SpaceX’s $7.8B Q2 Revenue, AMD Stock Falls Despite 50% Revenue Growth, China’s New AI Gold Rush

0:12Welcome everyone to the Informations TITV. My name is Stephanie Palazzolo and it's Wednesday, August 5th. First up, we have reactions to SpaceX's first ever earnings report and we'll discuss the company's big ambitions for its Starlink mobile business. We'll also dive into AMD's second quarter results. Shares have fallen since the report despite strong revenue growth. Plus, we have new reporting from our Asia Bureau about a new trend emerging. Chinese founders are looking to launch their own AI labs focused specifically on so-called world models. We'll hear from Jiro Osawa, who wrote the story, and our Asia Bureau chief, Jingyang.

0:51It's going to be a great show, so let's get on into it. LESLIE KENDRICK, SpaceX's first quarterly earnings report since its IPO showed the company is spending heavily on AI. The company did report$7.8 billion in revenue, up 92 % from the prior year. Here to comment on the report is Tim Ferrara, president at TMF Associates. Tim, welcome back to the show. TIM FERRARA, Great to be here. LESLIE KENDRICK, So initial reactions, what are your kind of first reactions to the report? Well, I think the headline numbers are very good in terms of revenue growth. Subscriber growth was higher than I expected at 1.7 million new Starlink subscribers.

1:32That's all great, especially when they raised prices at the end of May, which many people expected to cause a slowdown. So that's all great, and they expressed a lot of ambitions for huge amounts of growth next year. I think the challenge is really, Elon Musk said an awful lot of things that mostly seem quite crazy. Things like launching Starship every day next year. And it's hard to pass what's real and what's just him stating ambitions that are not going to be realized. And that makes it really hard to predict what's going to happen next year. But they said a lot about taking on the telcos and competing more aggressively there with Starlink Mobile.

2:19And I think that's a really interesting story that's going to run and run in the coming months. Before we get into Starlink Mobile, I am curious, if you had to rank kind of the top craziest comments from the earnings report yesterday, from maybe most believable to least believable, I'm curious how that ranking would kind of fall. I know there's some big numbers Yeah, I mean, they talked about a trillion dollars of revenue in 2013. They talked about 10 gigawatts of AI data centers next year. They talked about daily Starship launches. I mean, I think the hardest one to believe are the numbers for next year.

2:59People can take a view on revenue growth over a four-year period, and it probably won't be anywhere close to a trillion dollars. But, you know, that's not going to matter in the short term. What is going to matter in the short term is getting Starship ramped up and building more data centers. And I think both, in my mind, you know, the ambition to do what he said for next year is not credible. And you mentioned Starlink Mobile, which is obviously a very important business for them, something that they're talking a lot about. We didn't actually get revenue numbers for that business segment yesterday, but we've previously reported that.

3:40It's a pretty portion of their total sales. What do you make of SpaceX not disclosing that number yesterday? Yeah, I mean, they didn't give much of a breakdown at all between the enterprise and government and mobile segments. I mean, that's not a great sign. They didn't even give the metrics about monthly average users that they gave in the S1. And so, you know, a lot of this talk about competing with the telcos in some ways is sort of posturing to get a better deal out of T-Mobile. You've reported previously that that deal is only worth about$100 million over four years. And clearly, SpaceX needs an awful lot more than that to justify the$20 billion it's investing in Spectrum, not to mention the billions of additional dollars that they're investing in new satellites.

4:34So, you know, there's this big knife fight going on with the telcos and the announcements that they're going to try and compete with those telcos and build a terrestrial network are all feeding into that debate on the SpaceX side. And on the telco side, we're seeing a lot of comments about how this isn't really, D2D isn't really that necessary. It's a complementary service. It's only for national parks. It's a tiny percentage of usage and it's not a core part of the business. So, you know, we're seeing a lot of that posturing on both sides, a big fight. And at the end of the day, it's about how much the telcos are prepared to pay.

5:10And if they're not prepared to pay enough for SpaceX, then SpaceX is going to have to compete themselves. You know, it sounds like, as you mentioned, maybe a lot of posturing on the call yesterday. I mean, again, as you said, they were pretty aggressive on the call, saying things like, you know, So, you know, Gwen saying that she thinks that Starlink is going to get customers of companies like AT &T, Verizon and T-Mobile. What are the chances of that? Do you think that's a realistic goal for them? Well, I think that there will be a lot more competition in broadband going forward. I mean, although they raised the prices at the end of May, they could come back with lower prices.

5:47They just built a new terminal. It's going to be cheaper to manufacture. And they're going to launch more satellites and more capacity in the coming months and years. So on the broadband side, yeah, I think they will be aggressive with regard to the telcos and be more competitive and take subscribers there, although probably not a lot from fiber and other high-end terrestrial services. The mobile side, I think, is much more questionable. And they talked about building out a small cell network, put things on rooftops, along with sterling antennas, and build out that way, which would be more cost efficient.

6:27I mean, I think that's a belief for analysts that they're not going to be spending hundreds of billions of dollars to buy a telco or tens of billions of dollars to build a tower network. But the question is whether that's going to be really competitive and compelling for subscribers. If D2D is all about coverage in national parks, well, we don't go to national parks that often, and it's not really a good reason to choose your mobile service. And if this small cell network doesn't measure up to what the telcos are doing, then it's going to be hard for them to gain share. And I think one thing that you've, you know, an argument that you've made multiple times on the show is that, Starlink overall is a great business, but maybe not enough by itself to justify SpaceX's valuation.

7:13Did the numbers from yesterday overall, you think, weaken or strengthen that argument? Well, as I say, the numbers yesterday are very good. The growth is great. But we're not necessarily going to get to the hundreds of millions of subscribers on the broadband side. There's no indication that that is on the agenda right now, especially after the price rises. And on the mobile side, it's still all very tricky. So, you know, amazing what they're doing relative to the rest of the satellite industry. I mean, they're just beating everyone hands down and people like Amazon are struggling to catch up.

7:50But competing with the terrestrial telcos is a different kettle of fish. And we're yet to see them really make a big impact on the terrestrial market, whether that's in the U.S. or elsewhere. Let's talk a bit about the kind of AI side of the business. It feels like these days, you know, what was once, you know, XAI, now SpaceX AI, is almost, it almost feels like this kind of cloud provider business at this point. Are investors as excited about that? You know, do they feel like that part of the business is going well? And how sustainable is that sort of business model going to be in the long run?

8:26Well, they were talking a lot in the past about, the orbital data center part of the business. But the focus yesterday was more on the short-term build-out well ahead of what most analysts were expecting in terms of the gigawatts of AI data centers. And we're yet to see what the orbital data center deployment plan is going to look like. I mean, sure, if Starship is launching every day, then there'll be a bunch of capacity for these orbital data centers, but they still need to do a lot to get that ready in terms of the chips to go in those satellites and the launch site. I mean, there's been rumors in the last few days about a new launch site in Louisiana that will be suitable for launching to the orbits that they need for the orbital data centers, but that will take years to stand up and get working, as will the TerraFab to make the chips to go in these satellites.

9:20So, you know, the focus in the short term is on doing stuff on the ground and musk has been very good at making his data centers more quickly than some of the competitors but to what degree are they going to be able to sell all of that capacity to other people even if they can build it in the next year or two that's that remains an open question i think everything you just mentioned you know launching data centers into space building them on earth i think one thing all this has in common is it's obviously very expensive and burns a lot of cash. I mean, yesterday, even with the strong revenue growth that we saw, you know, they still burned twice as much cash as they brought in in revenue, and they had over$18 billion in CapEx.

10:03How worrying is that? Is there kind of a path to profitability that investors see? What's going on there? Yeah, that's right. I mean, if you take Musk's statements about the daily Starship launches and the 10 gigawatts of data centers next year at face value, then that investment, CapEx, is going to go up even faster. It's going to end up in the hundreds of billions of dollars, not just tens of billions of dollars a year. Is that sustainable? Will they be able to access the funding to do that? Remains to be seen. Or is this just really talk and they're going to come out a long way short of that next year?

10:39We're waiting to see what happens on the next Starship launch. We're waiting to see what happens in terms of the data centers that they're going to build. They've got this new deal with NVIDIA. That's going to help to some degree. But yeah, building stuff on the ground, we've been talking about why all these data centers need to go into space because it's hard to build stuff on the ground. But now they're saying they're going to build all these things on the ground in the next year. So let's see what happens. And it looks like this week might be a volatile one for SpaceX. You know, curious, what did you make of the kind of stock market reaction yesterday?

11:14And then obviously we have the first kind of post-IPO lockup expiration date coming up tomorrow. You know, a lot of people are speculating that's going to lead to a lot of moves in the stock price. How's this week going to shake out for SpaceX's stock? Yeah, and we're also hearing rumors about a merger between SpaceX and Tesla as well. That's sort of sitting there in the background. So, you know, there's a lot of moving pieces here. I think that, as we talked about, valuation is a major concern for this company. And Musk is trying to live up to that valuation by making these enormous promises of revenue growth and data centers and Starship launches.

11:56It's going to take some time before we see exactly how that shakes out. But yeah, in the near term, I think things are going to be volatile and people will be focusing on the merger potential between SpaceX and Tesla as well. And that will cause a lot of upset. I think if they have to get a price that is acceptable to Tesla holders, then that probably is a downside risk on the SpaceX side as well. Mm-hmm. And we talked about this a bit on our show yesterday, but it seems like there are pros and cons for the fact that SpaceX's stock has fallen since the IPO for the Tesla, you know, for the potential Tesla merger on one hand.

12:38The stock is maybe less valuable, so it's harder to use that for an acquisition. But maybe now that they're both trading at similar, you know, more similar levels, that could also help the possibility of a merger as well. So, you know, but it seems like things are very unclear and people are very much still speculating on whether this is going to happen or not. Yeah, that's right. I mean, I think the valuations being more comparable definitely helps. You know, I think ultimately. SpaceX is the center of Musk's public universe going forward and Tesla will get less attention. So it does make sense to bring it into the fold and bring it under SpaceX.

13:16And, you know, there's a lot that they're trying to do, things like TerraFab that could be, you know, have commonalities between the two companies. And this would boost the revenue substantially because obviously Tesla generates a lot more revenue right now than SpaceX. But it does just introduce an awful lot of uncertainty into this going forward. And I do think we'll see lots and lots of volatility. you know, Musk will presumably keep throwing things out there on X about his ambitions and factories on the moon and all sorts of crazy things. And, you know, is the reaction to that, oh, well, he's just talking or is the reaction to that, oh, my goodness, that's more CapEx for the company going forward?

14:01Well, I'm sure we'll have you back on super soon for, you know, who knows, the Tesla merger, the stock price craziness later on this week. Either way, lots of news coming for SpaceX. Thanks so much again for coming on. And that was Tim Ferrara, president at TMF Associates. AMD shares are down following earnings despite the company reporting 50 % revenue growth in its second quarter. Here to comment on the report is Ben Palladian, founder and CEO of BEP Research. Welcome back to the show, Ben. Hi, good morning. Thanks for having me here. I appreciate everything. Of course. So, you know, AMD seemed to have a great quarter on paper, but obviously the market did not react in that way.

14:44Why do you think the market was reacting so badly to the earnings yesterday? I was at the Advancing AI event a few weeks ago in San Francisco, and a lot of hype and promise for the company. They've been really trying to push the AI hyperscale narrative, which they have. But I think the challenge is that Helios, which is their premier double-wide rack scale product, will be effectively in high ramp probably the end of this year or really into the first quarter of next year. And that really pushes out the revenue and numbers for a lot of the estimates that Wall Street has and a lot of these analysts.

15:27And the growth isn't really there. Coupled that with what Elon said yesterday on the SpaceX call, he said we're exclusively going to use NVIDIA GPUs on the Verirubin architecture. So another blow for AMD from a product customer standpoint as well. And how bad is it that the Helios revenue is getting pushed to later this year or even next year? I mean, I think speed is of the essence here, obviously, in AI and all the AI chip startups and companies are trying to move very fast to catch up with NVIDIA. It's time to market. So if you look at data center developers and, you know, powered land, it really starts at that point, maybe like a year and a half or a year ago, where you're thinking about what architecture you're going to put into that data center.

16:21So if you're picking AMD and you want to use those products, you probably plan this out maybe two years before or a year and a half before. And getting that supply at the right time at the right moment to work is important because you want to energize as fast as you can to get those tokens online. And the longer it takes to get that ramp, the more behind you are, even though you say you have a better product at the time. Mm-hmm. And, you know, obviously AMD also makes money from CPUs as well. That's a big part of their business today. They're increasingly making money from CPUs specifically in data centers.

17:01And they did say yesterday that they are taking market share there from Intel. However, we're also seeing, you know, increased competition from NVIDIA as well, who's also now selling standalone CPU server racks. Where do you kind of see all this shaking out in terms of these market dynamics for CPUs in the next, you know, 12 to 18 months. There's definitely a big inflection in genetic AI, and that's where the CPU usage is getting a lot of its demand from. And you've seen companies like AMD and Intel raising CPU prices 35%, 40%. Traditional, you know, server makers like Dell and HP are marking up servers that cost businesses, you know,$20 ,000,$30 ,000.

17:45Now there are$100 ,000 Forex because of memory and storage, everything. There's this huge compute inflation going on because demand is so acute and supply is so tight. AMD has a great CPU franchise, and that's probably what they're going to lead with. But it's really not enough to carry them into what is going on on the AI infrastructure buildup. That's just one wedge to basically open up other things. but they need their helios rack system and other systems to really ramp up and get there fast i think one of the challenges that people really miss is that helios is sort of what blackwell to nvidia was a few years ago the the rack trace has like a lot of these big thick cables and it's it's pretty complex and it's double wide so when you try to actually assemble that scale like you're going to have issues.

18:40And NVIDIA actually had delays and went through this whole learning curve to understand how to assemble these things because they're so big and massive. And on the next generation for Verirubin, they actually went cabled this. So from a physical product generational standpoint, you could effectively think or say that AMD Helios is almost like a year behind in technology knowledge on that. So it sounds like both, you know helios is going to take maybe longer than some people expected to ramp up but even whenever it does there's it sounds like there's a pretty good chance that they might run into issues that they weren't expecting around like actually getting the physical server racks up and running similar to what we saw with nvidia and blackwell as you pointed out yes i mean that's kind of the the way we see it and uh the way at least as you approach the call as far as the ramp You could really hear that things, a few items are shipping this quarter and then a little bit more next quarter.

19:38And then the big ramp, according to her, was Q1 of 2027, which in AI data center land is almost like a generation behind from our competitor as far as time to first token. Yeah. And taking it back to CPUs for a second, can you just break down to me, you know, why exactly are CPUs important for agentic AI specifically? The CPU is sort of like the conductor. It tells the GPUs what to do. And because you're running multiple agents at the same time doing these processes, you need the CPU to sort of schedule these things. and you see a lot of CPU tool call usage and coding workflows, obviously like Cloud Code or Codex, running multiple agents at the same time, building something, running a web search, adding to a repo.

20:37So you have these things all working in parallel for you and the CPU is orchestrating that. A lot of the time the GPU isn't really doing anything. It's waiting for the instructions from the CPU to do those. And as you add multiple call loops and tasks, you increase your CPU usage. And I think the big debate or the question in industry between all these companies is how many, what's the ratio between the number of CPUs and GPUs? I think, you know, going back to what NVIDIA has done with the Vera Rubin and their architecture, the Vera CPU, I think it was purpose built for this agentic moment when it was co-designed with OpenAI and other companies.

21:20And it really has the performance for these long horizon agentic tool calls and tasks that are usually happening with these new AI models. Basically doing longer work at longer periods of time without less interruptions. And that's where all these things play in. So it sounds like because of all these reasons, AMD is facing a bit of an uphill battle. I think even more than that, AMD obviously is seeing its revenue grow quickly, but NVIDIA is growing faster and it has a larger revenue base to begin with. I mean, what do you think it'll take? What is it going to take for AMD to sustain that growth and even stand a chance at catching up with NVIDIA or getting closer in some way?

22:10What does it need to do and really execute on in the next 12 months? I wrote about this last night on my research sub stack. I feel like AMD is sort of squeezed in the middle because on the top, you have NVIDIA, who basically dominates on training and inference. You have Google TPU, and then you have basically every hyperscaler, like Amazon Tranium, Microsoft Maya. Even Meta has its own custom chip and then OpenAI, Jalapeno. So you have that custom stack and many of these guys on the bottom, like these new startups, or like Cerebrus or Etched or Positron. They're all trying to get a piece of inference and they're going to hyperscalers and saying, hey, can you just put our equipment in your data center?

23:00We'll give this to you for free. We just want to show that we have customers. and AMD is sort of stuck in the middle because they're getting compressed on both sides from a margin perspective and a margin profile because they're not going to win on training and they're trying to be the best, lowest cost inference player but then you have all these other guys trying to do it as well. So I think for them to really find their value proposition is they really need to lean in on their CPU franchise and really build a system that works really well for specific agentic workloads and dominate on that before the startups really try to eat their market share and nvidia i think will still have a large dominating market share of what it's doing but i but the market is getting so big i think lisa sue at the event put by 2030 the total addressable market for advanced hpc computer data center ai is about 1.3 trillion dollars So I think there's enough for everyone to play and get a piece.

24:09The question is, is how much of that do you get and what's your margin profile behind it? Kind of the entire pie is growing. So everybody's share is growing, even if you're not eating into NVIDIAs. I mean, I think along those lines, taking a step back, you mentioned, obviously, there's AMD, there's the hyperscalers that have their own chips. There's the model developers that have their own chips. And then there's just like the dozens of AI chip startups out there, you know, like Etch and all these others that are trying their hand at this too. I'm just curious, you know, compared to 12 months ago, would you say that you're more, like, are you more or less optimistic, I guess, about all these other players, you know, eating into NVIDIA's share?

24:50I think competition is obviously healthy for the market. uh obviously when going back 20 years and video was always always competing against other graphics card companies of 3dfx and silicon graphics it's the nature of of uh you know silicon valley and startups and disruption if you see a big market and there's a huge opportunity why not uh try to attack it and a lot of these investors and venture capital firms their exit strategy for these companies isn't an IPO it's it's creating that nuisance factor that eventually one of the larger companies will absorb them or the team so you have you know Intel and Samba Nova that are already close together you already have this uh cerebrus AMD partnership that's happening and then maybe have a host of other startups that will probably go to Qualcomm or Marvell or one of these other larger chip companies will absorb the teams to further their custom ASIC and AI designs.

25:54And I think that's just how things go in the cycles of the Valley. It's just another cycle. Very true. Well, again, thank you so much for coming on, Ben. And that was Ben Puladian, founder and CEO of BEP Research. A flood of Chinese entrepreneurs are rushing to start their own AI labs focused on models that can power robots and self-driving cars. That's according to reporting from my colleague Juro Osawa. Juro spoke to Asia Bureau Chief Jin Yang about what he learned. Here's that conversation.

26:29Juro Osawa:Juro Osawa Hi, Juro. So you reported that there's been a bit of like a new AI gold rush going on in China. And it's all focused on work models. Can you tell us what first, what is a world model and what's the use of a world model? Sure.

26:45Jing Yang:So world models are AI models that basically understand the physical world. So it can simulate, for example, how objects interact with the environment. And this is important because it could lead to, you know, breakthroughs in robotics, especially, you know, humanoids can understand and operate the real world. in the real world situations.

27:10Juro Osawa:And what's the reason for this new generation of Chinese Neo labs, Neo AI labs, all crowding into the world model area?

27:23Jing Yang:So they see a lot of potential applications for this at home, because world models are very relevant to China because China has the world's biggest supply chain for humanoid robots, autonomous driving. And there's a lot of government support for robotics in China, automation. So if world models can really develop in China, then there could be a lot of opportunities in the future. And even though the LLMs, the AI models are already very crowded, but startups are now seeing new opportunities for this area. I see.

28:08Juro Osawa:So I remember that we saw some sort of, you know, gold rush in China not so long ago with the explosion of China competing in late 2022. We saw many startups launching China, working on large language models. What's sort of the difference between the current work model gold rush and the LLM gold rush we saw back in 2023?