SpaceX's $23B Debt, DeepSeek's $20B Valuation, and OpenClaw's Growing Pains

0:12Welcome, everyone, to The Information's TITB. My name is Akash Pasricha. It is Wednesday, April 22nd. First up today, The Information has exclusive reporting that Tencent and Alibaba are in talks to invest in DeepSeek. Our Asia Bureau chief will join the show to talk about that. We'll then talk through our story about OpenClaw's growing pains. We also have some analysis on the SpaceX cursor deal. We have some exclusive reporting on SpaceX's debt profile. We'll then be joined by a cybersecurity CEO to talk through recent hacks at Mercor and Vercel. And we will wrap the show with our senior finance editor, who will discuss his latest column around the opportunities and risks for CoreWeave.

0:56It's going to be a fun show, so let's get right on into it. The Information published exclusive reporting about DeepSeek's funding round that is coming together. Our Asia Bureau Chief Jing Yang joins me now to break it all down. Jing, welcome back to the show. It's great to have you here. What do we know about this funding round? Yeah, so we reported just on Friday that DeepSeek is opening up to external capital for the first time ever, which is a big shift for this AI lab. Up until then, Dipsy has been running pretty much as open source AI lab, free from commercial pressure, free from external investor pressure.

1:32And remember, when we reported on Friday, the valuation that Dipsy was seeking is$10 billion plus. And quickly, that, you know, maybe potentially the effect our story has had, the company has met with immense interest from potential investors they've held discussions with. So now they are seeking a$20 billion plus valuation. That's how fast the story is moving and how quickly things are evolving. And then, so talk about Tencent and Alibaba. So they are interested in investing? Very much so, based on our reporting. Okay. So this is kind of interesting because I'm putting this into context of what we've seen here in North America.

2:16We've seen the big hyperscalers like Microsoft, Google, Amazon. They've invested in these high-flying AI startups like Anthropic OpenAI. You can't really call them startups anymore. I mean, these are hundreds of billions of dollars. But is this the same thing playing out in Asia? Is there a circularity loop with respect to business that DeepSeek is buying from Tencent and Alibaba? Walk us through it. So I would just say the short answer is yes and no. So I'll go with the no part first. So DeepSeek, as you know, is mostly open source AI developer. That means that any cloud provider can apply for a license to host the models on their cloud services.

3:01From that perspective, I would say that the logic for the hyperscalers in the US to invest in the likes of Anthropik is a little bit different in here. However, the other end of the spectrum is that we know that Alibaba, Tencent and Biden have all been developing racing to develop the best models in both China and on par with the US. And DeepSeek, even though they have not published or launched a new model for over a year now, still is considered as quite an asset and quite competitive technology-wise. So for the tech giants in China, investing in a company like DeepSeek has two benefits. First, that, yeah, of course, for these companies that have their own cloud computing, investing in Upstart can potentially help them acquire a customer.

3:59We've seen this played out previously, actually. Alibaba, being the cloud computing leader in China and Asia, has invested several Chinese AI Upstarts. and then they would, this like sort of a very easy way for them to acquire customers because you can invest in the startup, which will naturally lease compute from you. And then you use the money to gain a very loyal and sustainable customer. That's the first logic for these companies to want to invest. The second one would be, you know, DeepSeq still is very popular. And then if you believe that DeepSeq will be able to continue to innovate, then being an investor would then help you to deepen your potential collaboration down the line with Deepsea whenever they launches their new model or even more AI-related products.

4:51And these companies need the best technology they can get to continue to enhance their user base. Because these are companies that have hundreds of millions or even more than a billion user base of the various tech products and apps in China. So I would say these are the top two logics for these companies. Let's go back to the valuation. So$20 billion is the new revised valuation that we are reporting that DeepSeek is seeking. $20 billion plus. Okay. All right. So it could go even higher. We've seen it already double, I guess. So, I mean, sky's the limit in terms of DeepSeek. But tell me,$20 billion, where does that stack up in terms of competitors in China?

5:43Does that feel high, low? Just put it into context for us. Yeah. Yeah, so I understand 20 billion compared to the valuation that OpenAI and Fobb has been able to fetch feels like so minuscule, right? But you have to bear in mind that first, DeepSeq is like open source AI lab that has not generated meaningful revenue. And then on top of that, the math that based on reporting, it could go, you know, it basically works like this. The other heavy contender in the ALM space in China that has not gone public is Moonshot AI, which is behind the very popular Kimi series of models. And then they, as we reported, are finalizing their latest funding round, which values the company at$18 billion before the round.

6:34And so from that perspective, it doesn't seem too far-fetched that Deep Sea wants a slightly higher valuation. And then there are other data points that suggest the valuation could even get higher because we have two AI model developers that went public in Hong Kong in January, Minimax and Zipu, also known as Z.ai. And then these two companies went public in January at a valuation under$10 billion, but now they are traded somewhere between$30 to$50 billion based on recent market data. So it really depends on how you want to cut and paste the math and how you want to decide what is your worst. But then I do want to point out for the two companies that have recently gone public, they do have been able to generate revenue.

7:22Right. Well, I think as we've seen, revenue and profitability is not necessarily an indicator of how much these companies fetch. So at this point, anything goes. $20 billion plus, sure. Why not? uh jing i want to thank that week it's gone up to 50 billion i mean i don't know but well then you can come on the show and you can tell us more about if they have any more revenue in the last week or something like that i don't know all right well jing i want to thank you for coming on that is jing yang our asia bureau chief here at the information on ti tv open claw consumed silicon valley when it first came out a few months ago now the open source initiative is going through some growing pains.

8:06That is according to a story out today from my colleagues Juro Osawa and Rocket Drew. I want to bring on Rocket to chat more about it. Rocket, welcome back to the show. It's great to have you here. What is going on in OpenClaw land? That's exactly what I wanted to know going into this story. And what I come away with is the sense that OpenClaw is kind of in an impossible situation here. So if you remember, OpenClaw exploded all of a sudden, but it was just this open source project that Peter Steinberger had been plugging away at for months. And then all of a sudden it went viral and users started complaining that it had security issues.

8:42And Peter said, of course it has security issues. It's just my side project, but here's what I'm going to do. I'll give you my best advice for how to avoid security issues. And then people didn't follow the best advice. And he said, I'm going to add as many new security features as possible. So he's adding in all of these new security features, but of course that makes significant changes to the software so people's systems break and then they complain. They said, ah, you changed too much stuff and my system broke. He's also trying to reinvent, I mean, invent what agents should even be for the first time.

9:12It's a brand new product category. So he's making pretty sweeping changes to the software and it breaks people's setups. So some people are complaining that their setups are breaking. Some people decide to not even update at all. So they're just using out of date versions of the software. He says, okay, here's what I'm going to do. I'm going to build up a team as fast as I can. He recruits this lean team of volunteers to work on OpenClaw all the time. And other people say, well, I don't know. I don't know your team. I don't trust your team. This isn't a venture backed startup. I'm supposed to use your software.

9:43So he's up against all of this. Meanwhile, there's a whole raft of new competitors on the scene. You've got other open source agent projects. You've also got these offshoots of OpenClaw, like NemoClaw and other corporate offerings that have been adapted. And then at the same time, you're also getting cut off from your access to Anthropix models. Meanwhile, Anthropix is happy to take and copy all sorts of features from OpenClaw and incorporate them into Clawed Code and Clawed Cowork. So OpenClaw is in kind of a tough position here. They're hanging on. They're still getting new users, but growth has slown down a bit.

10:21And we wanted to take a look at all of these constraints that it's running up against because it really is in a tough position right now. Okay, so there's a lot there and I want to get into it. But before we get there, so Peter Steinberger, he was hired by OpenAI, right? Is he still working on OpenClaw while he's a full-time employee at OpenAI? That's right. That's right. It's hard to get a really great sense of where his hours are going these days. And he's probably working, you know, maybe the equivalent of two full-time jobs. But But from everything that we know, he's still able to spend quite a bit of his time on OpenClaw.

10:53He is a very active presence. He's still running the team there. The team is maybe 10, a dozen core maintainers who are these volunteers who are spending maybe three to five hours a day working on this project. And then there are also some maintainers that work at companies like Tencent or NVIDIA. And those companies have dedicated some employees to working on the product. but they tend to work on the parts of OpenClaw that touch their systems rather than the core parts of OpenClaw that everyone is using. So let's go back to the developers you talked about, the people who are upset at this impossible situation that Peter is trying to get his head around.

11:32So there's the idea that, hey, look, we have to make it more secure, and that will require updates. But also we acknowledge that you probably liked V1 of the software or whatever. But my question for you is, I mean, you speak to these developers. I mean, software updates are not anything new in software development, right? Like this is something that happens every day of every week. So why is it that these OpenClaw updates are infuriating them so much? Yeah, that's right. I mean, I think a lot of the software updates that people are used to is like when I update the iOS on my phone and all things considered, it goes very smoothly.

12:10It just works out of the box. But Apple, of course, is able to spend an enormous amount of resources on making that possible. OpenClaw, in contrast, is much more under-resourced. And that is kind of the nature of the game for open source, to your point. I mean, people do expect that when they're updating an open source package, a lot of the time, things will break. Now, open source products that are more mature have found some workarounds here, like Linux releases versions that they expect to maintain for the long term. These are called long-term support versions. But nothing like that exists for OpenClaw yet, because they're still building so rapidly and with such a small team.

12:48So I think you're exactly right that in some sense that should be par for the course, but people had unreasonable expectations when OpenClaw went out. It was discussed as though it was a fully polished, mature product from a company. And so people expected it to be a smoother experience than maybe they should have. Okay. And so now let's talk about the other angle you mentioned, competition, which is that we've seen Anthropic is coming out with their own answer to OpenClaw, I guess. And I'm not sure if it's a direct competitor, but I know that NVIDIA, they're also innovating in the space here with NemoClaw.

13:26And we also have our own reporting that Microsoft has been paying attention to all this. So what evidence do we have that these closed source products, maybe if you will, the actual legacy company products, are they competitive with OpenClaw? Is this a threat to them? Are people still using OpenClaw over the alternatives? What do we know? Yeah, I think they are competition in some sense if we think about these as different products. In other sense, it's a win for OpenClaw. I mean, it's OpenClaw influencing the ecosystem. Sometimes these other products are inheriting code directly from OpenClaw.

14:03So it's definitely OpenClaw having an influence on everyone else. I think to some extent, OpenClaw, faces a big battle in trying to get to the point where it's on the same tier as something like Claw Code or Clawed Cowork, because that's coming from such a name brand company. It has the reputation and some of the credibility built in, and Anthropic is able to throw those resources at making it a smoother experience to use. They're able to make it easier to install, easier to set up on your computer, and they're able to add in a bunch of security features that, frankly, some OpenClaw users wouldn't want because they want the freedom.

14:44They want the ability for their OpenClaw agent to take whatever action it wants on their computer, any action autonomously, go on the internet, browse any web page it wants with very little supervision. So in some sense, they're playing in different lanes here, but they also have some advantages over open claw so so if i understand you correctly you're saying that open claw is actually they're not necessarily threatened by this it's actually validation that their product uh should be adopted in other words yeah that's right interesting rocket last question for you what are you watching for with this story as it continues to unfold are there you know should we be watching for more money pouring into the space here, more high-profile developers joining the OpenClaw Foundation as maintainers.

15:35What are you looking for? I think both of those could go a long way to helping OpenClaw. What I'm looking for in terms of the future of OpenClaw specifically and whether it's going to develop into a mature, stable product or exactly that, you know, once they set up the foundation, are they able to raise the money to hire some full-time maintainers who can really take OpenClaw to the next level, that might be the thing that's missing for them right now. They might need to move from this sort of scrappy volunteer organization to something a little more established. And they're in the process of doing exactly that.

16:09So that's sort of the first question for what happens to OpenClaw. But as far as what I'm watching for on a much higher level, I think is what does OpenClaw tell us about the future of open source software, especially in this AI era that we're entering? People have a lot of questions about what the next few years hold for open source software. And there's a theory out there that we're about to enter an open source renaissance, that as AI allows the cost of writing new code to go to zero, that open source software will flourish. Just a thousand open source flowers will bloom. And OpenClaw, in some sense, is the beachhead of that.

16:49OpenClaw is where we should see that first. So I think we are starting to see it. All of the OpenClaw maintainers are using codecs heavily. They're using AI to write the new features that will allow OpenClaw to improve. But, you know, given that they're still facing these growing pains, it makes you wonder, you know, how far can AI really go for allowing a scrappy team like this to develop a product that's, you know, maybe the fastest growing open source product in history. And the sense is there's still a lot of challenges there. Open source has this fundamental asymmetry compared to these closed source proprietary products like you mentioned, which is that Anthropic is free to look at every line of code in OpenClaw and pick and choose the features that they like and adopt them.

17:33But it doesn't go the other way, right? It's difficult to know what's going on with ClaudeCode under the hood. Unless Anthropic leaks the source code to ClaudeCode, which they did. But that's the question here. Does open source stand a fighting chance? And is it going to really take off in the next few years? Okay, so you mentioned Anthropic, and you mentioned Leak, and this is not a leak that we're about to talk about here. But we do know that, according to a Bloomberg report, that Anthropic's mythos model has now gotten into the hands of some unauthorized users, which was the thing that we were worried about.

18:13I mean, you know, they gave access to a couple companies via Project Glasswing. They said it's going to be a controlled release, if a release at all. I wondered what your reaction was to this. I mean, people in your orbit, are they freaking out, scared that, oh my gosh, the cat is out of the bag, or is this not such a big deal? You know, I think for some people I've talked to, this is in line with their expectations. They knew that all of the AI labs have a ways to go in terms of their cybersecurity. So this wasn't a shock. But to your point, there's an irony here that this model that is so cyber capable, so dangerous in terms of cyber attacks that it had to be kept under lock and key was itself accessed in an unauthorized way.

18:57There's certainly an irony there. It seems like what happened is that the model was in an environment that Anthropik sets aside for third-party testing to help them develop their models before they're released. The model was still available there. Someone had access to that environment. They were able to use Mythos and also other unreleased models from Anthropic. And it sounds like they and their friends just used this model, not for cyber attacks, but to make fun websites for themselves and to play around with it. So that's what we know about how that situation went down. But I think it's not too surprising, even if it is ironic.

19:32Yeah, well, the irony here is like actually crazy, because if you think about it, I mean, I don't understand. Anthropic could have just used Mythos to analyze its own code base, its own cybersecurity, right? I mean, the whole thing was that, hey, we've used Mythos to find vulnerabilities in other systems that we couldn't have found. What, you didn't think to use it on your own system to figure out that maybe this was the back door that, you know, this thing could get leaked through? like yeah yeah it's a great point you should definitely point the system back at itself to some extent the problem is is intractable because it's also a question of permissions you do have humans and you do have to let them interact with the model if you're going to have them tinker with it before the model goes out and so maybe those permissions were left open too long maybe the permissions should have been restricted more quickly but uh there that's an aspect of it as well.

20:27Okay, well, Rockin. I would add one other thing, which is that it does kind of undermine their cybersecurity strategy, right? Like you said, their approach with Project Glasswing was, we will choose who are the trusted good guys, and they will have access to it to make their systems more robust. And we'll do that so that the good guys get access before the bad guys. But if their security doesn't meet the bar, then the bad guys are going to get access to the model and be able to use it for nefarious purposes and conduct cyber attacks. But it will only be available to a limited set of good guys, the ones that Anthropic handpicked.

20:59So if that's the trade-off, you might actually come down and say, well, I would have preferred them to just release it publicly and everyone gets access to it at the same time, both the cyber defenders and the cyber attackers. It's not totally clear, but it does undermine the strategy that they were going for. Or maybe there's a mole. Maybe there's a double agent. I mean, permissions, human error, that is the thing that we cannot control for here in this maybe the mole is opus 4.6 that's jealous of all the attention Mythos is getting we should be clear we're just riffing here we're writing a script for our next action movie it's a good name for a double agent I will tell you that much Rocket I want to thank you for coming on that is Rocket Drew our AI robotics and cybersecurity reporter today here at The Information.

21:56SpaceX has quite the deal with Cursor. The Elon Musk-led company could buy Cursor for$60 billion. To break it all down and also share more about their reporting on SpaceX's debt profile, I want to bring on deals reporter Valida Pau and deputy bureau chief of finance Corey Weinberg. Welcome to you both. It's great to have you here. Valida, I want to start with you. Walk us through this Cursor deal. so huge news yesterday about SpaceX agreeing to have an option to buy Cursor for 60 billion later this year or they can pay a breakup fee to Cursor for 10 billion dollars for the work they do together Cursor was just in talks to raise two billion dollars out of what paying 50 billion dollar valuations until this deal with Elon Musk yesterday so um the round is not going to happen anymore but then we still don't know when this acquisition would happen you know either it's before like after the IPO or how far away is um the deal will happen this also like adds a twist to Elon Musk IPO pitch for SpaceX in June it was largely going to be based on like putting data center in space and about its fast growing Starlink business but um now like with cursor in the mix in the formula you know like how Elon Musk will position SpaceX is like remain to be seen but of course like you know cursor generate like billions of revenues from developers and that But it's a market that XAI has struggled to keep up with its Grok products.

23:23Corey, did this surprise you? I mean, this is another deal under the SpaceX umbrella. What does it tell us about XAI's business? Of course, it surprised me. Just in terms of you don't really see an M &A announcement like this, even though it isn't a finalized deal yet. You don't see an M &A announcement like this on the eve of the largest IPO in history in a business line that is, let's just say, adjacent to SpaceX's core, not exactly in the core. So it was shocking. To me, it illustrated just this key point that I keep hearing over and over again from folks that are watching this IPO very closely, which is this is going to be in large part an AI pitch.

24:15which we talk so much about Starlink and the satellite internet that SpaceX has. We talk about its dominance in launching rockets. Elon is going to be making this about AI. Well, and the other thing, Corey, that I've been thinking about is, look, XAI, we have heard rumblings. Well, Elon has come out and said, look, we didn't build XAI right the first time. We are rebuilding it. We've talked a little bit on this show about how coding is one of the most ripe markets for AI companies to actually make money. I mean, it sort of seems to suggest to us, Corey, that XAI's coding endeavors were not going very well at all.

25:00No, I mean, nothing really with XAI has been going all that well. And they have a ton of hardware, a ton of chips. They have these huge data centers. they have the infrastructure in place to be a real force in AI. But you have to do the computer science well. You have to build the models correctly. You have to have the right people in place. And I definitely think that's one read on this deal that I've heard from sources in the past. What's it been? Not, you know, maybe 12 hours, 15 hours since this deal was announced. is that you really have a situation where you had XAI needing to sort of find a great reason and a great business partner for all the compute that they have.

25:56And Cursor needs compute. It's building its own coding model. It's trying to wean itself off of OpenAI and Anthropic, which poses a ton of strategic risk for them. and XAI also needs a nice story to go with the SpaceX IPO. Okay, so Valida, I want to move to a story that you and Corey broke yesterday. You had some details about SpaceX's debt profile ahead of the IPO. What did we learn there? So new numbers from SpaceX IPO documents is showing us the impact of how Elon Musk is merging its companies together to its debt. So SpaceX increased its debt low by nearly two-third last year to about$23 billion at the end of 2025, and that's up from$14 billion the year prior.

26:47And much of the debt was tied to a lease deal they struck with, like, Valor Equity, that, you know, to rent out, like, AI chips for XAI. And SpaceX recorded that as, like, a$4.5 billion of debt on its balance sheets. and valida do we have any idea how investors will feel about this debt well i thought it's pretty um they are expecting that like you know to build these ai businesses like you need to take on like you know lots of capital investments and like debt to fund these so i feel like that won't affect his like credit ratings um most likely but then they also share with investors like you know in recent weeks that like you know by going public they will have a full range of like you know debt and equity financing solutions and that could help you know fund future investment so but like some investors down in the south meeting with new spacex right now was kind of like surprised about how you know how much debt it's had took on and then like you know now with the cursor deal and like xai deals you know how much cash it will burn like in the futures are big questions for investors.

27:54Corey, are you hearing anything about what investors are thinking about this? And also, there was a detail that I wanted to ask you about in the story about how the CFO is being compensated that was kind of interesting. Walk us through those angles. Yeah, the real finance and accounting nerds will love this little Easter egg, which really just illustrates the situation that SpaceX is in now, going from this company that had this very profitable sort of internet business, satellite internet business in Starlink, and a pretty break-even to profitable business in rocket launch. And previously, the SpaceX board had compensated their CFO, Brett Johnson, based on the, you know, gold star metric of any CFO, which is free cash flow.

28:40You know, that is the, everyone would agree, the purest sort of measure of profitability, or maybe not everyone, because it's finance and people debate things all the time. But free cash flow, we know that's so pure, and that's how the board is going to compensate Brett Johnson based on hitting those metrics. And recently, it changed the metric that he is shooting for to adjusted EBITDA, which of course takes out any sort of accounting measure for capital expenditures through depreciation or interest payments on the debt. So to me, that just sort of showed this pivot that SpaceX has been in, in terms of what are its finance leaders actually incentivized to reach.

29:27So to put it more simply, Corey, so is the CFO now being more compensated towards how they pay down the debt or profitability? It's sort of forgiving him and forgiving the company for taking on a more capital-intensive approach to running the business. Okay, so it's saying you can issue the debt if you want. It's fine. We won't knock you for it, essentially. That's the way of the way. I mean, every company building AI models has taken on a ton of debt. The difference in SpaceX is, you know, unlike sort of the other publicly traded companies, you know, like Meta, like Microsoft, like, you know, I mean, Apple has a ton of debt.