Claude Hacked Real Companies During Tests, AWS Revenue Growth Accelerates, What’s Next for Apple?

0:13Welcome, everyone, to The Information's TI TV. My name is Akash Pasricha. It is Friday, July 31st. Today on the show, we're going to get to Amazon and Apple's quarterly results. We're going to start with a string of headlines around the big AI labs yesterday, including an anthropic hack and OpenAI's new pricing structure. We're then going to turn to our editors for this week's edition of The Editor's Cut. We'll take a bird's eye view on big tech's quarterly results at large and also look back on Tim Cook's tenure at Apple. And to close out the show, we'll discuss our weekend big read on an experimental new program in Montana that will allow almost anyone to access drugs that haven't yet been approved by the FDA.

0:57It's going to be a great show, so let's get right on into it. Anthropic said yesterday that Claude hacked into the systems of three companies during an internal security review. That review came after OpenAI disclosed that one of its models had breached hugging face. Here to comment on the latest on those attacks is Anastasios Angelopoulos, co-founder and CEO of Arena. Anastasius, welcome back to the show. It's great to have you here. Happy to be here. Excited to talk about this. So we got a lot to get through. So let's start with the Anthropic hack. I mean, funny thing to me was they didn't discover it until OpenAI had its hack and then it decided to go back and say, we're going to audit everything.

1:36But they found that it did. And I wonder what are your thoughts here? Well, listen, there's two stories here. There's the OpenAI story and the Anthropic story. Both are incredibly entertaining. So let's talk about the opening eye story first what happened is that let's let's talk about let's talk about anthropica that's that's the latest one yeah yeah let's talk about anthropic first so basically in the context of like some security uh some security tests that they were running internally they found that these anthropic models that actually hacked three real companies and so the way that it worked was like this one of the companies uh was a company that had the same name as a fake target that they were running a test against inside their environment so they said you know go hack the information and the information was a fictional target they had made inside their little sandbox and then the model was like oh i know the information the news outlet and then it escaped the testing environment that anthropic had made and one hacked the the real company so it wasn't obviously the information but it was a different company and that's right okay honest mistake yeah honest mistake well i don't know i mean like who's honest here the model's not the model sort of deceived the yeah okay into you know the thinking that it was like it basically like escaped its guardrails right so what we what we're seeing is that these models both open and anthropic are in these escape from prison scenarios where they started inside like a locked cell to perform these experiments and then are finding their way to sort of like scratch and claw their way out um the second incident that the anthropic model you know this was mythos it published a malicious package to pi pi that you could actually download um and the third was that an unreleased internal model it scanned a bunch of internet targets and uh you know basically you know almost did a sequel injection type attack um so i'm you know this this is like the the types of threats that are possible models.

3:35So these are the menu of hacks. You know, we've talked to people, people say, look, stuff like this will keep happening. The models will get better. I mean, this is all the cost of development. What are the implications, you know, that you see for businesses here? Is it literally just you have to beef up your security practice in a world where these mistakes will continue to happen? Or do we have implications in terms of which model to use? You're in the business of ranking these models i wonder if you take that or these events into consideration like what are the ripple effects of these hacks do you think yeah well certainly at arena we think a lot about how we're going to do security uh as a result of this and the reason is because like the you have to you know and this is probably the narrative that these companies want to promote by talking about this is that you have to start thinking about these models as potential nefarious actors that may even within the security guardrails or benchmarks that you set for them they may even know that they're being benchmarked but then when you put them out in the real world they could really do anything because and it's and it's basically a big win for the effective altruist guys that were saying this paperclip maximizer story from the beginning that if you set a goal and you're not specific enough about all the guardrails that you need along the way that the model might go and do something else crazy so you know the paperclip maximizer story is that if you set the goal of producing the largest number of paperclips, then the model might like eat the whole world to make paperclips and kill everybody in order to make paperclips out of them.

5:05Right. So this is kind of what we're seeing is that you told the model to do something inside of the small security test and then instead of hacked a real company. I mean, how the heck were you supposed to know that was going to happen? And so what you really need is somebody looking at the post deployment consequences of this model. Probably there needs to be several steps of auditing that says that's looking at the model all the time and saying hey are you doing something nefarious are you misleading your environment are you uh doing something malicious are you hacking are you uh scheming more people i mean more people basically i mean you know this is where the hiring is helpful that you need more people monitoring the stuff basically at these companies agents right and i think that the agents are more scaled but the agents are run by the models themselves so it's a little it's the agents that you have to have an attacker and a defender agent There's no way a human's going to be able to look at all this.

5:55I mean, things are happening at the speed of light. They're happening on a computer. And so having a human able to audit these things is just going to be impossible. We're going to be in a world with more agents than humans. You can't have a human looking at all of this. You're going to have to have automated defensive capabilities. Right. I want to get through a couple other headlines with you about models. So OpenAI changed up their pricing structure for a couple of their models. So Luna, which is the cheaper model in the 5.6 family, that is 80 % cheaper. Terra is the mid-tier model. It's 20 % cheaper.

6:29Take me inside what you think OpenAI is trying to do with this pricing strategy. And, I mean, look, on the surface here, they're just trying to win the market. I mean, that's obvious here. But where do you think they're trying to position themselves in the market compared to Anthropic? because, you know, we had folks on the show come on a little bit earlier this week talking about how Anthropic, with Opus 5, they're kind of moving up a weight class at the same price performance that they had with their older models. This is OpenAI coming down with price. So are we sort of seeing a delineation here between what ends of the market these two companies are targeting?

7:07Yeah, so I think the secular trend that's happening here is that there was the token maxing era earlier this year, and now everybody's talking about efficiency. Everybody's talking about what model is right, size, and best for my task. Of course, it's something that on Arena we have the best information about because we do all this real-world benchmarking. And so what we see in the new pricing for Luna is that it is really an extraordinary Pareto optimal model in the sense that there's no model in its price range that's necessarily better than Luna. And that's really the game that OpenAI wants to win.

7:41They don't just want to build the biggest, baddest model that's going to cost you$100 a token and is going to be so, so smart necessarily. I mean, they want to win that too. But another huge part of the narrative is going to be winning the cost performance trade-offs. Is my business getting ROI? Am I getting the outcomes that I need at the prices that I need? And this is what's on everybody's mind because everyone's done spending$1 ,000 asking an LLM to tell it about the weather. Do you think that this was a direct response to the open source, closed source discussions that have been going on and the fact that there is a conversation about spending less?

8:19How can there not be a response? How can there not be a response when a model comes out it's number one in the code arena and then it's less expensive than the frontier models from the labs. It's the same price as Sonnet. Of course, it's driven in part by competition. Competition is so good for us as consumers of LLMs. It means that all these labs are going to be competing to get the lowest and lowest and lowest price because of adoption reasons and when you talk to folks using these models customers etc i mean i already saw some businesses putting out bulletins saying just so you know you know that uh you can now use this many more tokens for your subscription price so what have the reviews been in the developer community of this pricing change how much of an impact will it have for them it'll have a big impact but you know the open source thing is also a huge factor here because the it's not just about pricing an open source the promise of the open source is hey you can own your own intelligence the model is going to live on your server you're not sending your data to some other company some other company that might eventually copy your business model you're not like in you don't have to worry about the government shutting off your source of intelligence that your business has been building on right so it might not do everything i mean just because it's cheaper it's not like people are going to switch right away exactly and And there's structural reasons why the open source models are going to be better for businesses.

9:46They're just going to decrease so much risk and they're going to make people feel more ownership. And that's the whole story around AI sovereignty. Every company wants AI sovereignty. And this is such a funny word. It's like, why do companies need sovereignty? They're not countries. But the point is that everyone wants to own their own intelligence. And even if you went on price, you really can't win that race when you're a closed source model provider. So they better win on price at least. Right. I know you got to run. Let me ask you one more quick question. Thinking Machines Lab, they released their latest model.

10:19It was a small model in their Inkling family. Just help us understand what is different about this model and any early reactions you're seeing from people. Yeah, absolutely. So if you look on the arena leaderboards, the new thinking machines, the small model, the inkling small, is like number 90 overall. Then when you subset to open source models, it's like number 21. But that doesn't tell the whole story, because it's actually a very small and efficient model. So when you subset to models that are within the same weight class as that model, which is a little model and a cheap model, it's actually top three.

10:56which is who would you compare it against like who are the smaller gemma 4 model and so on there's a few of these um okay and it's and that is an incredible achievement because if that's the type of model that you could potentially even run on a small smaller computer or maybe even locally that's where we're pushing towards um and they're absolutely they've absolutely built a frontier model in that regard that's the early reaction right great well uh i want to thank you anastasios for coming on. That is Anastasios Angelopoulos, the CEO of Arena here on TITV. Apple investors were spooked by the supply constraints the company flagged on its quarterly results last night.

11:37Total revenue growth is expected to decelerate. It grew 16 % in the most recent quarter, and it's expected to grow between 9 % and 11 % in the current quarter. For more color on that, I want to bring on Barton Crockett, Senior Research Analyst at Rosenblatt Securities. Barton, welcome to the show. It's great to have you back. Great. Thank you. Happy to be here. What were the headlines from last night's results for Apple? Well, I think that the headline is that their iPhone cycle continues to be historic, tremendous. We're near the end of that. And, you know, next year they have to comp it. And so I think that's a big question.

12:13I mean, historically and reasonably, you know, growth would slow after the 20 % plus type surge you've had in this iPhone generation. And, you know, that's coupled with this really crazy macro, right, where supply chain is challenged. It's hard for a company that has historically been able to get what it wants, when it wants, Apple to get what it wants, when it wants. And, you know, that's a new challenge for them. And, you know, with the stock at really historically high kind of valuations, the setup is really kind of interesting, I think, for a great company. Why was this iPhone cycle so strong specifically?

12:54You know, that's a really great question. I think that Apple makes great products and I think they made great iPhones here. I think that a lot of people bought an iPhone 12 during the pandemic. and a lot of those iPhone 12s need to be replaced. And so it's kind of a baby boomlet, kind of an echo of that boom that we had. So phones have died. They've had to be replaced and Apple's made some great devices and people have gone to them. Right, and then there's also the issue of which iPhones are actually able to run the Apple intelligence services. I mean, I can't remember exactly what it was, but I think it was the last model was the one I think you had to get to get the Apple intelligence.

13:41The other headline, obviously, that came out of the results was the supply constraints that they talked about. And the interesting thing was Tim Cook, he very much seemed to delineate between the memory chip crunch and then the supply constraints that they're seeing at TSMC for these advanced nodes or the advanced processors, I guess, that iPhones need. Right, right. How do you factor in supply constraints into your model when you think about the long-term valuation of the company, given that it's so cyclical? Yeah. Look, the first thing I think is this is not something you normally have to think about.

14:18I mean, really, you know, I can't really remember when there's been this kind of sustained pressure on Apple. And for really a generation, they've been the big man on the block and they're not right now. And so it does, at the end of the day, though, seem that this is unlikely to be permanent, that the market will find a way to make much of what Apple needs and that the world will normalize. I just don't know when. And I'm in my thought about what the terminal value is for an Apple and a long-term kind of DCF. I'm not going to have that there for a couple of interim years. It's going to be there and it's going to be a question mark about how long that continues.

15:01Right. When you say Apple is not the big man on the block now, I mean, what's the context in which you're saying to this? There's certainly the AI story, but how else are you referring to that story there? Well, it's really, it's the AI story, right? I mean, AI sucked up so much capacity in memory at TSMC. They're not the main driver of growth at those companies today and the main source of profit. And that's just a different place for them to sit. You know, it does, at the end of the day, get reflected in, you know, Apple having to charge more and not be able to get as much product as it wanted.

15:39And ultimately, that's the consumer kind of feeling the AI pinch in that way, but that's what's going on. Right. And so that's actually something I hadn't think about, which is that when you think about leverage that companies have with TSMC, it very much depends on the size of the orders that they're placing and where the iPhones used to be the dominant product. I guess now everything is focused on GPUs and that's making it harder for them to get that capacity. They talked a little bit about rising prices last night on the call. How are you factoring that into your modeling? Well, you know, you kind of stick your finger in the wind and try and get at it.

16:21You know, nobody really knows precisely. There's no great secret sauce to it. What we do know is that you can go back for years and years and see when they have these occasional spikes in iPhone sales one year, the next year, it's always been a much slower year, flattish or down. And so the pattern would seem to argue for that. And then the noise is they're going to be charging more for phones. There's some price sensitivity. people probably are willing to be less price sensitive for Apple products and others and they've got a new premium priced foldable phone that's going to be in the mix yeah you might be almost I mean 20 2500 bucks up to that much I guess what people are talking yeah yeah and you know there's definitely a niche I don't think it's mainstream um but you don't think you don't think it's gonna it's gonna be big for the company I think it's gonna be you know a few percent of sales, you know, a low to mid single digit percent of sales at, you know, a real premium price and, you know, kind of 2x the price.

17:29And so it might be, you know, higher than that kind of percent of sales. And, you know, within that, though, there's going to be some cannibalization from other devices and there's going to be this difficult comp. So I think it still settles out that Apple, after the September quarter, when you have to face the music of strong comps, sales are going to be much slower than they are today. Right. We'll come back to the foldable phone in a second. The services revenue was the other interesting part of last night's results. So there was some – there was expected to be some deceleration in that business.

18:11And the interesting part was that I believe the margin also came down for services, which they attributed to a different product mix in services. So, I mean, this was kind of interesting to me, the fact that growth was coming down, margin was coming down. Were you satisfied by the explanation there? What's going on in that business? Well, you know, I'm not terribly concerned. I mean, they were growing, I think, 12 % in the quarter in services. You know, I think the implicit in the guidance when they talked about foreign exchange is that they decelerate maybe to 10 % in the September quarter. You know, so things are slowing down some.

18:52The margin on services is really high, you know, and so a couple of points here or there up or down is really a rounding error. It doesn't matter relative to the, you know, mix of services versus products. And, you know, so all that's fine. I do think that, you know, the question of services growth is one of, you know, geez, you have this platform. you should be able to milk at 10 Ways to Tuesday for services revenue. But there is a little bit of a governor in that, you know, there's some antitrust concerns. And there's also just what can they do well? They've been incredibly strong at things you wouldn't think.

19:31Like they've done great with movies. They've done great with TV shows. And you wouldn't have expected that. I wouldn't have expected that from a Tim Cook-led company. But there we have it. and uh uh but at lasso i mean you know he he was he's clearly a big fan yeah he's a big fan of lasso they did the formula one movie uh they got that in the fold and um you know they're uh they're doing great but um um you know the i think the real way you make money in services is you know through the app store and um you know through um the cloud i mean the cloud storage that they sell and I think that the app store is really in this great place, right, with so many people vibe coding new apps and they can't get them all out onto the app store.

20:18Presumably, you know, that should be a healthy backdrop for revenue. But then there's the imponderable is what happens if the future is some type of smartphone without an app store or everything's an AI interface. what happens if Johnny Ive makes that work through open AI or some such? So there's that important question. And I mean, that's what I was a little bit confused about. It was on the backs of the vibe coding, easy to make an app boom. That's why the services revenue, the app store, I guess, I mean, they didn't seem to break out exactly what was going on with the app store exactly. Exactly.

20:58I just, I would have thought, you know, it would have set the stage for acceleration in that category. Yeah, you know, it's, we'll have to kind of probe a little bit more, you know, but if I was to guess, you know, without having done survey work in this point, exactly. You know, the consumer is doing a lot with apps, but maybe a little bit of time is being spent with some apps in particular, your LLMs, your AI apps. and maybe that's eating into some of the other times than gaming, which is really what drives the app revenue. Brent, before I let you go, let's talk about what wasn't discussed as much on the call.

21:37So this whole debate around open source and closed source models, we know that Apple has taken a bit of a different tone with AI, a little bit of a less aggressive tone in terms of its own build. You know, Apple hasn't really seemed to opine on that at all. And I wondered where you think they stand on the issue, what you think their calculus here is on supporting the Frontier Labs, but then also the reality that open source is popular right now. Right. Well, you know, I think that what they put in the app stores is one question. And I'm sure, you know, they're going to be as open as the regulations, you know, would be comfortable with.

22:17They're not going to push the envelope. But, you know, and I think there's a regulatory question in terms of what they use themselves. Obviously, you know, they've wedded Gemini as the foundation and then they do their own customization on top. You know, I think the the longer term question is, you know, is Gemini the right answer or does it need to be in the future? You know, something different outside of China, obviously in China. They've you know, they've got a deal with Alibaba, with Quinn and another deal. And, you know, there's this regional balkanization of LLMs, which is forcing their hand to a degree.

22:52But, you know, in the future, maybe they'll be able to do what the rest of the world's moving towards, which is choose the cheapest model for the job, a mix of very inexpensive, you know, open source models from a range of players and then the edge case on the frontier. and they've been jacking up their spending on R &D and maybe that'll be one outcome of it is they can build some more efficiency and effectiveness on the service end of it. And I missed that one detail. Did they have a deal with Alibaba in China for using models there? Yeah, that's my understanding is that's one of the two kind of sources in China for them, but strictly for China.

23:31Cool. Well, Barton, I want to thank you. Oh, I forgot to ask you. Are you going to get the foldable iPhone, do you think, at$2 ,500? That's what I was going to say. I'm probably, it's probably not my first spend. Okay. I don't know if I'm kind of expensive. I'm going to get it. I'm going to get it. I'm going to be one of the only people, apparently, who are going to get it, because most people I talk to are actually not as enthusiastic as I am, but I think it should be interesting. Cool. Well, tell me about it. Great. Okay. Well, Barton, I want to thank you for coming on. That is Barton Crockett, Senior Research Analyst at Rosenblatt Securities here on TITV.

24:07Amazon shares jumped after its quarterly results last night. AWS growth accelerated by nine percentage points compared to last quarter's growth rate. I want to bring on Lloyd Walmsley, Internet Equity Research Analyst at Mizuho, to talk more about what we should be paying attention to. Lloyd, welcome to the show. It's great to have you here. Great to be here. Thanks for having me. So let's talk about the AWS acceleration. What is driving that right now? Yeah, so I think it's two big things they called out. One, obviously AI, the growth there is tremendous, and they're bringing capacity online from Project Rainier.

24:44And so the minute they bring it online, they can monetize it. And then the second thing they talked about was chips. The training in chips, which is obviously part of AI, so it's overlap, it's not completely different. And then the Graviton chips, they're just seeing tremendous demand for those chips, largely also from AI. So those are the two big things. And then, look, they're also improving their own services around AI, sort of moving up the software stack, whether that's security, whether that's call center product. They're doing a lot in and around AI that's helping them as well. Right. So they obviously, they said these two figures,$25 billion run rate for AI related revenue, I think it was, and then$25 billion for the run rate for chips.

25:37And they didn't give us a whole lot of color in terms of what's inside those two$25 billion businesses. Let's start with the AI revenue. Do you have any sense on your end based on your research? I mean, what are they including and excluding in that segment at all? Yeah. So there's a bunch of different components. I think the big ones are, you know, Tranium chips are part of that. Yeah. That would be the chips revenue, right? Well, so these are not two separate buckets of revenue.

26:11Martin Peers:There's significant overlap. Oh, okay. Okay. And so, you know, to your point, they don't disclose that much, but there's significant overlap. And so a lot of AI consumption involves chips. So I think that's a big part of it. And then Bedrock is a huge part of it, obviously, their platform where their enterprise customers can deploy any model they want and have it managed and hosted by Amazon. I think that is on a complete tear as well as more and more companies who have been experimenting, running POCs in AI start to move into scaled inference. That is Amazon's wheelhouse, right? They have very low cost inference and they've got all their customers who already have the bulk of their data sitting on AWS.

27:08And so it's a natural place to run your AI inference. And that's only gotten more compelling now that they have OpenAI's models as part of bedrock as well. Right. And I'm seeing here, you know, I should have caught this, but there has to be significant overlap because the total revenue for AWS was 42, which is lower than 25 plus 25. but i want to move to sort of you know the the broad comparing amazon's results to the other big tech companies hyperscalers that reported because i mean if you look at who was burning cash and who was not microsoft came out they said we are not going to burn cash amazon was google was uh google investors initially sent the stock down after results but Amazon, they didn't seem so concerned.

27:58Both businesses had acceleration in their cloud businesses. So what do you attribute the difference to in the investor's reaction? Yeah, so I think there's been a lot of bad, I wouldn't even say reporting. I think throughout the investor community, the press community, people have attributed the weakness at Google and Meta to this like CapEx and free cashflow issue. I don't think that surprised anybody. And I think what led to the weakness at Google was it wasn't a clean quarter, right? They were talking down search growth in the second half. It was the first quarter where search actually missed, and they'd been clobbering numbers.

28:42So I think there were some factors at Google that were unrelated to the CapEx. Like every investor I've spoken to before and after the quarter expected CapEx to go up this quarter. So I don't think that was the surprise. I think the surprise was marginal weakness in the core business. And then you turn to Meta. Again, everyone expected CapEx to go up. The surprise was everybody expected them to announce some sort of rental deal with someone like Anthropic. and not only did they not announce a deal, they got on their call and said, you know, we've gotten a lot of offers and we might do this, but we might not because we'd rather sell a higher value intelligence than like just our own infrastructure.

29:27And so whereas Amazon gets on, and yeah, they spent, the CapEx went up, but like to your intro point, nine points of acceleration. And I think one of the problems that has bedeviled Amazon stock over the last year plus has been, They're spending all this money in CapEx and they see this small, little, nice, but not that impressive acceleration at AWS. Right. It was three points. Then it became four points. You know, good. But like you're spending two hundred billion dollars over the last two years. Like we need more than four points. And this quarter, we got nine points and we got record margins in a quarter where margins usually come down.

30:09and everything they said on their call suggested that these margins weren't unusual and should continue to be very healthy and that the AI margins themselves are on track to, you know, tracking ahead of the core cloud business. So like, we're finally getting everything we've been wanting for a long time at Amazon. Whereas you look at Meta, you look at Google and there was just not, they weren't clean quartered. And so I don't think it, the bottom line, I don't think it was the capex i don't think it was the negative free cash flow i think if um if meta was was knocking the ball off on an ai product no no one would be as concerned on the capex spending the problem is no one even knows what they're building really um beyond like another chat let me ask you so on the aws front do you think the growth is sustainable the margin expansion is it sustainable what are you expecting yeah so we're expecting growth to pick up again in the third quarter.