SpaceX’s IPO Filing Incoming, OpenAI Kills Sora, Arm Launches CPU Chip, Meta Layoffs

0:13Stephanie Palazzolo:Welcome everyone to The Information's TITV. My name is Akash Pasricha. It is Wednesday, March 25th. We have a big story out of our newsroom. The information is exclusively reporting that SpaceX is aiming to file confidentially for its IPO as soon as this week or after the next week. later next week. There are a few interesting details that I want to highlight to you about this story. The first thing is that SpaceX could raise more than$75 billion. That is up from the $50 billion that we previously thought that the company was going to raise. And the amount here is significant because it actually would be more than all of the other IPOs in the past year combined.

0:51Stephanie Palazzolo:That is according to data from Renaissance Capital. And it really highlights the importance of this offering. The second piece that we are reporting is that normally banks limit individual investor allocations to about 10 % for an IPO like this. But for SpaceX, we have reporting that individual investor allocations could be more than 20%. And if you think about this, I mean, Elon Musk has a lot of fans in the retail trading community, and you can imagine they would be quite excited for an IPO like this. And so it certainly is a strategic move. The third piece that I want to highlight is that this IPO might not have a standard six-month lockup period.

1:29Stephanie Palazzolo:Normally, these lockup periods are there so that early investors can't just dump their shares right away. We don't know the details about what the lockup period will be, but what we do know is that SpaceX has raised a colossal amount of money privately, and so there has to be some careful considerations about when those early investors can eventually sell. We will make sure to keep be posted on that, but the full story is on our website at theinformation.com. Now for some more exclusive reporting. Today on the show, OpenAI is doing away with Sora, and CEO Sam Altman is making some big changes to his own responsibilities at the company.

2:06Stephanie Palazzolo:We'll hear what he's shifting his focus to. Plus, Arm is getting a boost today after the company unveiled its first in-house CPU chip. We'll talk about the news with a longtime chip startup venture capitalist. It's going to be a big show, so let's get right on into it. Sam Altman is making some big changes at OpenAI to continue to focus the company on a narrower set of ambitions. The information reported exclusively that he is changing his own focus at the company. He is shuttering Sora, and OpenAI also has a new model coming. To talk more about what we know, I want to bring on Stephanie Palazzolo, our AI reporter.

2:43Stephanie Palazzolo:Steph, welcome to the show. It's great to have you back. Thanks. Great to be here. Okay, so you got a hold of this memo that Sam Altman, the latest of many memos that Sam Altman puts out there. What did the most recent memo say about all these changes that the company's making? So there are a couple of key things in this latest memo. I think first, a big product shift where the company is, you know, Sam basically told everyone that they're shutting down all of their Sora-related products. So Sora is their video generation model. So this includes the kind of TikTok-style mobile app. This includes the API for developers, and they're also reversing on plans that we had originally written about earlier this month to move those video capabilities into ChatGPT.

3:28So now the Sora research team is going to be focused more on world models and kind of use cases that are related to the field of robotics. Another interesting change was Sam had also told employees yesterday that he'd be moving the safety and security teams, which previously reported up to him, to other parts of the org. So specifically, safety will be moving under research, reporting up to Mark Chen, and then the security team will be moving to the quote-unquote scaling org, which is run by co-founder and president Greg Rockman. And then the last change was Sam kind of hinted a bit at a new model that the company has finished pre-training.

4:06Its codename is Spud, a continuation of weird codenames from OpenAI. But he kind of hinted that he expects it to be a really strong, strong model. So something to look out for there as well. You're not a fan of Spud? You don't like the name? I like Spud. I'm just like, what is the overarching theme here? You know, we've had garlic, we've had like shallot peat. I'm just like... Strawberry. There was strawberry too. True. Right. So enjoying all the food-related code names. Right, which end up just being decimal points in the long run in the model hierarchy. Okay, so look, I want to rewind. So let's talk about Sora.

4:45So they're doing away with Sora. You said they're going to focus on world models. What is a world model and what does this tell us about their strategy at large? Yeah, so a world model is a type of model that's become increasingly more popular in the last year or so. Essentially, it's a model that's able to kind of like recreate the physics of the real world. So you might have seen from like some Sora videos, some awkward physics kind of errors going on where maybe you knock over a glass of water and the water doesn't come out of the glass. So a world model would, you know, theoretically be better at mimicking the physics of the real world so that in a video that it generates, If you knock over a glass of water or, you know, a mirror breaks, it will kind of look like how it would if that happened in the real world.

5:33So, you know, with this, you know, that has very strong use cases in robotics, which is something that OpenAI has also been working in for quite a while. It's actually kind of one of the earliest projects that OpenAI ever went after. And it's like very, very early days. So it kind of just feels like they are cutting back on divisions of the company that might be taking up a lot of compute, like video generation, and trying to put those resources elsewhere that might be more promising. So this is kind of interesting because the world model, I mean, I figure something like that would be very applicable to the hardware play that OpenAI could be working on potentially.

6:15I don't want to skip the Sora piece, though. I mean, this is huge. I mean, the Sora was, I mean, it was very controversial. We saw that Disney said that they were going to invest a billion dollars in OpenAI. That is no longer happening is the latest update. And I mean, you know, the core reason, I mean, from what you're reporting is, I mean, it was draining a lot of compute and there was not a lot of ways to monetize it. Or what do we know about that? Yeah, I mean, I think it's essentially boiled down to that, which is, you know, yeah, it took up a lot of compute, which, you know, OpenAI needs a ton of to serve its models.

6:53And then, as you mentioned, we have reported in recent months that usage of the Sora app has kind of dropped off a bit. It wasn't really clear, I think, that the Sora app was going to become a huge success in the way that TikTok has or some of these other kind of video, social networking type apps. So I think, yeah, I think they kind of just decided to stop the bleeding wall early and move those resources elsewhere. Okay, and so let's go to the new model then, Spud. What is it supposed to be good at? Yeah, so details on this are very vague. I think all Sam really said is this new model, Spud, has finished pre-training, which is kind of the first stage of model development.

7:36He also said that the company expects to have a very strong model that's based off of Spud in a few weeks, and the team believes this model will be able to accelerate the economy, whatever that kind of means. He also said that things are moving faster than any of us ever expected. So obviously, this all sounds great. But, you know, of course, take this with a huge grain of salt just because we haven't actually seen this model in production. We haven't gotten to test it out yet. And, you know, it's easy for leaders to say very optimistic things to employees. But definitely, I would say overall, like not a negative sign.

8:08I mean, it's definitely something for us to be excited about and, you know, a new model to look forward to, I think, in the next couple months or so. And then let's go to Sam Altman's new work focus, new refined focus, I guess. So remind us again, so what is he doing away with? What is he allocating more responsibility to with respect to Fiji Simo's role? You know, how is all that looking and what do you make of it? Yeah, so this is kind of a continuation of a trend we've already seen over the past year where Sam is kind of taking a step back from these like day-to-day operations of the company And so this all kind of started, you know, about, you know, a year ago or so, whenever Sam brought on Fiji to handle kind of product and operations.

8:52And he kind of took a step back from that to focus more on research and compute. And then this is kind of another step in that direction where you see him moving safety and security under other leaders so that he can, as he says in the memo, focus on things like raising capital, supply chains, and building data centers. So obviously, I think OpenAI is a huge, sprawling company. And I think Sam, as a single person, cannot handle everything. So I think he's just really trying to focus on what he's good at. And as somebody that has been in tech and in the investing world for quite a long time, I think that is, you know, raising capital is something that he's clearly very good at, that he's been able to do a ton.

9:33And then also just generally kind of rubbing shoulders with these leaders around the world, whether that's business leaders or people to help him build data centers. And so it just seems like he's doubling down on kind of what he knows he's good at. So when you look at these changes, and they will continue to come, I mean, I just wonder, Steph, do you think this is a positive change for the company? And again, when we say positive, you have to really define what criteria on which you're evaluating it. But I think really, as we look ahead to an IPO, and we look at the company's financial profile, we look at usage metrics going up in the right categories.

10:12I mean, they're saying it's a smaller subset of products. They're really going deeper.

10:17Stephanie Palazzolo:Feels like the right move to me, but I don't know. What do you think? Yeah, definitely agree here. I think even though it is a painful move, I think for many within the company who were excited about video models, excited about Sora, I think ultimately this is the kind of right choice for OpenAI just because, as you're saying, They're gearing up for an IPO. Investors want to see a company that is focused, that knows what is working well and what's not, and is able to kind of invest its capital in the right types of products. So I do think that, you know, this might be painful in the short run.

10:53I think, again, researchers feel very strongly about what sorts of projects they get to work on. So maybe we'll even see some of the sore researchers leave in the next couple weeks. But ultimately, I think it is the right choice. And I think it shows that, you know, ever since VG came in, she seems really willing to cut things that folks have been saying for, honestly, for quite a while aren't really working in favor of the company. Right. Well, Steph, I want to thank you for coming on. That is Stephanie Palazzolo, our AI reporter, here at The Information. Arm announced it is launching its own CPU chip developed in collaboration with Meta.

11:28It is the latest semiconductor to join the heated chip race that is seeing just about every major company expand its product suite to overlap with many of its own customers.

11:38Stephanie Palazzolo:To explain some more of the dynamics here at Play, I want to bring on Umesh Padwal, a longtime chip startup venture capitalist and a managing partner at Seligman Ventures. I'm really excited to talk about this. So, look, the way that I understand this new chip that ARM has, ARM traditionally has been the blueprint designer for these chips. It hasn't really made them itself. Of course, nobody's really making them because TSMC is making them all. But I mean, it hasn't really done its own chip. It's really given away its blueprints at a cost, of course, to other companies. Now they're saying, we want to be in the market ourselves with the CPU chip.

12:17Stephanie Palazzolo:Is this a smart strategy for them, Amesh? What do you think? Akash, it's a very bold move because essentially they're competing with their competitors. ARM started off as an embeddable core. So you customize the chip around the ARM core, which is the CPU, customize things like DSP, memory, and other IOs to interface with the external world. So it was offered as a core, and then customers around the world, thousands of them, customized it up to their needs and sold into the market. They were used extensively by Apple. by Samsung, by Intel, by many people for the embedded market. But now they've gone up the stack and they're making standard product, which is a complete CPU going after the x86, Intel, AMD, as well as some of their licensed producers like Vera from NVIDIA, as well as Qualcomm, Apple, which are licensees.

13:23and some of these CPUs using ARM core are developed by the hyperscalers like Amazon, Google, as well as Microsoft. So now they're entering a field where they're not only competing with the x86 architecture with Intel and AMD, but they're also competing with ARM-based hyperscalers as well as formative competitors like NVIDIA. Okay, so, and that's sort of what my read on it was, was, I mean, they're making a bold bet here. Do we know anything about how this chip will compare

13:58Stephanie Palazzolo:to those ARM-based chips and to the x86 architecture? Is there any evidence that it will be better or worse? Or is there just so much demand that people would need to diversify? How are you thinking about it? I think that we don't know the performance numbers yet. We'll find out over time. But I think this is trying to have alternate sources for the hyperscalers, which want choices. Right now, you have Vera and Rubin, which are coming from Jensen at NVIDIA. And there are custom chips being developed by many of the hyperscalers themselves based on ARM, as well as they do the AI accelerators, which are the GPUs.

14:39So any given data center has a CPU as well as a GPU. And I think meta strategy is to diversify their sources from NVIDIA as well as other suppliers where they have the AI accelerators from three, four suppliers. And then they have the x86 and now the ARM-based CPU. So I think it's a diversification, not relying on the dependence of anything. That's what the hyperscalers would like. It's an alternate source for the CPU. Can you explain to me, Amesh, I mean, you've worked with a number of chip startups over the years, and this is not the first time that any company has begun to compete with their customers.

15:27And it's not just happening in the chip sector. It's happening in various different places in the AI ecosystem.

15:33Stephanie Palazzolo:I mean, I guess I'm just curious, you know, how much tension that brings into any kind of these dynamics when you're negotiating contracts with customers. We saw that NVIDIA, certainly the hyperscalers, I mean, they all provided very flattering statements for ARM in the press release saying, we're super excited about the CPU. And yet there's competition there. I mean, does this have any tactical bearing on negotiations? Yes. I think this is really a bold move to compete against your customers. So you're going off the value chain. you are enabling your customers to design chips which are customized for your applications around the world.

16:18And there are millions of customers. Now you're going to go up the stack and directly compete with people who you have license to. So you really need to have a very strong conviction that your architecture is better than any one of the other people competing with you from x86, as well as ARM-based licensees like NVIDIA, like the hyperscalers, like Qualcomm, like Apple. Right. But I mean, do you think that this chip could cannibalize

16:48Stephanie Palazzolo:the existing business that ARM has in selling those blueprints to those customers? I think people are going to be wary. And there are alternatives to ARM, which are also coming in the market. There's something called RISC-V's ARM instruction set compatibility core, which is available without any royalties. So now the customers are going to look at this move and say, they are competing with me. Do I go to open source licensing of a RISC-V engine for many of these companies which are in the market right now, which are their competitors? It's like companies which have provided these new competitive cores.

17:34So do you think that happens? I mean, let's make predictions here. Do you think the business suffers? My prediction is that there are going to be customers who are going to look at alternate source just to make sure that the person they're licensing from is not going to compete with them. So they're going to have their second sourcing strategy. So it's a brilliant market dynamic going on right now. I'm coming up competing in some segments with x86 and also their licensees. But then their competitor in the embedded market, which is the RISC-V companies, are going to say, hey, look, you can't rely on just ARM.

18:15You should also use our RISC-V architecture and make sure you're not dependent just on ARM. So it's a fascinating time in the industry in terms of, they call it cooptition. And I think the best architecture and the best chips are going to win in the market. Great. Well, Amesha, I want to thank you for coming on. That is Umesh Padoval from Seligman Ventures here on TITV.

18:40Stephanie Palazzolo:That does it for today's show. A reminder, we are on this stream Monday through Friday at 10 a.m. Pacific, 1 p.m. Eastern. I want to thank you all for tuning in. We really do appreciate your viewership. Make sure to subscribe to the information on YouTube and follow us on X, Instagram, TikTok. Check us out wherever you get your podcasts. I'm already excited for our next show tomorrow. Have a great rest of your Wednesday. Bye-bye for now.

19:07Thank you.