Nvidia $100B OpenAI Credit Support, Tesla’s Radical ‘Flying’ Roaster, Dario Addresses AI Backlash

0:13Welcome, everyone, to The Information's TITV. My name is Akash Pasricha. It is Monday, August 17th. Today on the show, NVIDIA is backing OpenAI and SB Energy's massive new Ohio data center with roughly$100 billion in guarantees. We'll unpack that deal and NVIDIA's SB Energy investment with our NVIDIA reporter shortly. We'll then unpack new reporting that we have on the Tesla Roadster. And we're going to close out the show with a discussion on Dario Amadei's latest thoughts on AI regulation and the debate around AI centralization that has made waves on X for much of the weekend. It's going to be a great show, so let's get right on into it.

0:58NVIDIA announced this morning it will provide a financial guarantee for the lease on a new Ohio data center that is expected to amount to 10 gigawatts of new power capacity. The company also announced it will take a state in the data center developer, SB Energy. My colleagues Phoebe Liu, Valida Pau, and Ann Davis Vaughn had exclusive details about much of that investment over the weekend. I want to bring on Phoebe to walk us through what we need to know. Phoebe, welcome back to the show. Happy Monday. Happy Monday. Good to be here. Okay, so we are getting right to it this morning with a big deal that NVIDIA has announced.

1:37There was a lot in this press release. And of course, you and Vlita and Anne reported many of these details over the weekend. What did they announce this morning? Let's start there. Yes. So NVIDIA, OpenAI, and SB Energy basically announced that they've finalized the agreement for OpenAI to lease data center and power capacity that SB Energy is building. And that will house NVIDIA hardware. So that's kind of the incentive for NVIDIA to say that it will guarantee a lot of the powered shells, the data center power and land lease agreements that OpenAI has signed. Because basically OpenAI doesn't have the credit to be able to kind of lease that on its own and convince lenders to back it.

2:23So it would need kind of an investment grade company with a strong balance sheet such as NVIDIA to back it. And yeah, so they're partnering to build... NVIDIA didn't say the amount, importantly though, they didn't say the amount today in the press release, right? It didn't in the press release, but it actually came out in a regulatory filing a few minutes afterward. The total of exposure is about$105 billion, which importantly is not the amount that NVIDIA is paying, but it's the amount that NVIDIA would help guarantee and that would lower over time. We can get into that later, But it's pretty close to the amount that Valida, Ann, and I reported on Friday of around$100 billion.

3:04And just to make sure that we get the details here correctly, so OpenAI has contracted – well, SB Energy, the energy company that is backed by SoftBank, they're the ones that are developing the data center. They need a tenant, which is OpenAI. OpenAI has committed to a lease for this facility. And in doing that, they've said, build it. We will come. they need some sort of a guarantor, I guess, as we've talked about on this show. It's much like the New York City rent guarantors. NVIDIA has come to the rescue. Has NVIDIA ever made a guarantee of this size before? I think this is pretty unprecedented in terms of size.

3:47However, NVIDIA's previous guarantees are a bit more airtight than this one. So for example, in September 2025, NVIDIA said it would buy back all of CoreWeave's unsold GPU capacity. So this means if CoreWeave doesn't sell this capacity, NVIDIA would have to buy 100 % of it back at some kind of rate that it's determined. The way this one works is a little bit different. So if something were to happen, if OpenAI, hope this doesn't happen, but defaults on its lease in some way or can't make xyz payments uh nvidia wouldn't directly take over the lease uh they would try to they as in kind of the companies collectively would try to find another tenant to lease whatever capacity that open ai can't pay for um at some rate and nvidia's guarantee covers like a very low base rate so what nvidia would do in this case is to step in and kind of cover the difference which hopefully would be much, much lower than the$105 billion.

4:52That is kind of its maximum exposure as part of the deal. Now,$100 billion, you had some good reporting over the weekend on how we got to that point. Was it always$100 billion? There was a discussion earlier a couple months ago that it could have been even more than that at one point, right? Yes, it depends on who you ask. Some people really don't. Yeah, basically, in late July, the Wall Street Journal reported that the guarantee could be as high as$250 billion. Some people are very insistent that this amount has not been scaled back, but rather phased into two parts. So now the initial guarantee covers the first 4.5 gigawatts of the data center.

5:34There's a second phase that would be 3.75 gigawatts that NVIDIA is technically not on the hook for yet. So basically, if it's been phased, the initial amount that NVIDIA is guaranteeing is inherently much less than what was described before. And I think different people have different interpretations as to whether this means NVIDIA is kind of worried about or kind of taking precautions as to not use too much of its balance sheet directly up front to help its customers buy NVIDIA hardware. But I think NVIDIA is more and more putting out the message that it wants to make sure and signal to investors that it's being very financially disciplined and kind of capping the amount of exposure of any particular deal that it itself would be liable for.

6:26So in the Ohio deal, it's about 25%, according to a couple of people I talked to who are involved in the deal. You talked about the NVIDIA hardware, so the chips in all this. I just want to make sure I understand here, where do the chip purchases fit into this story? I mean, I think you mentioned it would take$500 billion in total to construct this facility. So$100 billion, it's a proportion of that. So they're not covering the whole thing. But regardless of however much they're covering, are they covering loans that would then be purchases of their own hardware? Or how do the chip purchases fit into this?

7:07Do we know? Yeah, so today's announcement, I think, specifies that the$100 billion would just cover the PowerShell. It's kind of unclear where exactly the chips are. So just the PowerShell means just the data center? The data center, the power, the land, kind of the buildings that would be constructed, which is also pretty expensive. But the chips are important because they're even more expensive. And yeah, it would take a lot of debt because OpenAI doesn't have that much cash right now to be able to afford. So I think my sense is that they would need separate financing, either slightly overlapping with this deal or completely separate, which is part of the broader discussion around kind of how NVIDIA announced last week that it's partnering with these five Wall Street giants to help raise third parties with non-NVIDIA capital to help NVIDIA customers finance purchases of its chips and other hardware.

8:04okay so this is this is actually kind of staggering i didn't really put the pieces together here but so you've got this 500 billion dollar data center that is being built in ohio nvidia is saying hey we'll put up 100 billion dollars if things go south to cover the lease commitments but then there's 500 billion dollars in additional financing that nvidia is basically working with wall street on and we know that chips are you know i think there's a stat out there chips are like 60 of the cost of a data center in the end. So this is kind of a staggering amount of capital that is being committed here.

8:39I want to talk about the second half of the story here. So NVIDIA is also taking a stake in SB Energy, the data center developer. Who is SB Energy and what should we make of this investment? Yeah, so SB Energy has its roots in kind of power generation, power development, but it acquired a data center company late last year. So now it's kind of doing both the power and the data center development itself. It is majority owned by SoftBank, trying to go public kind of shortly, probably this year. And also kind of OpenAI and now NVIDIA have pretty major stakes in the company. So it's kind of this central player, one step removed from kind of NVIDIA's direct customer in the supply chain that would be necessary to make this 10 gigawatt Ohio project possible.

9:35And part of SV Energy's connection to the project also involves relationships with the Japan government and the US government as part of trade negotiations from earlier this year. So there's kind of a government component to this as well. So NVIDIA's$1.5 billion investment, I wouldn't say it's exactly circular because it's not directly to an NVIDIA customer, but it's all kind of pretty intertwined, if that makes sense. Right. Although you could say, I mean, it's not to an NVIDIA customer, but if they're going to build the data centers that tenants take leases in to buy NVIDIA, it's not as circular.

10:17Let's put it that way. It's not as circular. Yeah, but I think some people think that this equity investment would help incentivize NVIDIA to also guarantee the second phase of this project, which is also key. And you also reported that it's$1.5 billion that they've committed right now in terms of an investment. There could be more money coming, perhaps. Could be another$1.5 billion at the IPO, according to some of the people we spoke to. And that would, if the numbers that SB Energy is trying to raise that the IPO doesn't change very much, that would make NVIDIA one of the largest investors in the IPO, which is also interesting.

10:56Right. So let me ask you a broader question here, Phoebe. I mean, you know, we see all these deals coming together. How risky are they? What do they tell us broadly about NVIDIA strategy? Is it more of the same or are these now nuanced ways that NVIDIA is trying to insert itself into the AI build-out? Yeah, so I think both for this PowerShell lease guarantee and the$500 billion partnership NVIDIA announced with the big asset managers, a little while ago, NVIDIA is trying to cap its risk at 25%. So I think they're making a very intentional step to kind of signal that they want to cap their direct risk to all of this.

11:39And I think if you asked Jensen, he would say there's no risk of DePaul. Like they're just, because demand so far outpaces supply right now. I think from NVIDIA's perspective, the chance that they will even have to pay any of this backstopping is close to zero. I think other people may see it differently. I think bond traders' assessment of NVIDIA's credit risk has gone up over the last few months. And that kind of is a signal of investors being a little bit worried that maybe the whole AI ecosystem is becoming too hinged on NVIDIA. So that's a metric that we should continue to watch going forward.

12:18Great. Well, Phoebe, I want to thank you for coming on. That is Phoebe Liu, our NVIDIA reporter here at The Information. Will your Tesla be able to fly? The Information has exclusive reporting that Tesla has big plans for its Roadster vehicle. Our Elon Musk reporter, Grace Kaye, wrote that story. I want to bring her on to walk us through some of her great reporting. Grace, welcome back to the show. It's great to have you here. Hi, thanks for having me. Okay, so give us a bit of the backstory here. I mean, the Tesla Roadster, this was a car that initially was unveiled over a decade ago. You know, we heard a little bit about it, but then we sort of drifted into Model S, Model X, Land, etc.

13:00We stopped hearing about the Roadster. Give us a little bit of the history, and then we'll talk about your reporting shortly. Yeah, so the Roadster has arguably the longest history of any product at Tesla. The very first Roadster was unveiled in the 2000s, and it was Tesla's first car. It was also, So, you know, the very first vehicle was delivered to Elon Musk. And then in 2017, Elon Musk surprised everyone at Tesla by unveiling, you know, all the Tesla fans by unveiling the Roadster, you know, a revamped version of this Roadster in 2017. At the time, they started, you know, accepting reservations and saying that the car would be available for delivery by 2020.

13:372020 came and went and the car has been delayed at least eight times since then. So it's kind of Tesla's longest running, you know, delayed product. And we should say, I mean, the Roadster division here was, this is sort of the sports car version of the Tesla, which was supposed to be the higher end, you know, Lamborghini type of vehicle, right? Yeah, it's a very niche product. It's supposed to be like their top of the line performance car, exactly designed to compete with Lamborghini, McLaren. How many did they end up selling ever in that first generation of the Roadster? The first generation of the Roadster, they sold a few thousand units and it actually helped put them on the map.

14:19It led up to their IPO. It was a major product for them. This revamped Roadster has kind of been less of a priority because they've since moved into making mainstream cars for the average user, much cheaper vehicles like the Model 3 and the Model Y. So the Roadster is much more niche, you know, and isn't going to sell as many units because it's expected to be hundreds of thousands, if not for this, you know, specialized version that they're making millions of dollars. And so let's talk about the specialized version. So you have some exclusive reporting on what Tesla and Elon are working on with this specialized new vehicle.

14:56What have you found? Yeah, I found that this project has gone through a lot of different changes over time. So there's the 2017 vehicle we saw, you know, which seemed like a traditional sports car. Musk said at the time it's going to be the fastest production car ever made. Since then, the design has completely evolved. It's gone through a few different stages. At one point, they tried to stretch out this and make this a true four-seater on top of the Model S Plaid. I think they decided that looked a little goofy. And then they went into trying to make some designs inspired by the Lamborghini Countach, which is this very angular vehicle.

15:33And then at one point, they were even exploring designs similar to an SR-71 Blackbird, which is this spy plane from the Cold War era. It's very unique looking. Now I'm told that the design is a little bit more conservative than that, but very different from the 2017 car we saw. But OK, so we should. But let's not bury the lead here. So, I mean, it's they're trying to make it fly. Like what is that? Is that what you mean by spy plane? So Elon Musk has been saying that this car is going to fly for a while. And he said pretty openly that it's part of a project with SpaceX using these cold air thrusters, you know, that are going to send it into the air at Tesla.

16:13I think their immediate reaction to that was like, what are we doing here? What is this going to look like? um it's obviously a very unusual car um in that idea what is what does fly really mean in this car like like i mean what should we expect here like are we literally are we talking about a runway are we talking about i mean you said something about like hovering in the story like what does flying mean in this context yeah that was a question i had for a lot of people and it sounds like the idea of what it means to fly evolved over time so i think initially they were thinking like almost like an aircraft car, you know, very similar to this spy plane.

16:51And then I think, you know, those ideas became a little more conservative over time. And the idea now is it's more hovering in the air. So as recently as like earlier this year, when they were practicing for this elaborate stunt to unveil, you know, the specialized version of the Roadster, they had it driving down a ramp, going upside down, you know, detaching from this ramp and hovering upside down, on writing itself, you know, and then eventually coming back down. You know, it wasn't necessarily going to fly, you know, miles and miles. Like, sorry, so am I... I don't know if I'm understanding it.

17:24It goes like this, and then it kind of like hovers like that? Explain to me it's kind of like a roller coaster structure. It's going to go on a ramp, and then it's eventually going to be... And then it stays upside, maybe it flips around. Yeah, it flips upside down. This ramp is magnetized, so the magnets release. It's upside down. It writes itself in the air. I mean, this was as of earlier this year. It sounds like some of the stunt plans might have evolved since then because they've showed Elon Musk the demo a few times, and he's obviously had some notes. So, okay, let's talk about, I mean, the reality of the situation on the ground.

18:01So you had reporting that they are working towards this, maybe an event of sorts where they could unveil this. When is that event expected? And ultimately, what does that mean for, you know, the actual product? I mean, I went on the website today, you can still reserve the Roadster. It's$5 ,000. And then within 10 days,$45 ,000 in a wire transfer to secure your spot. You know, I think people have been waiting here, you said in your stories for almost a decade now, is this thing ever going to get sold? I mean, that's a great question. So this stunt was supposed to happen you know at the beginning of the year elon musk said it was going to happen in april now you know i'm hearing it could happen as soon as this month but you know i wouldn't hold my breath if it got delayed just because it's such an elaborate stunt and this product has kind of proven not to be a high priority for tesla okay and you also said in the in the story i mean this vehicle that they've designed as per now i mean it's it's not a street legal vehicle i mean this This is very much, I mean, what, it's going to be collectible, I guess?

19:08Yeah. So I think the idea is there's two versions of the Roadster. So some people put down$50 ,000 deposits. That would be for a Roadster that, you know, wouldn't fly. Some people very early on put$250 ,000 deposits down for a specialized version. And the idea is this would be the specialized version. And there has been some internal discussion at Tesla about having this be a track-only car, that you could use it a track, you keep it a track, because it's not street legal. And right now, you can't even ride in it because the cold air thrusters have so much noise and vibration that it could burst your eardrums if you were too close.

19:45I hate it when that happens. I hate when I get in the car and the cold air thrusters are just so dang loud. Although I will say, here in New York City, I'm pretty sure they have cold air thrusters on some of those vehicles out there because they are pretty loud already sometimes. times. Okay, let's talk a little bit about the strategy here. So Tesla, SpaceX, we've written extensively about the collaboration between the two companies. Does this hint anything to us in terms of a possible merger between the two? Yeah, I mean, I think it's just another example of them collaborating together. There were rumors at one point at Tesla, you know, like, what if this could be the perfect event to say that we are going to have a merger?

20:27But obviously, that's just speculation. And, you know, this is another one of many examples of them working together. Okay. And very quickly, Grace, before I let you go, I mean, speaking of SpaceX, so the cursor deal has now closed. So cursor is officially part of SpaceX. And just to pivot here slightly, what are the big reporting questions that you're trying to find answers to about this acquisition? Because this is one that we've been watching for a long time as well. Yeah, I think the biggest question is kind of how will, you know, will Cursor be able to allow XAI to catch up to all its competitors because they've been lagging behind?

21:03You know, what will that look like in practice and, you know, how will that play out? I'm also curious about the leadership structure. How will, you know, Michael and Amon and the leaders at Cursor fit into XAI now and what will that look like going forward? And I think, you know, I'll definitely be keeping an eye out to see how is this going to impact the product roadmap going forward. Great. Well, Grace, I want to thank you for coming on. It was a great story. I encourage everyone to read it. That is Grace K., our Elon Musk reporter, and for the day, our aviation correspondent here at The Information.

21:37To close out the show, I want to unpack a debate that has been brewing over X this weekend. A string of posts from the All In podcast and technology investor Gavin Baker prompted Anthropic CEO to publish a long post on his thoughts around regulation. To break that down, I want to bring on Leo Schwartz, our tech and politics reporter. Leo, welcome back to the show. Happy Monday. Thanks for having me. Always fun to talk about subtweets. And the meta debates. We were talking this morning about the debate and the meta debate that is going on in AI right now. I want to get to all of it. So look, what prompted this post from Dario over the weekend?

22:17So Dario, not a big, I guess I can't even say tweeter anymore, subposter, poster on X. But he got prompted by this appearance by Gavin Baker on All In and this tweet from or post from Gavin Baker where he basically said that he'd heard from multiple reliable sources that Anthropic and Dario have said that Anthropic is going to be the only company in this new AI future. And this just goes with this broader fear that you're seeing across tech that Anthropic not only is building the best models, but is trying to do this form of regulatory capture in such a way that there will be no other winners, that they'll make this future where they're the only company left either by their own technical capability or the lobbying they're doing in D.C.

23:04in such a way that they create such constraints that no other companies can even compete. Dario, of course, did not agree with that, this idea that they're trying to do regulatory capture. And you saw this long post, although, interestingly, he did not dispute the idea that he has said at some point that it would probably be the only company left. Well, founders say a lot of things when they're fundraising. Okay. So, I mean, you know, I don't know. Is it that big a deal? It's not the greatest thing to say, but he's got to raise a ton of money. So whatever he needs to do to get the bag, as they say.

23:39You know, so the one of the things that stood out to me about this post was this whole idea that people equate regulation with regulatory capture with centralization of power. And I guess this is the idea that when there are regulations, companies with the deepest pockets, you know, they have the most lobbyists in D.C. They can sort of bend the rule or not bend the rule. They can shape the rules to benefit them the most. I guess as you've been covering D.C., is this generally the case? And is there evidence that this is the way that the A.I. story is going from what you've seen? And I think the debate around this AI framework that we talk about ad nauseum is a good lens into this question, where basically if you only have three companies really at this point that are building at the frontier level, OpenAI, Anthropic, and Google, and then whatever regulatory body or agency or branch is creating regulation to deal with the frontier, then it's going to be written with those three companies in mind.

24:43And I think the fear is if basically Anthropic is defining what Frontier is and the system that's created is written with Anthropic in mind, then it might be impossible for any other company to even build a Frontier model. Or because Anthropic's already at the Frontier, if another company has to start going through these regulatory hurdles, then they'll never actually catch up. What Dario said in response is that's not true. Not all regulation is this idea of regulatory capture where the leading companies are the ones setting the rules. He said that everything they've lobbied for has been trying to make sure that other companies can catch up and that the regulation is only really restricting the ones that are at the leading edge.

25:24In other words, they're the ones who have to go through these extra hurdles. And the smaller companies, the ones building open weights, will have less constraints as they're trying to catch up. This is still an open question. I think David Sachs, obviously, who used to be the lead AI policy person, the Trump administration, had an interesting retort where he said, essentially, no, it's clear that in history, regulation isn't only done to protect the public. It's also done because the companies with the most power, with the most money, with the most lobbyists have undue influence in shaping how supervisory programs by the government are actually put into place.

26:04So I don't know. It is an interesting debate. And I mean, I guess on the flip side of this idea where regulations are supposed to help the smaller companies grow and, you know, maybe equalize the playing field a little bit. There's also the flip side of this argument, which is that no, any regulation for American companies, period, will make it easier for companies from China, for example, to grow without regulation. and that inherently puts America at a disadvantage. I mean, this is kind of the other side of the argument that a lot of people have been talking about, right? Yeah, and what David Sachs said was funny because he said, I'm not against all regulation.

26:44Obviously, he champions this hands-off approach. He said, I am looking at antitrust as one potential area and maybe that would be good for anthropics. So this idea that one company shouldn't get too powerful. But no, that's completely right. I mean, if you have regulation in the US and it only affects U.S. companies, but you have a lot of U.S. companies using Chinese open-weight models that won't be touched by U.S. regulation, then that obviously disadvantages all these other U.S. companies that are trying to build here. Let me ask you one, you know, slightly different question. So we talk about the pushback against AI labs here, and Anthropic is at the center of it.

27:23Anthropic is very much in the lead right now, so they have become sort of the punching bag, I guess, if you will, for the industry. I mean, we just finished talking to Phoebe about NVIDIA and all the ways that its tentacles have stretched into the AI ecosystem. NVIDIA's product is the undisputed leader. I mean, there's no race here at all, right? And I wonder if people in DC that you talked to, people in that orbit, do they have any of the same criticisms of the big players in the chip sector of NVIDIA? because it feels like, I mean, NVIDIA, people are kind of throwing their hands up in the air saying, yeah, they are the clear winner here, but there's no solution to it.

28:05But also maybe there's not anything that can be done. I don't know. Well, I think it depends on what you're talking about the winner in, right? Like most of the AI policy debate the past few months has been about cyber risks that people realized after the release of Mythos. So it's very much been about frontier models and the potential of what models can do and what happens if you have these incredibly powerful models in the hands of just a few private companies and right now that is really just open ai and anthropic and maybe google so that's what this debate is around i think nvidia as you said is the leader in chips they have the best product and chips they're incredibly important on this infrastructure level and i do think you've seen this criticism when it comes to chip exports this idea that maybe nvidia is really determining how chip exports are being handled under Trump, this idea that they're really trying to open up the sale of their chips to China, because obviously that's profitable for them and maybe very important to their bottom line.