any LLM. You can choose Anthropic Cloud. You can choose OpenAI's models, various open source models like LLAMA.

17:53You can even use their own fine-tuned perplexity sonar model, right? But the most important thing about the browser is that it's giving you memory and a user interface that is quite different. And this is really important. It almost begs the question, is the browser the right name for this product? What would you call it? Well, the browser used to be about websites and websites were relatively static and they weren't dynamic apps when the word browser was coined. And now today, you know, the web is not only a bunch of websites, but it's also most of the applications that you use. Right. And so really, it's it's it's not just browsing apps.

18:37It's allowing you to use access and and use these apps. And so, you know, you use them for, you know, there's SaaS apps that you use for work that are very complex. There's also fun consumer websites and apps. And so the agentic actions that Comet has introduced are really interesting. They have the context of what app or what website you're on and has the context of what's available on that page. It also has the accumulating memory of what your preferences are over time. And then the third one is the user interface. So some people might say, hey, I can have my search open here. I can have a website open here and I can have perplexity or chat GPT or cloud open here and I can just move things around information.

19:29Isn't that good enough? But actually, it's not good enough because it's not as fluid and as integrated as possible. And so the actions are also made available because you can actually use the browser. And the Comet browser today is able to organize your data. It's able to change your spreadsheets, answer emails at scale. It's able to parse through your LinkedIn and figure out who you should respond to or accept. Right. I want to jump in here. I mean, before I let you go, I just want to go back to me. Why do you think perplexity wins against OpenAI in the browser wars? Well, I think what OpenAI recently released is an agentic product.

20:13It doesn't feel as much of a browser. It's not meeting the user to me. It's not meeting the user where they are. And I think the browser, the construct of the product is, let's go from a browser that is relatively static and doesn't allow actions. And then Perplexity's Comet browser is proposing, let's let you continue to do that. Let's let you continue to use the web as you were. But what we're going to add in there is amplify the knowledge you get from each web app or website that you're on and then take actions directly in that. And I think if you look at the Comet UI and experience, and then you look at the agent experience that OpenAI has rolled out, I think you'll see a difference of focus on product and design and experience.

20:59And then you'll see a difference in sort of testing out and flexing muscle on the LLM itself, right? In this agent framework. So I think they're both really interesting takes on it. I think neither take is unavailable to the other. Right. And so I think it will be definitely a product competition out on the market that'll be interesting to watch. But you can see sort of two different philosophies in the product and UX in both those companies. Right. Well, it's very fascinating. Like I said, I mean, you've touched just about every corner of tech. So we appreciate you coming on to talk about robotaxis and about perplexity and the browser wars.

21:38Next time when you come on, we'll have to get your take on generative AI media. I know that's something that you're very passionate about. Thank you so much, Steve, for being here. Thanks so much. Thanks for having me. Bye.

21:51Well, OpenAI made a big announcement this morning. It is expanding its data center deal with Oracle. and we published a quick update on it this morning. And the line that stood out to me from that update was that Oracle data centers for OpenAI would consume as much power as cloud provider like Microsoft uses today. There's also some new updates about Stargate. To break it all down, I want to bring on our cloud reporter who has been getting to know the data center space very well lately. Anissa Gardizi joins us from San Francisco. Thank you for being here, Anissa. Thanks, Akash. So what did you make of this new deal this morning?

22:26Yeah, I mean, it was quite a big update for Oracle and OpenAI to say that they've agreed to develop 4.5 gigawatts of capacity. Like you mentioned, that's almost as large as some hyperscale cloud providers today. So what stood out to me was honestly the scale. And then the other thing that I noticed was that OpenAI said that this was a part of Stargate and also sort of redefined what Stargate really is. So redefine Stargate how? I mean, tell us a little bit more about that. Sure. So earlier this year, people might remember Sam Altman from OpenAI, Larry Ellison from Oracle, and Masa from SoftBank throughout the White House, announcing Stargate as this joint venture that would raise money and then put that money into data center assets for OpenAI.

23:15and what we found out this morning with OpenAI's announcement was that over the past six months, they've sort of expanded the definition of what Stargate is and now any compute pretty much that OpenAI gets for itself, it's going to loop in under the Stargate brand even if SoftBank and this investment vehicle are not involved. That's not what it was to begin with, right? I mean, this is a pretty significant change. Right, and if you really want to go back to what Stargate was, it was in early 2024, this idea between Microsoft and OpenAI to build a giant supercomputer. So it's not too surprising that the definition of Stargate changed, but I think it's pretty important and it just shows that they realized maybe the joint venture route isn't the way to get some of the first data centers up and running.

24:03I mean, you said it's not surprising that it has changed and I mean, I agree with you there, but I mean, I wonder, you know, from what people have been telling you, does it surprise you that this is the new definition of Stargate, that all of OpenAI's capacity is included in it? Yeah, it's a really good question. And it actually does not surprise me because when Stargate was announced the second time at the White House, they pointed to a project in Texas that we had reported had no ties to SoftBank or Stargate. So almost from the jump, how they were using the word Stargate wasn't exactly describing data centers within the joint venture.

24:44What had been going on in Texas had been happening before the Stargate joint was announced. There was a report out of the Wall Street Journal this week that Stargate is off to a slow start. I mean, as it relates to that, what are your sources telling you? Yeah, everyone is sort of sharing that story and talking about what they think is really happening with Stargate. I think when Stargate was announced, they sort of, that team met with everyone in the industry and talked about their vision to build out 10 gigawatts of data center capacity in the US. And so, of course, in the past six months, we have not seen that kind of capacity built out.

25:24This is a long-term plan. But the journal did report that the Stargate joint venture is maybe looking at doing one data center this year. And I can kind of understand, given the fanfare earlier this year, how that seems a little underwhelming, perhaps. Right. Well, look, I mean, it's been a fascinating story to follow, and I think we are still very early in it. There's still a lot of work left to do. Thank you, Anissa, for coming on and explaining to us what that latest deal was and also how Stargate is changing. That is Anissa Gardizi in San Francisco. Now, one of the most exciting parts about this show is that when founders raise money, we can bring them on to talk about their latest funding rounds.

26:07And so we had news this morning that RECA AI is raising$110 million in a funding round that included Snowflake and NVIDIA. The company's valuation is over a billion dollars now. We don't have exact figures of where it sits, but the company is building multimodal AI models and its CEO, Danny Yogatama, joins us. Danny, thank you so much for being here. We appreciate it. Happy to be here. Thanks for the invite. So look, the way I understand it, you know, the pitch of your company is that your multimodal multimodal models, I should say, are supposed to be more efficient than other AI labs. What is it that makes your technology more efficient?

26:46Yeah. So we are an AI research and product company. We build ultra efficient multimodal language models. We use them to power our end-to-end solutions, Rekka Research and Rekka Vision. We train everything that we have from scratch. It's optimizing on certain use cases that we care about that allows us to deploy efficient models for our customer base. So for example, Rekka Research is our research agent that browses the web, private documents to answer questions. We use our own model, Rykel Flash, to orchestrate the plan. This is something that a couple of front-end models can do very well as well, but they are expensive and a bit slower due to their sizes.

27:31Using our in-house model, optimizing it for these tasks allows us to achieve present cost performance at a much more attractive cost profile. But do you not think, I mean, you know, the bigger tech companies who are, you know, multimodal models are kind of all the rave today. I mean, the bigger tech companies are surely doing work to get them cheaper. I mean, do you not think they're going to get there? What gives you confidence you can take on them?

27:58We've shown so far that we're able to deliver, you know, best-in-class performance for our customers at a fraction of the cost. Our models cost us maybe a few million dollars to train each. We've been focusing on Rika Flash more recently for Rika Vision and Rika Research. They are also, I think, trying to bring down the cost for their agile purpose models. But as I said, we're only focusing on two very specific use cases. RegaVision is for analyzing, searching, and editing video content at scale. This is due to our powerful multi-modal understanding capabilities. So on these two specific things that we focus on, we're not worried.