Apple Explores Ways to Welcome AI Agents in App Store, Cerebras IPO Winners, Modals’ Raise

0:13Aaron Tilley:Welcome, everyone, to The Information's TITV. My name is Akash Pasricha. It is Thursday, May 14th. First up today, The Information published exclusive reporting that Apple is exploring ways to better incorporate AI agents into its App Store. Our Apple reporter will join us shortly to share with us what he learned. Plus, we have more exclusive reporting around the Cerebrus IPO. The company is making its debut on the NASDAQ today. Our VC reporter is going to join us to discuss who the big winners are in the IPO. We'll then get to our daily update from RocketDrew around the Elon Musk OpenAI trial.

0:49Aaron Tilley:We will also discuss startup Modal's big funding discussions and wrap with a conversation with Gokul Rajaram. a well-known name in Silicon Valley. It's going to be a fun show, so let's get right on into it. Apple is trying to figure out its approach to agentic AI on its App Store. My colleague Aaron Tilly published exclusive reporting on that topic this week. I want to bring him on to talk all about it. Aaron, welcome back to the show. How is Apple dealing with the agentic AI platform and technology on its App Store? Yeah, so it has a team working on ways of incorporating AI agents better into its app store.

1:33It's going to be trying to be a much more safer and secure way of deploying agents. These are things that can take action on behalf, take instruction and orchestrate sort of actions on its platform. So that's doing things like planning a trip and scheduling hotels and flights. And yeah, it's certainly one of the, you know, the currently awesome thing going on in AI. And they've really taken off on the Mac platform. So OpenClaw, a really popular open source tool for agentic AI, has really taken off on the Mac platform, the hardware. But Apple's looking for ways to open this up on its platform, a very closed platform, the iPhone.

2:24Aaron Tilley:So a couple weeks ago, you came on the show and you told us about your reporting that a number of apps, iPhone apps, I should say, on the App Store, they were having trouble getting approved because they offered vibe coding tools in their functionality and this was a discussion that you and i had about basically apple was a little concerned about you know how much power these apps could have you know potentially creating new apps that could then make the app store obsolete altogether uh you know that was one concern that came up is that still the case or you know have we come around has apple come around to approving these these uh apps yeah so vibe these vibe coding apps there's certainly um you know a type of agentic ai you ask uh these vibe coding apps to do something build something it'll go off and do it um the problem for apple is this is all a little too freewheeling freewheeling You know, these vibe coding apps could build anything you wanted, stuff that Apple would consider improper for its platform or dangerous for its users.

3:39So it's a little bit too open for them and they want more control over it. So I think it's a balancing act where they're going to have to kind of open the door a bit to allow these some sort of agentic AI to be functioning on its platform. We're also not allowing a user to do anything they want with a vibe coding app, which is, you know, they could, again, introduce some sort of security risk or improper material that they just don't want running on the iPhone.

4:14Aaron Tilley:So, and I just want to make clear, we were talking about Vibe Coding apps. We're talking about agentic AI tools in your story out this week. Are we talking about the same group of apps right now or are these different groups insofar as how Apple thinks about them? um i mean i'm not sure exactly how apple is thinking about these vibe coding apps at this very moment but i think it's a subset of sort of like ai agents like ai coding has been a really key explosion for the current ai uh ai companies and it's again a form of agentic ai but i don't think it's the only kind of thing we're talking about here right right and the point you make in your story is that one of the things that Apple gets nervous about is the idea that if I have an app that can then assert control over other parts of the operating system, you know, or other apps even to do the task that, you know, it prescribes to do, that is one of the things that is concerning for Apple just based on security guidelines and how they've traditionally thought about it right for sure i mean i i think apple's always their sort of like line they go back to is our platform is more private and secure than any other sort of consumer hardware platform on the planet and that's why we deserve a massive cut of the transactions that take place on this platform right now let's let's talk about the flip side to this argument which is that then And what is the, you know, the risk if Apple does not go with where everyone else is going, certainly developers, right?

6:03Aaron Tilley:If they say that we're not going to allow these tools and functionalities on our platforms, then the risk is that they will stop becoming sort of the developer ecosystem of choice, right? That's the concern? I think that's 100 % right. I mean, you know, they may never end up kind of opening the door for a genetic AI. I'm just talking about like there is efforts internally to do this, move in this direction, but maybe it doesn't gain traction. Maybe they think it's too insecure for them to move forward with. And that is definitely top of mind like risk for them is do they become irrelevant in this new generation of of software, of AI?

6:42So I think they're really thinking about this as like, Apple makes the best hardware and they need to be the best platform for this new generation of AI, even if it's not something they do internally in-house. How do they make this accessible for developers outside the company?

7:00Aaron Tilley:Last question for you. Let's talk about Siri. So we are expecting the revamped Siri. What implications do these discussions around how to handle AI agents on the Apple operating system. What implications does that have for Siri and inevitably how good it could be? Because you and I have talked. I mean, this is the thing to watch, right? Is Siri going to be good or not? Yeah, Siri is like the thing that Apple has to do well on at this upcoming developers conference next month. I mean, they've been promising for the past two years a more personalized, agentic type of chatbot that they've been promising since it launched in 2011.

7:49But this is their first party experience that they need to nail. um but yeah as far as how it affects this other conversation we're having like this is the this is opening the door for third-party agents to be more um you know functional on the ios platform so but they need to nail this first-party experience they've been promising for the past two years to work to launch something more more meaningful and um there's a lot of pressure but I think they also know that they can't do all on their own. Like, this is a company that is not making the kind of investments it needs to make to do something world-class AI.

8:31They're working with Google Gemini now to kind of bring up its capabilities, but they're going to have to work with outsiders more. Great.

8:40Aaron Tilley:Well, Aaron, I want to thank you for coming on. That is Aaron Tilly, our Apple reporter, here at The Information. now cerebris is going public today pricing its ipo at 185 dollars a share well above the initial expectation of between 115 and 125 bucks a share as is tradition with big ipos the information has exclusive reporting on who the big winners are in this ipo this has been a staple of our coverage for a long time and so for the cerebris edition i want to bring on julia hornstein who was one of the reporters on that story this week. Julia, welcome to the show. It's great to have you here.

9:18Thanks for having me.

9:19Aaron Tilley:So let's talk about the, well, actually, before we get into the names, and we're going to talk about the specific winners in a second here. Broadly speaking, you went out, you did all this reporting, you talked to some VCs. What was the, you know, what were the broader observations you got from who stands to gain in this IPO? I think that one of my biggest takeaways was just the excitement around AI infrastructure and inference. I mean, we can see, as you just mentioned, the IPO priced at$185 a share, though it's indicating that it may trade up to$350 today. So I think that there's just broad investor excitement for AI inference, not only on the VC side, but also among retail investors as well.

10:07Aaron Tilley:Okay, so let's talk about the big winners then. Who did you find in your reporting stands to gain the most here? I think one of the first ones that stood out is Benchmark. You know, Cerebrus wasn't a typical investment for Benchmark, a firm known for its early bets in Uber and Snap. But the firm was intrigued by Cerebrus' moonshot idea of running AI faster and better and redesigning chips. Benchmark first invested in Cerebris in 2016. It co-led the chipmaker Series A with Foundation Capital. Its stake at the IPO is nearly 10 % after it purchased an additional three or so million shares last year and this year.

10:48That makes it more valuable than Benchmark's stake in Snap at the time of the social media app's 2017 IPO. We expect the firm to 12x its$268 million investment, turning it into roughly$3.3 billion. Benchmark also quietly puts some money into later rounds and special purpose vehicles, which we learned through our reporting, meaning that its returns could be higher than what we expect based off security filings. It could also be higher if the IPO pops today and trades at 350 plus a share. Another firm that stood out to us was Foundation Capital. We expect the firm to have a staggering 76 % or 76X return on a$37 million investment.

11:33And we estimate that their stake at the time of the IPO at that 185 share price will be worth$2.8 billion. They've also been in the company since the Series A. They even let Cerebris employees actually work out of their office back when they first invested. They backed the 2018 fundraise as well. So those were the two firms that stood out the most to us.

11:56Aaron Tilley:Now, what about the founder, Andrew Feldman? He's a name that we've talked about on the show. How much does he stand to gain? So Andrew Feldman owns a 5.5 % stake in the company. And at that 185 share price, his personal stake would be worth roughly$1.9 billion dollars. He's also not, he's one of a few, you know, people on the cap table. Some other ones that stuck out to us are the, some OpenAI co-founders, Sam Altman and Greg Brockman. They each own tens of thousands of shares, which we now know from court filings from the Musk-Altman trial. We know that Cerebrus kind of struck this interesting deal with OpenAI at the end of last year.

12:40Those two are sitting on stakes worth 16 million and 14 million apiece. So the founder stakes are much larger, but still there's some other people on the cap table as well.

12:50Aaron Tilley:I wonder, as you think about reporting questions, what comes next in this story? Obviously, there is the question of the share price and how that will oscillate or how it will fare after the IPO. There's also the question around cerebris and whether or not the technology potentially could get bought by another big chip firm. But for the venture investors, I'm just thinking, what do you think comes next for them? What's the next big opportunity? This was chips. These are firms that probably invest in AI applications too. What's sort of hot on their mind right now? I think there's so much that's hot on their mind right now.

13:38I mean, chief among them is still inference. I think that investors are really excited not only about inference, but also AI infrastructure. I mean, there's so much talk about compute and data centers and what these smaller neolabs will, and even big players like Anthropic and OpenAI, will need in terms of compute. So I think that's really hot on investors' minds as well. I think that we'll see some more talk of consolidation in this market, as you were kind of alluding to. I mean, we can point to Grok late last year. So I think that we'll see some more interesting themes play out kind of around the inference era.

14:18Aaron Tilley:Great. Well, Julia, I want to thank you for coming on. That is Julia Hornstein, our venture capital reporter here at The Information. The information that Rocket drew is covering the Elon Musk OpenAI trial from the courtroom all week long. Here is Rocket's latest update. All of the evidence is now before the jury in Elon Musk's lawsuit against OpenAI. Both sides were running low on time by the end, so OpenAI's lawyers rushed through questioning their last few witnesses before resting their case. Those witnesses included an accountant, a business professor, and a legal expert, a professor who said that in his view, OpenAI has acted in ways that are consistent with the best practices and customs for non-profit governance.

15:02Maybe the most interesting witness of the day was Josh Achiam, who is OpenAI's chief futurist. That's his actual title. He was previously head of OpenAI's mission alignment team. And he has an interesting perspective because he's been at OpenAI since 2017. So he was even at OpenAI while Musk was still there. And in fact, he told a story, a story that has come out before, that on Musk's last day, Sam Altman hosted him for a Q &A session at an all-hands. So basically, most of OpenAI's employees attended, 50 or 60 people. Sam Altman asked some questions, and then employees had the opportunity to ask questions.

15:39And Josh asked Elon at the time, why is it that you care so much about seemingly recklessly catching up to Google in AI so bad when you think that AI could also be a risky thing to pursue? Aren't those things kind of inconsistent? Doesn't it create risks to try to race so quickly with Google? And Elon was not sympathetic to the question and called Josh a jackass. So that resonated with some people at the company, and a month later at the next All Hands, some other employees, one of them was Dario Amadei, who went on to found Anthropic, presented Josh with a trophy, an actual trophy of the lower half of a donkey that said, never stop being a jackass for safety.

16:27And this trophy was present in the courtroom because OpenAI's lawyers held it up to the judge in the morning and sought to admit it into evidence. Ultimately, they didn't show the trophy to the jury, but they discussed this anecdote and they discussed the trophy and a photo of it was admitted, which was something that did not please Elon Musk's lawyers, as you can imagine. So it ended up being an interesting day, if for that reason alone. Up next in Elon Musk's trial, both sides will deliver their closing arguments, which is something that will really benefit the jury. Because the jury has just been flooded with all sorts of old email threads and mountains of evidence from across the history of OpenAI.

17:11And now the lawyers in their closing statements have the opportunity to take these pieces of the puzzle and assemble them in such a way that they depict a clear picture to the jury that the jury can then follow in their deliberations, which could begin as early as Thursday and could even begin on Monday as well. So that's what's up next. And then on Monday, the judge will begin to hear a short remedies period of the trial at which other experts, damages experts, economists will testify about what sort of remedies would be fair in a trial like this if the jury does end up finding that Sam Altman, Greg Brockman, OpenAI, or Microsoft are liable.

17:56Aaron Tilley:another company that rents gpus out is looking to raise money at a steep valuation my colleagues stephanie palazzolo and anissa gardizi have reporting on modal and i want to bring on stephanie to tell us more about what she's learned about the startup stephanie welcome back to the show what did we learn about modal thanks akash yeah so first is that modal on the back of pretty strong revenue growth is in talks to raise a new round of funding. And so this round would value it at around$4.5 billion. Excel and Redpoint are in talks to invest. And of course, all this comes on their revenue increasing by more than fivefold since last fall.

18:38So last fall, it was at around $60 million in annualized revenue, and now it's at around$300 million. And so that's obviously pretty impressive growth there.

18:48Aaron Tilley:And what exactly does this$4.5 billion,$300 million annual recurring revenue, whatever, big numbers, what does it do? Yeah, so Model does a couple of things. I think chiefly what it does is it, you know, rents out GPUs, which are the very important AI chips that people need to run and train models. It rents those out from bigger cloud providers, and then leases those out to AI developers that want to then use those chips to either run or train their own AI models. So lots of people have referred to this kind of model as like a GPU reseller, as an inference provider, but they also offer a lot of other software that these developers might need.

19:29And so they offer software that will help you, you know, run your model faster or, you know, a big driver behind their growth is one of their products called Sandboxes. And so this is basically a way for companies to run their AI agents in this kind of closed-off, safe, isolated environment. So that way those AI agents, if they make a mistake, don't, you know, kind of wreak havoc and mess up a lot of other things on your computer or in your actual company's code base.

19:57Aaron Tilley:So inference provider, GPU reseller, does this classify them as a NeoCloud or are they not a NeoCloud? So I would say they definitely do fall in that NeoCloud kind of category or area. I would say that Modal itself probably would argue that they are more than just a reseller of GPUs and that they offer a lot of software to developers in addition to those chips themselves that either help them, you know, run or train those models faster or, again, any other, you know, helpful software like these sandboxes that, you know, helps companies run their AI agents, for instance. And then if we think about competition here, I mean, the names that come to mind on the NeoCloud side, we certainly have names like CoreWeave and Nebius.

20:43Aaron Tilley:But then on the inference provider side, what it's like, companies like Base10 and Fireworks, is that right? Yeah. So companies like Together, Base 10, Fireworks, many of these companies have really seen their revenue skyrocket in the last year as, you know, developers are, you know, using AI models a lot more. These AI models are generating a lot more tokens because they're being used for things like AI agents. So they're really part of this just super fast-growing space more broadly that has seen a really big uplift from the AI boom of the last year or so. So, Stephanie, are we at a point now, do you think, where winners are going to get picked?

21:28Aaron Tilley:Or are we still at a point where there's so much demand for compute and for GPUs, as we've reported on? You know, the GPU crunch is very – it's a real issue. You know, is there still way more demand than supply? I, you know, because at some point I think, look, does the world need 10 different inference providers or GPU resellers? I don't know. I mean, I think they're, you know, they're either going to consolidate or they're going to get acquired. The question is when. And if there's so much excess demand right now, I don't know. Maybe this goes on for a couple more years. Like, what's your read on this?

22:07Yeah, so that's the big question that I think a lot of investors are asking too. because, you know, as you're saying, these investors are trying to back the winner in these categories. But I do think right now it's very, it's still pretty hard to tell which one's going to come out on top, which is like kind of funny because you literally see these companies raising at, you know, four, five, six, seven billion dollars, which used to be kind of a sign that you were the winner of your category. But now it just kind of feels like every startup is getting these multi-billion dollar prices. So in this case, it does feel like because the demand for chips and for you know using ai models is just so strong that all of these inference providers all these companies are really seeing their revenue grow um i do think that in some cases like modal for instance there is still room for differentiation so uh with the sandbox product for instance it's something that they offer that not every inference provider offers and as i mentioned before is really important for things like ai agents um and so i do think that there are things that these inference providers can offer to make them kind of stand out a bit more from the rest of the crowd.

23:11But in general, we're just going to really see, you know, the rising tide lift all of these boats in the AI inference sector.

23:20Aaron Tilley:If you had an interview with the modal CEO right now, what would be some of the questions you would ask him? Yeah, so I think a big question that's on our mind is this kind of GPU crunch that we've been writing about a lot. So, you know, there's been just insane demand for these chips from the biggest model developers like Anthropic and OpenAI. And that has really pushed up the price on GPUs for everyone. And so that can be a problem for inference providers like Modal because they, you know, they're basically going out, renting those GPUs from cloud providers and then renting those GPUs out to other developers.

23:58So if the prices are going up for them, that's going to be kind of squeezing their margins and it's going to be making making it harder for for them to make a profit on those gpus that they're that they're going to go out and and rent out to other developers so i'd be curious what this gpu crunch is doing to their margins and if that makes that you know part of their business maybe less attractive and what are they going to do about it are they going to move into other areas are they going to are they going to consider maybe buying those chips themselves as we've seen other inference providers like together do.

24:31So yeah, lots of questions around, you know, margins and the compute crunks that I'd be very curious to ask them about. Great.

24:37Aaron Tilley:Well, Stephanie, I want to thank you for coming on. That is Stephanie Palazzolo, author of our AI Agenda newsletter here on TI-TV. Our next guest has firmly planted himself in many different corners of the tech sector. Gokul Rajaram is the founding partner at Marathon Management Partners. He has invested in a number of big names, including Figma, Deal, Vercel, and Rubrik, to name just a few. He is also on the boards of Coinbase and Pinterest, which sets us up for a wide-ranging conversation today. Gokul, welcome to the show. It's great to have you here. Thanks, Akash. It's great to be here. Thanks for having me.

25:13Aaron Tilley:So we've got a whole host of topics to get through with you, given all the areas you play in. I do want to start with crypto, because crypto, it's very newsy this week. We've got this bill, which is we are waiting to see if it gets approval from the Senate Banking Committee. You're on the board of Coinbase. And so I just want to get your framework for how you think about this bill. And specifically, I'm hoping you can sort of explain to us from your perspective, the ripple effects of this bill, if it inevitably gets approved, and then the ripple effects if it doesn't. So I think the digital, I think it's called the Clarity Act.