OpenAI's $115B Cash Burn, Nuclear Energy for AI & Trump's Tech Dinner | Sep 8, 2025

0:14Welcome, everyone, to The Information's TI TV. My name is Akash Pasricha. It is Monday, September 8th, and we have got a fun, fun show planned for you today. We are talking to the founder of a fast-growing nuclear energy company about the future of that sector. We're then going to dive deep into anthropic settlement from late last week that is making waves in the land of AI copyright issues. We're also going to bring on our Washington, D.C. correspondent to help us figure out what the president's meeting on AI with the world's biggest technology CEO's means going forward. But I want to start with a big story that the information published on Friday about the latest on OpenAI's finances.

0:55If you thought the pace of OpenAI's revenue growth was staggering, the information is first to report that its top line is projected to grow even faster, but so too will its cash burn. My colleague Shree Mupiti reported that story, and I want to bring her on to talk all about it. Sri, it's great to have you. Welcome back. Gosh, this was a big story. Tell us about the numbers that you found. Of course. It really is a mind-blowing story. But what we found is that OpenAI had good and bad news for investors. They projected higher revenue growth by the end of the decade for 2030, where they projected roughly to end at$200 billion in revenue as compared to$174 billion from its Q1 projections that we had previously written about.

1:43But also, they had much higher costs than expected. So their cash burn from this year to 2029 is roughly$115 billion. That's$80 billion higher than previously expected. And that's total cash. So that's through the end of the decade? Till the end of 2029. 2029. Okay. Okay. Okay. And so this is considerably higher than what you had found they initially projected a couple quarters ago, right? Exactly. It's$80 billion higher from when I had previously written about it, about its Q1 projections. Okay. And so where are all these extra costs coming from? Is it compute? What is it? Yeah, exactly. So burn roughly this year is$8 billion.

2:28That's about$1.5 billion more. and for Neftshira, it's$17 billion. That's about$10 billion more. And so a lot of these costs are actually coming from compute as well as other data-related costs. And so the big uptick is from R &D and these data-related costs, which they talk about to investors as well. Now, there's a great chart in the story, and we're going to link the story in the show notes, and the numbers go much more into detail. But with the chart, I see that it does eventually plan to get to cash flow generation. What is his plan to get there by 2029 or 2030 even? Exactly. They plan to get to cash flow positive by 2030.

3:10And so despite some of their uptick in costs, they expect revenue to continue growing. And they see that largest revenue changes from ChatGPT, where they expect it to go faster than previously projected. So in their projections to some of their investors, they told them that they would expect about$70 billion in more revenue over the next six years compared to its prior projections. And you see that with the pace at which ChatGPT is growing. So, for example, they're at 500 million weekly active users across its products as of March, I believe. And now they're at 700 million weekly active users plus.

3:48And so when you see fast revenue growth, I think they've adjusted their expectations for ChatGPT. But it does mean that their expectation for profitability goes to 2030, then 2029, which they had previously expected. And so all the money that we've heard about them raising, is that going to cover all of this? Or do you expect they'll have to raise even more money? How are they going to finance all this? I expect them to continue raising money. But one of the concerns is around how to think through dilution for existing investors as well as for the company itself, like the founders, employees, things like that.

4:28But I think also related to their compute-related costs and data-related costs, there are other types of capital that they can raise. So, for example, debt and also the fact that they are continuing to build Stargate, which is their own facility. There are maybe other ways to think through financing that. And so I think there will have to be other ways that OpenAI will continue to raise, but it's obviously yet to be seen. All right. So let's just take a step back here. I mean, these are projections, right? I mean, projections change. The numbers are what they are. The numbers are huge. One of the things I wanted to ask you is just big picture.

5:06I mean, the fact that revenue projections have increased, the fact that expenditure, you know, the cost side has also increased. What does this tell you about Sam Altman's sort of calculus saying, hey, we're willing to burn all this money? Or what does it tell you about OpenAI's strategy and saying, look, you know, we don't care. We're going to keep raising. We're going to keep burning. I mean, like, help me get inside their head a little bit as to what they're thinking. Of course. I think Sam Altman has said before that they expect this to be one of the most capital intensive companies. And you see that with the amount that they're raising as well as the amount that they're spending.

5:45But I think that what it shows me just giving and tracking the company closely across both its revenue projections as well as like its cost projections is the fact that this is a unique company. And Sam and the company will do very unique things to make it happen. The fact that they're, for example, have talked about infrastructure as a service in a way where they liken it to AWS, where they can maybe sell different types of compute or that really shows that their ambitions are beyond just like the chat GPT product. And they can actually diversify. I mean, this is one of the things you wrote in the story is that they are really looking to diversify revenue, although it's a little bit cloudy as to, you know, because one of the things you said in the story was that revenue from API might be a little bit less, might be baked into the ChatGP team.

6:36And we don't actually know how that's going to be diversified, but it strikes me that they're going to have to get beyond just the chatbot sort of thing. Of course. And I think that you say that with their projections too, where they talk about new products. And we've written about this in the past where it could include things like free user monetization, which we've interpreted to be, for example, affiliate fees or maybe even ads, which both the CEO, Sam Altman, and CFO, Sarah Fryer, have talked about. And so this might look like, for example,$2 to$15 as an average revenue per user and have gross margins similar to Meta's in terms of 80 % to 85%.

7:14And so those are other avenues in which OpenAI could also think through monetizing more. And so they generate about$1.10 billion from these products that they've told investors from now to the end of the decade. Right. Great. Well, Sri, thank you so much for coming on. Anytime there are numbers to do with OpenAI, I'm racing to go click the story because the numbers, they change so quickly. And as we've seen, they just get bigger and bigger. So we'll be watching your byline for more updates on that. That is Sri Mupiti, our OpenAI and Anthropic reporter from The Information. Okay, well, we've all heard about the power that AI is going to consume over the years.

7:56It is going to be a big expense for AI giants. And one of the emerging technologies that big tech firms have increasingly been looking at is nuclear energy. Now, there are tons of companies in the space. One by one, we're going to bring them on to talk about what they're building. But this morning, I want to bring on Matt Lozak, the CEO of Alo Atomics, a company that just raised another$100 million in a funding round led by Valor Equity Partners. Matt, it is great to have you. Welcome to TITB. Thanks so much for having me. Great to be here. So, look, I just want to take a step back here before we get into Alo's business.

8:31I mean, talk to me about where we are at in the story of nuclear, just broadly speaking. Yeah, so I think it's a record good time for nuclear right now. There's this kind of record levels of public support, investor support, government support. And like you pointed out, arguably, most importantly, support from the right type of customer. So that is AI data centers. They present an immense amount of demand. We're talking about 40 gigawatts in the U.S. alone in the next five years, 100 gigawatts globally. And that is just an unprecedented amount of power. So that's like creating 40 new cities in a time span of five years.

9:17And the last time that happened was basically never. So it's a really great time for nuclear. And we are very excited to help push this way forward. So one of the things about nuclear is that nuclear takes a long time to build, whereas one of the biggest consumers of this power, AI companies, that's happening right now. So talk to me about where ALO Atomics fits into that story and how you plan to satisfy that demand that it's not going to be five years away. It's literally happening right now. Yeah, exactly. So that's a big part of why we exist. So right now it takes around seven years in the best case to build a gigawatt scale nuclear plant, maybe 10 years or even longer for the most recent ones.

10:02And the reality is it takes around one year to build these data center buildings that house all these GPUs. So really what you have to do is you have to have a factory that can mass manufacture slightly smaller nuclear reactors and power plants and mass produce them at a faster clip. And so that's what we're doing. It's an advanced reactor with a lot of great inherent safety characteristics and equally importantly, something that can be deployed at the same pace as these hyperscalers. So ALO's pitch is that you can build these plants faster than they ever have been able to build. Is that the idea?

10:40Yeah, that's a big part of it. But I mean, you know, I really think it's worth highlighting nuclear is kind of like this holy grail energy source. So for advanced nuclear, it doesn't need water, doesn't need a magma pocket, can be located anywhere. It uses 100 times less land than renewables like solar or wind. And right now you see the two biggest data center projects, the XAI Memphis project and the Abilene OpenAI project is basically, you know, they cited at those locations because there was a lot of natural gas there. But the problem is for one or two gigawatts, yes, there's existing pipeline for 40 gigawatts.

11:21That pipeline simply does not exist. So if you have to do new fracking, new pipeline expansion, et cetera, you might as well go with the ultimately best solution, which is nuclear because it's clean, baseload, works anywhere, and when mass manufactured and done right, can be very quick to deploy. And that was one of the things I wanted to ask you about was, a lot of the time when these nuclear companies were found, I think you guys were found fairly recently, 2023, if I'm not mistaken, right? That's right. So you were founded two years ago. Even if it was three, four, five years ago, I mean, the scale of data center that nuclear companies were building for was much smaller, I think, before the advent of the most recent AI boom, at least.

12:03Now we've got these data centers that are huge and that they're bigger than I think a lot of nuclear companies might have fathomed. So how do you see a company like yours sort of scaling up to meet that size of demand? Well, we've made a lot of decisions that really are well-suited to help meet this scale. And I'll give you maybe a few examples. One big one is our fuel supply chain. So a lot of nuclear companies chose to use HALU, which is higher enrichment fuel, or TRISO, which is a fuel that's really kind of well-supported or kind of well-suited for remote diesel applications. But the problem with those fuel supply chains is they're basically non-existent.

12:49And so if we stand any chance at meeting the demand that these hyperscalers present, we decided to go with off-the-shelf uranium dioxide, which is used to power the entire water-based fleet around the country today. So we're essentially doing an advanced reactor using traditional fuel for supply chain reasons. And the advanced reactor component essentially gives it a little bit better kind of inherent safety, even though nuclear is already very, very safe. And it also helps with mass manufacturability. So those are the decisions we made to make sure that we basically did what a startup is supposed to do, which is go talk to customers, ask what they want, and design around that.

13:31And how does that then address the issue that many nuclear companies have had, which is these facilities inevitably going over budget? We wrote this great story. I'll link it in the show notes. It was one of our weekend big reads on the nuclear sector. and the editor, Nick Wingfield, who took the role of the reporter writing the story, he wrote about how these projects have really gone over budget. How do you aim to address that issue? Well, first of all, our first plant is going to be certainly more expensive than our eventual plants, or end of a kind, as they're called. And that's kind of to be expected.

14:09So if you're mass manufacturing something, you're going to come down a cost curve, if you're going to have learning and your first plant is going to be more expensive. The good news is this customer has a kind of higher willingness to pay as long as you can get things done quickly and efficiently. And so - This customer, meaning the people that, I mean, you're not talking about any customer in particular, you're talking about big tech companies. Yeah, this customer profile, right? So the hyperscalers have kind of shown that they're willing to pay 10 to 20 cents per kilowatt hour or baseload power that can be delivered very rapidly.

14:46Basically, for them, it's existential because if they don't move quickly, they'll lose to another hyperscaler. And so speed is the name of the game there. And they're willing to pay this premium for that speed. And that allows us to move quickly, start a little bit higher in the cost curve. And then, of course, as you master something, you can bring it down at cost. So you're saying that we're not concerned that it's going to cost us more than we might think, maybe to build one of these plants. We're still going to be able to make enough money, you know, on the revenue side from these companies.

15:19That's not a concern for you in the short term. In the long term, you're saying you're going to try to bring that cost down. Yeah, I mean, we have a pretty good sense of how much this will cost because we're actually building things. So most of our cost estimation is based on real quotes and estimates and real data from the things we've built so far. And so, you know, we're just making sure that the costs that we're forecasting and that we're estimating are in line with what the customer is willing to pay. But that also making sure that we have a very clear path to cost reduction over time. Right.

15:51Last question for you. I want to get a picture of the way you think about regulations right now for nuclear. We've seen the Trump administration has been largely supportive of this technology. This is, as we've talked about, it's not just a four-year journey. It's going to be a 5, 10, 15, 20-year journey. How do you think about the direction in which regulations are going and how do you think about if the next administration might be as supportive or not about this technology? Well, yeah, fortunately, nuclear has pretty solid bipartisan support. So Biden was very pro-nuclear. Trump has been very pro-nuclear.

16:30And what's really exciting is the recent executive orders from Trump have really served to kind of streamline the pathway for new nuclear to get built. So these executive orders, one, for example, has pushed towards achieving criticality under DOE authorization by July 4th of next year, which would be a very nice birthday present for the nation. So we have a bunch of companies pushing after that. And we actually have a very uniquely qualified team to go that route because we have the only team in the country that's ever previously achieved a DOE approval for construction for a nuclear reactor.

17:09and so we feel very kind of confident in our ability to meet that timeline and we know how to do this we've done it before and yeah we're very excited because this could be potentially the world's first co-located and co-built nuclear plant and data center the AllOx our first power plant that we're on pace to go live with next summer so we're very excited and yeah we can't wait to show you the world we can accomplish there. Great. Well, Matt, thanks so much for coming on the show. I appreciate it. Next time you come on, I think we look forward to having a bit of a longer discussion about how this technology is evolving and also the different stakeholders that are at play.

17:51But that is for the next time we have you. Matt Lozak, that is the CEO of ALO Atomics. Okay. Well, Anthropic last week agreed to pay$1.5 billion in a settlement with a group of authors that alleged the AI company trained its models on 500 ,000 books without permission. If you're doing the math,$1.5 billion divided by 500 ,000 books comes out to about$3 ,000 per book. Whether that's a lot or a little is up for debate, but we have heard that many, many lawsuits like these from content companies are still going on. We've talked about it a lot in the show, and so I want to bring on someone to tell us more about her view on how these issues could play out.

18:32Cecilia Zaniti is the CEO of GCAI. She is coming to us. It's her first time on TI TV. Cecilia, it's great to have you. Welcome to the show. Thank you. Big fan of the information. Thanks for hosting. So this settlement came out late last week. You've had the weekend to reflect on it now. What were some of your weekend reflections now looking back on it? I would say the settlement is a waypoint on our journey around what does it mean to train data and what does it mean for copyright authors. And the most important, I guess, takeaway is that Anthropic elected to pay for infringement that at least was ruled to possibly be there.

19:17And in terms of timing, it feels like a Napster moment in the sense that we're proceeding to more of a licensing scheme for training data. And so you think this is going to be the norm, these licensing deals are going to become more commonplace? I do. I think a market-based solution is where this was always going to go. And the reason I think that is all the history, right? So originally we had the biggest fair use case ever was the Betamax case, literally over VCRs. And at the time, there was a question if copyright owners could make money on VCR tapes, then of course we know they did. And it happens kind of every time there's a new tech, how does the market adjust and what do the economics become?

20:01So let me ask you this. If you look at just the deals between the news publishers, for example, and the AI companies as one category of licensing deals, lawsuits that are sort of in play right now, who do you think has the bargaining power in these arrangements? Mm-hmm. Where is society going to put that balance and who gets paid? In the case of the news publishers, some have adjusted very well. So New York Times found other revenue streams. They've got the New York Times food app, Wordle, that kind of thing, in addition to just the print journalism. Outfits like, of course, the information, do the high-quality journalism behind a paywall.

21:09But how those business models evolve and how, you know, American journalism or just copyright in general moves forward, I do think the balance of power is quite shifted to the AI companies. And do you think that's a good thing? I'm a tech maximalist, tech optimist about AI. I run an AI company. you know, there was Mark Andreessen's, you know, kind of time to build, you know, uber techno-optimist view. I think in the long run, it will, AI will be like the internet and have incredible advantage. But I think creatives have to respond, right? And you see that, right? You see things like CAA is creating essentially digital twins of talent to be able to license.

21:56I believe we will see licensing schemes emerge. And that's also why this is exciting, right? You can imagine a system where authors get paid when their things get used for training data. It's not that hard to imagine, particularly if the appetite for this training data is so high and the AI companies have so much money. So I think a market-based solution will emerge. Do you think it remains with the flat fee deals that a lot of these licensing deals have taken a shape of right now? Or do you think it evolves into a usage-based model? I think that we're going to see more of a usage-based. So when I think about, for example, you could look at a model of online advertising, right?

22:37So Google obviously has figured out micropayments. Every time we click on a sponsored search, there is a payment. Right. So as a technical matter, I think it's possible. I think it's a red herring to say it's incalculable. But who becomes that iTunes? What is the system? Is it something like an ASCAP, which is the music licensing kind of regime? that I don't think a leader has emerged yet in that. Great. And last question for you. I mean, we see the licensing deals. We also see the settlements. One way that I've been thinking about this is, hey, do we see the AI companies just use the data and then settle?

23:14And do those settlements sort of happen at a higher rate or a faster rate than news companies can strike these licensing deals? I mean, I guess my question is, do you think the settlements will become more commonplace in the short term than the licensing deals and then licensing deals have to catch up? Is that how this is going to play out? So, you know, it's always a question as for permission or as for forgiveness. But I think that the model companies, just like Anthropic, in this case, they're staffed up with lawyers. Each of Anthropic and OpenAI have big legal teams now spending on the very best.

23:51But thinking about it from the standpoint of like planning in advance and getting creative with licensing deals, getting creative around what user data they end up taking in. We saw Anthropic last week change their privacy policies where they will default train on consumer information. Right. So I think that the companies will get more creative. I do think we're going to see huge foundation model companies have the stomach and the money to both litigate and to create big compliance frameworks. So my expectation is that we'll see fewer kind of small model company challengers. And the big ones will definitely have legal in mind, employ people like me to help them kind of do it in a copyright effective manner.

24:36Great. Well, Cecilia, thank you so much for coming on the show. Next time we have you on, we're going to hear more about what you're building at GCAI because I think it's a fascinating business and the intersection between law and AI and copyright and tech. I mean, there's a lot to get to. So thank you so much for coming on. That is Cecilia Zanini, the CEO of GCAI. Well, late last week, President Trump hosted a dinner at the White House for a who's who of the tech elite. More than 30 top tech executives, including Mark Zuckerberg, Tim Cook, Bill Gates, leaders from Palantir, AMD, and Figma. Oh my gosh, they all gathered together to talk about policy and technology.

25:11I want to bring on Sylvia Varnamoregan, who reports on all things Washington, D.C., and technology for the information to help us understand where this story goes from here. Sylvia, it's great to have you. Welcome back to the show. Hey, Akash. Great to be here. Thanks for having me. So, look, I do want to talk about the broader meaning of this meeting, but just give us the 20-second recap. Why exactly did this meeting happen in the first place? Yeah, well, it was a pretty stunning scene, wasn't it? Having all of these big CEOs coming together for a dinner at the White House. The White House billed it as an opportunity for the group to sort of discuss how America can get ahead in the AI race.

25:52And it also came after an event was held earlier in the day hosted by First Lady Melania Trump that focused around AI and education. And so they talked about AI and policy and sort of, I mean, what was even the discussion topic? At the dinner itself, as we saw, the CEOs went around and really praised Trump, really flattered him, pumped him up, talked to his leadership. I saw these compilation videos of them all saying, hey, thanks for that. Thanks for this. We like that. Right. Yeah. And it's actually really interesting to remember that some of these people, including Sam Altman, including Mark Zuckerberg, had previously been quite critical of Donald Trump.

26:35Right. So I think what this meeting really underscored is that those days are over and these companies and these individuals are really getting behind the administration as they seek to sort of build out these huge AI businesses. But in terms of what was actually discussed at the dinner, I'm told that President Trump really wanted to understand the obstacles that these companies view as being in their way as they try to sort of win this global AI race. they discussed China they discussed the enormous energy and power demands of AI European regulation and really just canvassed what what roadblocks these companies saw as having and how to sort of smooth the path for them to build out these businesses I thought it was funny I mean the analysis I was running just seeing the photos was like who's sitting where you know who got the seat.

27:33Zuckerberg was sitting right next to Trump. You had the people at the end of the table. I don't know if they felt like, oh, I was the last person drafted to the team or something. I don't know. Anyway, we could put that analysis aside. What I really want to ask you about, Sylvia, is what you think both sides got in this meeting. On one hand, there's what Trump got, and on the other side is what big tech got or what these CEOs got. Let's start with the Trump side of the equation. I mean, what do you think he was getting out of this meeting? Well, I think for Trump, firstly, he had the optics of having these huge names in business gathered around him, flattering him, as I said, praising his leadership and demonstrating loyalty to his administration, which, of course, we know plays well with Trump.