Anthropic Sues Pentagon, OpenAI IPO Investor Skeptics, New Groq Chip Reveal at Nvidia GTC

0:13Stephanie Palazzolo:Welcome, everyone, to The Information's TITV. My name is Akash Pasricha. It is Monday, March 9th. We have news this morning that Anthropic is suing the U.S. government over the fact that it has said the AI company is a supply chain risk. Our AI reporter will join the show to discuss the latest development. We'll then look at how OpenAI's latest GPT 5.4 model is faring and also get into some inside reporting we have about a new audio model that OpenAI is trying to develop. Next up, our financial analysis columnist published a piece this morning about the big questions OpenAI faces from the investor community as it marches towards a likely IPO later this year.

0:55Stephanie Palazzolo:And finally, ahead of Oracle's quarterly results on Tuesday, our cloud and compute reporter will offer some context on headlines that have been stirring about the company's data center progress. We'll also preview what to look for next week at NVIDIA's GTC conference. And we'll also talk to her about some inside reporting that she has on Google's TPU traction in the AI ecosystem. It's going to be a fun show, so let's get right on into it. We have big news this morning. Anthropic is suing the U.S. Department of Defense over its supply chain risk designation. I want to bring on our AI reporter, Stephanie Palazzolo, to discuss the news.

1:33Stephanie Palazzolo:Stephanie, so we were expecting this designation to come. Now that it's here, what does it mean for the story and how it goes from here?

1:41Anita Ramaswamy:Exactly. So as you mentioned, this is something that we have definitely been expecting. Anthropic has been signaling really all throughout this conflict with the Pentagon that if the Pentagon were to designate it a supply chain risk, which it did, Anthropic would sue them, basically. So kind of what this means for the story is that, you know, obviously now this will go to the courts and kind of be decided there. As my colleague Rocket reported this morning, and from all of our other, you know, just other conversations as well, it does seem like this lawsuit will probably go in the way of Anthropic.

2:21Anita Ramaswamy:specifically just kind of boiling down to the fact that the supply chain risk designation is mostly meant for kind of foreign companies that are meant to you know it's meant to prevent these companies from from from spying from doing kind of outright sabotage so these are kind of very extreme measures that you wouldn't really expect to be placed on a you know on a domestic like u.s based company like anthropic um so it seems like right now obviously things are very early, but a lot of lawyers that we spoke to do think that Anthropic is likely to kind of win this suit.

2:53Stephanie Palazzolo:And, you know, the thing that I'm sort of thinking about is how long this takes to play out. And I'm thinking about the IPO timelines that we have for OpenAI and Anthropic. And look, I don't think it's going to resolve itself very quickly. We know that things in the courts move slowly, but it does kind of cast a bit of a cloud on these IPOs happening at all later this year, potentially and what it says about the narrative here?

3:20Anita Ramaswamy:Yeah, I think this whole saga has really introduced a lot of uncertainty into Anthropik's IPO plans, which we've said, you know, both Anthropik and OpenAI are really gearing up to be IPO ready, potentially could go public as soon as later this year. So it definitely is something that, you know, investors have been very concerned about, you know, has raised a lot of questions. We've actually even seen, you know, some of Anthropics investors get directly involved with talking with the Pentagon and trying to work things out. So it definitely seems like for Anthropics, it's in its investors' interest to settle things as quickly as possible.

3:58Anita Ramaswamy:But obviously, courts do take a bit of a time. So we'll have to see kind of how long that process takes.

4:05Stephanie Palazzolo:Now, I want to pivot for a moment here to talk about another piece of news that you wrote about last week. You wrote about OpenAI's 5.4 or GPT 5.4 model and the new release of it. You actually came out ahead of time and said that it was coming out. It came out on Friday. What is different about GPT 5.4, Stephanie?

4:27Anita Ramaswamy:So GPT 5.4 really just kind of continues OpenAI's trend of making its models more useful for the sort of real work that knowledge workers are doing. So for instance, OpenAI really highlighted in this latest model that it's much better at working with things like spreadsheets, presentations, and documents, which you could easily imagine why that would be so useful for people working in finance or investment bankers, those sorts of folks. There's also a new X-high, is what they're calling it, reasoning setting, where the model can basically spend even more time and tokens kind of thinking about super tough problems to get those answers right, which is something that we broke in our newsletter days before the release.

5:12Stephanie Palazzolo:So it sounds like this is very much the model that is continuing to go head to head against Anthropic with that. We hear this term knowledge work come up. I mean, this is really like the white collar work that Anthropic has really gotten a lot of traction with, with its integration with Microsoft, with the Office suite. We've heard about Aaron Holmes reporting with PowerPoint, stuff like that. We obviously know the coding. So this is very much an answer to that, it seems like.

5:38Anita Ramaswamy:Yeah, you know, as we've seen, obviously OpenAI and ChatGPT has really been the breakout star with consumers. But we've seen in recent months, OpenAI is really emphasizing wanting to work more with enterprises too, which is what, as you mentioned, Anthropic is really focused on. So they're really trying to compete with the Anthropic models in that way to make sure that they can, you know, OpenAI models can work well with spreadsheets and presentations. And, you know, just a couple weeks back, we saw a similar announcement from Anthropic announcing specialized features and models for people working in finance too.

6:11Anita Ramaswamy:So it's very much like this very tight race between the two companies and a lot of back and forth in terms of one company making an update or announcing a new feature one week that will make it better for knowledge workers. And then you see the other company kind of doing the same maybe a couple of weeks later.

6:27Stephanie Palazzolo:Any initial reactions from folks? Are they liking it?

6:30Anita Ramaswamy:Yeah, it seems like overall developers are pretty pleased. it seems like people think overall it's more intelligent it's a lot faster which is always a good thing with these sorts of models I do think though that like generally the developer reaction to these new models have been kind of it's been more muted than in the past because companies like OpenAI and Anthropic have moved to this sort of like once a month model update where it's at this point it's kind of smaller tweaks and smaller updates to the models every time they come out with the new version. So because of that, there's not these giant jumps in capability that you would see kind of written about in tons of news stories.

7:10Anita Ramaswamy:And it kind of just feels like developers are just switching to whatever model is newest. So there's no loyalty really to any specific model developer.

7:18Stephanie Palazzolo:Yeah, I mean, it's almost like making a big press release about an Apple software update at some point. I feel like that's the direction that we're going in. But as we know, at one point, I mean, these were they were keynotes around new models. And so maybe the days of those keynotes are long gone. But a new model that I do want to ask you about is you wrote last week also about a bi-directional model that OpenAI is working on. And this is specifically to do with the audio models that it's developing. Tell us a little bit about what that means.

7:53Anita Ramaswamy:Yeah, so I need to give a little bit of a backdrop here. As you mentioned, this is an audio model. And audio models are really interesting because for a while they were kind of like the black sheep, I guess, of AI or kind of like the forgotten child, like not a ton of money and effort was going into it. But, you know, especially in the last year or so, we've seen audio models become increasingly more important. And lots of AI researchers are talking about them a lot more. And essentially that's because a lot of people are kind of hypothesizing that audio models, which is when you kind of talk to the AI out loud, like the way humans do with each other, is a more natural way to interact with AI and is a lot easier than having to text out every single thought that you have to a chatbot.

8:37Anita Ramaswamy:The issue is that today audio models are kind of trailing behind text-based models in terms of just how intelligent they are, just their general capabilities. But one issue here that's really important is that audio models today don't feel very natural to talk to. So you might have noticed when you turn on advanced voice mode in ChatGPT, for instance, and you're talking to it, if you try to interrupt the model or even just say something like, yeah, just to like indicate that you're listening, the model will just kind of completely stop talking, like pause for a second. It might get confused. It might ask you to repeat yourself.

9:11Anita Ramaswamy:It is kind of like will create this very unnatural kind of environment, I guess, for you to kind of be talking to this model in. So OpenAI's bi-directional model, or BiDi for short, is really kind of going after that specific issue. So this model is real-time, which means that it's always kind of processing audio even while it's talking to you. So in this case, if you do interrupt the BiDi model kind of mid-sentence, it will not pause, but it will instead be listening to you and updating its answer in real-time. So you can kind of imagine why that would be really important for a bunch of use cases.

9:49Anita Ramaswamy:But like customer support, for instance, like let's say you're on the phone trying to return some clothing. Mid-conversation, you decide you want to exchange sizes instead. You can interrupt the voice agent now, and it will just naturally kind of respond to you and be like, okay, sure, yeah. Here's actually the button that you press to do that instead. A much more natural kind of conversation.

10:08Stephanie Palazzolo:So basically what I'm hearing from you is, I mean, the audio models here, you know, one of the things that they have difficulty with is when you interrupt the device or the person to confirm what you're saying. Same way I do with you. It's like, yeah.

10:25Anita Ramaswamy:So like literally like what we just did, like where I said.

10:28Stephanie Palazzolo:What we just did. And so the model cannot handle that. And so this new model will hope to rectify that. And I imagine, I mean, this will become more important as OpenAI looks to build out its hardware ambitions. And, you know, you reported on this smart speaker-esque device that the company is looking to launch. I mean, you've got to be able to handle those yas and ahas.

10:51Anita Ramaswamy:Exactly, yeah. That's actually the smart speaker and the device work that OpenAI has been doing was actually, you know, one of the main catalysts for them, like really ramping up work on the audio models and on this bi-directional model specifically, because, you know, as you're saying, if you're talking out loud to your assistant, it's going to be really awkward if you have to like wait for the assistant to finish rambling on every time you want to ask it something and you can't even say like, yeah, or, you know, kind of just acknowledge what it's saying while it's talking. So you just kind of want it to feel more like you're having a conversation with a human rather than like this robot that we're used to whenever you like call customer support.

11:26Stephanie Palazzolo:And where is this model at in terms of progress? Do we know about when it's going to be released at all?

11:31Anita Ramaswamy:So it seems like it's still pretty early. I think originally OpenAI was aiming for a Q1 release, but it seems more and more likely that that will probably get pushed to the second quarter of this year. It does seem pretty early in that, you know, people I talked to for the story said that the longer you talk to this model, you know, maybe it works really well for the first couple minutes, but after several minutes of talking to it, it will start glitching or speaking in these weird, abnormal robotic voices. It's definitely still a work in progress, but I think it's one that a lot of us are really excited to test out, hopefully when it comes out in the next couple months.

12:07Stephanie Palazzolo:Great. Well, Stephanie, I want to thank you for coming on. That is Stephanie Palazzolo, our AI reporter here at The Information. As OpenAI marches towards its likely IPO, either later this year or early next year. A big question concerning investors these days is what valuation the company should seek, how much it's actually worth, and whether or not there is appetite from the public market to buy into Sam Altman's story. The Informations Financial Analysis columnist, Anita Ramaswamy, wrote a piece on those exact questions today, and I want to bring her on to talk all about it. Anita, welcome back to the show.

12:43Stephanie Palazzolo:It is great to have you here. I want to start with this big question. What is OpenAI worth?

12:49Anissa Gardizy:That is a big question, Akash. And my answer is really what the market says it's worth, which sounds like a cop-out, but we actually have a number because OpenAI is reportedly closing a fundraising round at an$850 billion valuation.

13:03Stephanie Palazzolo:Okay, so we have that number. Now, the number that I want to talk about, though, is, look, when it goes public, how much appetite there is from public market investors, what they're willing to pay for it. And you did some good numbers in your piece today that I want to walk through. What is the current multiple that OpenAI's valuation there is at? And how does that compare to the top AI companies today?

13:24Anissa Gardizy:Yeah, so it's a good question. I took a look at OpenAI's valuation multiple on forward revenue relative to some other companies. So we had reported earlier OpenAI's revenue projections. And so as a multiple of their projected 2026 revenue, OpenAI is valued at a 28 times multiple. If you take that and you compare it to, say, Anthropic or NVIDIA, we have Anthropic closer to 20 times. We have, where they last fundraised, we have NVIDIA closer to 12 times. So it really puts into perspective that OpenAI's valuation is looking rich. And this is something that investors are talking about or debating and discussing as they get ready for the offering.

14:03Stephanie Palazzolo:So the NVIDIA comparison is kind of interesting because these are the two most foremost names in AI. NVIDIA, of course, has the benefit of having free cash flow and having a lot of it. And so I wonder, when you talk to people, I mean, there are two different business models. It's still the AI narrative. How do they balance a 28x multiple for a company that doesn't have any free cash flow against a company that very much is?

14:30Anissa Gardizy:Yeah, I think that's one of the primary areas of concern for investors who I spoke to. So in this piece, my colleague Miles and I reported on all these different public investment managers, we talked to more than 11 people about how they would think about allocating their portfolios to OpenAI. And all of these people that we spoke to are all tech investors. They invest in the public markets. And really the big theme and the big concern that came up, even amongst those who were really bullish about OpenAI, those people who said, you know, they have a great business model and we really do see them getting to profitability at some point in the future, they were still concerned about the current valuation.

15:05Anissa Gardizy:If you think about NVIDIA, like you said, Akash, they have real cash flow. They have therefore a multiple of free cash flow because we can actually calculate one. OpenAI, meanwhile, doesn't plan on, and this is based on OpenAI's own projections, doesn't project that it's going to be generating cash until at least 2030. So it's still a way away. And that's why a lot of the investors are concerned that if this IPO comes out at a really high valuation, even if they like the business, they might not consider buying it because of that.

15:34Stephanie Palazzolo:And you spoke to some people who are actually planning to bet against the company, right?

15:39Anissa Gardizy:I spoke to some investors who said they would consider shorting the company. Yeah, exactly. Depending on what valuation they go out at.

15:46Stephanie Palazzolo:Right. The other comp that was interesting to me that you looked at was Palantir, a company with an astronomical valuation on just about every multiple. How did that comparison sit with people you talked to?

15:59Anissa Gardizy:So Palantir has a really high valuation. If I'm recalling correctly, it's trading at over 40 times forward revenue. So that is definitely richer than where OpenAI's investors have ascribed at value, at least in the current private funding round OpenAI is assembling. But there is a key difference between Palantir and OpenAI, which is something that some investors pointed out to me. The idea that Palantir has more flexibility to manage their costs. So with OpenAI, they're subsidizing a bunch of these free users, essentially. who aren't paying anything for their product, but OpenAI still has to pay for the cost of computing power to actually give those users the service that is ChatGPT.

16:37Anissa Gardizy:Whereas what happens with Palantir is they're a software company, they're software and services, and they have a little bit more flexibility. With the example one of the investors told me was, if you're OpenAI and you say, we're going to build this multi-billion dollar data center, to then say, never mind, and walk it back later on if you realize you were overspending is a lot more difficult. And so this investor compared OpenAI to sort of like a cargo ship, and he said Palantir is like a Formula One car because they can cut costs, they can easily trim headcount, they can find other places in their budget for operating expenses that can help them contain their costs, whereas OpenAI will have a lot harder of a time doing that.