EP 179: Apple's WWDC and Siri AI, Nebius Inflection, China Trip Takeaways

[MUSIC] Hello everyone, welcome to another episode of The Circuit. I am Ben Behren. [MUSIC] [FOREIGN] I'm Jay Goldberg. [MUSIC] So I want to call attention to something real quick in my video. I have brought back Bunny Suit Guy right here. And I'm doing a fun series on Twitter for anybody cares where Bunny Suit Guy goes and does fun things on my farm. So Bunny Suit Guy has helped me tend to bees and picked peaches and picked blackberries. And this weekend, the adventures of Intel Bunny Suit Guy continue. So there you go. [LAUGH] This is the randomness I think to entertain myself, Jay. >> Okay. >> And entertain myself. >> Thank you, thank you for that quality content. [LAUGH] >> I am, I am to please and subtly entertain. So there you go. >> Okay. >> Just so everybody knows, Ben is referring to an Intel stuffed figure dressed in a bunny room appropriate for a clean room environment. >> Yes. >> For some of you that could be a fab. Not the other bunny suit. Some of you may be picturing. >> Of a furry or some kind of ex. >> There are others out there. Don't Google it. Ben's talking about semi-ductor fab. >> For the audio only listeners. There you go. And you can see my Twitter of Bunny Suit Guy that I got probably in like 2004 from Intel. Okay, headliner for this week was Apple's worldwide developer conference where a lot of stuff went down. I was there. It was very interesting to be in the room. They also had a session afterwards that they let a handful of us into to kind of get the nitty gritty on what they did with the new Siri. So I do want to talk about that because I had a couple of interesting takeaways. But let's just sort of start headline. I'll give you kind of the thousand foot view and then we can dig into it. So a lot of people kind of mentioned to me, well, one, why did they not go platform by platform? In the last few years it was here's Mac OS, here's iPad OS, here's iOS and kind of went through a watch OS. They went through each platform specific update as well as including new features, new experiences, new dev kits, APIs. There was none of that this year. It was basically it started in kind of three, it had three pillars. The first was performance upgrades where they spent time talking about all the things they did to make the platform faster, more efficient, more issue with bugs and whatnot. And so two interesting things came from that. One, they called out the ever glorious topic of CPU scheduling with an operating system. And remember people like we're sitting there and they're like, why are they talking about a CPU scheduler? And the reality is every year Apple updates their CPU scheduler. In fact, we regularly track the performance efficiencies that happen on particularly Mac OS, if you will, and where things get faster, where it manages Apple Silicon better and a lot of that, right, just comes back to a CPU scheduler. But they called it out and they called it out as an important part of their platform update for speed. And interestingly, this one seems like a good one because even older Mac and Siri Silicon feel like they're faster. And we've had to talk to a couple people who've installed the latest on really old kind of performance. And the old iPhones, Mac's on my team was saying like an iPhone 11 feels exceptionally fast, which is something. So performance upgrades, a big part of it. The other one, which to me was sort of the biggest tell for where they were going with this hyped up concept of new Siri, was they fundamentally rebuilt the indexer. So anybody who's on an Apple platform knows that the second you get a new Mac or a new iPhone or whatever, it has to index. It goes through and basically collects all of the data, all of the information and what not. And I can take a real long time, which is why your battery life typically sucks for the first few days when you get a new Apple product. But they specifically called out, they rebuilt the index, right? And the reality is they needed to rebuild the index. Semantic search, how it understands the context of all your conversations and what had to be rebuilt. So Siri might actually work, right? And that was kind of the directional path, right? And then the second part of this was they doubled down on trust and safety, which was super interesting actually because obviously, you know, Apple's on the hill that they'll die on on user privacy. But my takeaway from that was similarly, you also needed to double down on that and prove that you have the most safe, secure and private platform in order to do the next step, which was talk about the all new Siri. So it was almost like those two kind of pillars had to flow, had to happen. They had to do that work in order to get to the point where this Siri works the way that they want in a much more conversational, deeply contextual, good with search, knows what's in your email, knows what's in your contacts, knows what's in your calendar, it's such, such, which we're all actively beta testing at the, I say we're all, Jay's probably not yet. We're all actively testing the new, the new series. So high level platform updates to two key pillars led to now what was possible in new Siri and new Siri is going over well. A lot of the demos worked very well. There was some compelling use cases largely around semantic search. How Apple's doing and we'll talk about this relative to Gemini. But that was, that's my thousand foot view of how they architected the platform to get to a point to where Siri can finally do the things that we'd all hoped and, and, and be a more intelligent assistant. Okay. Thank you. Thank you for that. I'm speechless. He was speechless. So, um, I got to say I was a little bit underwhelmed by the BWDC. Um, I, I appreciate your point that, um, schedulers are very, very important. I mean, who thought we'd be sitting there talking about CPU schedules? I didn't, but I wanted to highlight it. But it seems, you gotta admit, it seems a little bit underwhelming when the world is moving as fast as it is with AI, new features coming out every day. And they're talking about schedulers. Right? Now in fairness, WWDC is an event for developers. Right? They have lots of tracks going on, presumably this week. Um, they didn't really tell us what was going to be in those tracks, which I think is weird. Um, to your point, they usually talk about the platforms. Let's talk about things developers want to hear, give them reasons to go tune into different, different, you know, features and talks and attend. Um, so I don't have a, I didn't feel like there's lots of go on. The privacy stuff felt important. They have all these new parental controls, which I guess personally I care about. Um, but I've used a lot of their parental controls in the past and was kind of like, they need to do more. So maybe, maybe I get that. Um, I think the real highlight for most people was going to be what they said about AI. And I think I'll put an optimistic spin on it and say, this is where Apple should have been a year ago with AI. Yeah. Just making improvements to Siri. And they, they last year they made these huge announcements around Apple intelligence. And I think, you know, on hindsight, that is what they would like Apple intelligence to be. Agreed. But it was going to always going to take them a while to get there. And this is like, this would have been a good first step. Okay, we're going to make improvements to Siri. Right? All that made sense. You can argue it's a year late. It's fine. It doesn't, it doesn't matter. They have time. But I really made me wonder why they announced all that stuff last year. What internal process failed that let them talk about a whole bunch of stuff that was just not going to work? Um, that's what I kept thinking about. But that's not really what matters. Right? If you just look at where they're standing, okay, good. Siri's a little bit better. It's, you know, it's good. Um, they need to do more. But they're probably on the right track. Yeah. So, so let's, so let's get into, get into some of that. So, um, what was interesting was kind of where they tried to draw the line with this relationship with Google, with base Gemini. Um, and then there's a couple of clever things that I actually think are really interesting that they're doing inference wise that I just want to tease as a, I think things in general may go this way for how models run locally and then across the cloud. Um, but there's, there's a couple parts of this, right? They demoed kind of this better version of just, uh, world knowledge search. They call this Apple Broad Knowledge or Broad World Knowledge. It's one of those. I can't remember. Either way, that's their term for it. And they're very clearly stating that that is not Gemini searching the web. I don't know if that's, they got some index from them, but they're pretty much saying like, we've had a web browser for a long time. We've been able to index the web. So we built basically our own kind of based search engine. And Siri can use that for you to just ask queries of the web and whatnot. So they call that Apple Broad Broad World Knowledge and they're really emphasizing that's there. So that's, they own that. They also have a handful of other foundational models that run on device that they're very clear. None of that was trained or using base Gemini. It might have Gemini outputs so that it works with the cloud version of the web. The cloud version of Gemini, which I'll explain in a minute, but anyway, distinctly trying to distance themselves from it being Gemini or a white label of Gemini. That's must might take away from the kind of executive technical session they did afterwards. But the cloud model is where there's the used some Gemini. It sounds like base, base model. And then fine tuned it, trained it all on their kind of data set to go and do all of this contextual stuff. Hook into the operating system, which we knew needed to happen for this for AI and assistant to really work well like it needs to have applications and system level hooks. Like that's just inevitable. And so they did that. They did that with their with their version. But what I found was interesting was why. Like, why did they feel like they needed to take, you know, us into this conversation with Craig Fittery and his team. And do what they did. Like explain how much of this is really theirs. And not and not Google's. And so I thought that was just kind of a. A high level kind of taken part of me wonders. Like did they did they do this for geopolitical reasons. Like did you need to do this so that China, for example, is comfortable letting Apple intelligence and Syria, I somehow into that nation even though they have no they didn't talk about it at all. We don't know how they're going to bring it there. They only mentioned the EU and that they're in this kerfuffle with the EU trying to bring bring this there but no mention of China. But my point is, you know, they they went out of their way to kind of really make it and emphasize that they did all this work. It's really apples. Even if it was based on a base level Gemini weight. And the analogy that Max on my team came up with and then I'll carry it further was, you know, think of it like it's TSMC's transistors, but it's apples design right or it's RIP, but it's apples silicon. That's kind of the way to think about they took a base level thing and made it their own truly their own distinctly and uniquely their own. And they really wanted to nail that point home with everyone. And so that's kind of my lob like why do it? I'm just, you know, posturing that there's some geopolitical maybe there's some just a lot of their consumers maybe are freaked out of Google right maybe there's and they needed to do that but regardless they felt the need to do this and I thought that was interesting. Yeah, it's it's it's the latter. I don't think there's a geopolitical angle yet. The coverage in China was very low just because it was a lot of interest. It wasn't none didn't get censored. It was just little. The focus being, you know, come back in a few months. These features aren't available in China right now. Right. And so the Chinese audience kind of wasn't didn't care much. I think it's very much around protecting the Apple brand against the Google brand because frankly Google is not seen as being privacy, privacy friendly, quite the opposite. And I'm not criticizing Google. I'm just saying the perception of the brand is a pretty big split there. There's a lot of a lot and growing concern of privacy, especially around AI. And so Apple, I think it's just laying the groundwork for hey, this is our stuff. Don't worry. It'll be just the safe as you expect from Apple. But they certainly spent a lot of time talking about their models in like I would almost say too much like right. It's just like and I personally at the end of it was more confused. I read a lot of your commentary and a few other things about it. And I was I was still like they're parsing the language in a very strange way to say like they kept getting asked like are these your models or are these Google. So they kept saying it's our model. But as we all know, there's the exact there's a lot of room in that definition. Yeah. What's our model. It's not Google's model. I my interpretation is they somehow have distilled or post trained some version of Gemini. That they can say is theirs that they it's all run on their servers. They control they don't share the data back with Google. And that's supposed to give us comfort that it's as private and secure as we have come to expect from Apple. What that means in terms of functionality, though, it sounds like it's going to be a step behind where Gemini is right and Gemini's gotten pretty good. I don't think this is going to be using whatever the latest version of Gemini is. It's always it by definition. It's going to be a few months behind because Google Apple is going to want to take their take a version and post train it. Yeah. Yeah. And I mean, you can already tell this like I tried a lot of time on the the beta and maybe this gets better. So I'm not discounting that the fact that I'm on a dev zero beta of of Apple intelligence. But you know, broad world knowledge is OK to be honest with you. If it stays like this, like I'm not not using Gemini like Gemini is still going to be the thing I go and use for Internet search and information. But to me, where this, I guess there's two parts of this. I'm not I don't think it needed to be on par with Gemini in search. I think where their opportunity is is again integration at the systems level so that you can say, you know, one of the best examples, which we tried like it is it is it is delightful. How well contextual search works now or their semantic search works now. And so like, you know, max sends me, you know, reports or Carolina sends me like a picture from an event and you literally can just search my message now and be like, what was that, you know, report on this and this and this. And it knows it like it brings it up, you know, it was it was remarkable that it worked right to be honest with you or, you know, I'm going to to Vegas for HPE next week and all my flight information is in my Mac mail. And I can be, you know, what's my flight information what time do I need to be the airport on Monday and it worked like that's that's pretty neat, you know, so just but to me, like those are the kind of things that people will do and enjoy and find useful and but in my brain, they're not thinking. I'm using this like smart intelligent agent slash assistant to go do these things it's just I want to do this right I want I want this outcome and it's kind of invisible but but it works and that's that's the productization here that I'm optimistic about. That they're delivering on which again, you know, you can tell me if you hate this term I've thrown it to everybody to me, this is like this is applied AI so in the same way that robotics is applied robotics meaning it's gone from the lab and now it's into the mainstream and being used for mainstream use cases, we know that most consumers aren't using AI for anything other than internet search that's the broad use case of call it the vast majority of consumers that are using it today. So this moves it to that functionality that job you want to do find this in my email find that song Jay sent me and open it in music or Spotify right it's such that so this is why I think applied AI is an interesting application where Apple can bring these things to the mainstream in very useful ways while at the same time it's not like a chatbot it can be but it's not necessarily because it has system level functionality that it exposes and so that's that's making it useful for the mainstream and at least the use cases I saw and have tried it's got a lot of potential to deliver on that concept. I mean, I think I don't even think we need a special term for it like I mean applied AI is great I think that's that it's very it's very descriptive but this is just what I would just call simply making a product out of AI right and it's something that's been lacking in much of the conversation of AI for four years now where a lot of AI is going to be used in the background it's going to be a feature if even that is going to be part of the code of an app to make some app some software better. And so if you wouldn't have talked about agents we're not talking about applications this is just a feature that's going to be the good companies are going to use just sort of without thinking about it this just me come another tool in in the developers tool kit. And I think that's important right I think those are really important things productizing AI in ways and. Ultimately, consumers have no reason to care about this as AI. They care that they can do things better than they couldn't do before. Whether it uses AI or whether it uses hamsters running on a wheel, doesn't really matter to a consumer as long as it works. I know that sounds pretty obvious, but believe me, if you look at what all the other AI labs are doing, this is not something they've come to grasp so-and-so. It actually gives me hope about Apple. We've been saying this for years, which is that Apple is one of the best position companies out there to make the best use of AI for the average consumer. >> Yeah. >> Great. >> Fixing search so you can actually search in messages is a huge value ad for consumers. People will love that assuming it works. >> Yep. That's why I did highlight the push that they may have even said this exact term, but to completely re-architect the index. The way in which your device is now label information, label things that were unstructured and unlabeled before, had to happen in order for what we just talked about to actually work. That's why I think this is actually interesting. The other part of this is obviously, they talked a lot about how AppIntense works, and obviously I think we've speculated that AppIntense is the thing that exposes some sort of an agentic operation so that you could say, for example, go check into my Southwest flight for me. Those functions are now there for developers to go and expose. I'll be very, very interested to see where that goes. Even similarly, this doesn't work yet, but I hope it does. People who have smart cars, I'll just give Tesla as an example on the day that we're talking about this on their magnificent IPO first-space X. Go start my car and turn the temperature to 75 degrees, or come pick me up, whatever it is if you're using century mode. That stuff, again, is it agentic? Yes. Does the person necessarily have to think that their agent is going and doing things? No, probably not. I think that's just the usefulness of this bubbling up to everyday use cases. That's where I think we're going. In relatively short order, and it sounds like the system, the platform updates that Apple has made is making that a little bit more of a clear reality than it was before. I'll take it a step further and say, if the consumer has to know that their agents involved, there's been a failure. Yeah. Yep. Agreed. This probably just goes back to the very early conversations and what our friend Benedict Evan said on this podcast, which is like, is AI really just a feature of all of this stuff or a dedicated thing? And the reality is what you're seeing from Apple Intelligence is that they're applying it just as a feature. It's just something you use that does stuff for you and you like it. For now, for now. I think this is true. Of everything is, for now, it's AI is going to be a feature in other applications, and we're going to a lot of gain from that. But I do hold out hope that we're going to have much more sophisticated AI-based systems. Those are coming. Apple may have them soon too. I mean, I have some clues on what Apple is working on. I haven't even talked to you about these. I have some clues on what Apple is working on behind the scenes. And I think we're going to see them in products next year that go much further. That's what we're saying. Apple is just like good progress, good for them. They should have been here a year ago, but fine. They're on track. And hopefully they and everyone else will come up with even better things. Yep. I mean, exactly to your point, if this was what Apple Intelligence was, it would have gone over really well, day one. That's what's making this interesting. It was a fun bowl fine. You made too big of a deal of something that nobody used fine. But if it was this man, people would have been like, "Apples run in like off to the like so much more bullish," then we are. But regardless, it's here. So we'll see. Okay. The other thing I wanted to mention, because this is just entertaining for anybody who has followed this. Apple has said words at this event that I don't know if they've said it a long, long time, at least two decades. And that word is in video. Yes. I did not, to be honest with you, I thought I was going to have to go into this and like, Carole, one of their executives, the thing is like, "Hey, are you running this on in video?" And they cram straight out and said it. They went out of their way to say, "We are doing in video and until for confidential compute." And they really in this session made a big deal of confidential compute, which to me was super interesting, interesting enough that even Roden article, or report about it yesterday, talking about the idea of protected tokens versus commodity tokens. And what that means as a cost tier for infrastructure. Because my default way of thinking about this is like, obviously, when Apple comes out and does something, it catches on. Not that everybody wasn't already thinking about private cloud compute and confidential compute. But for enterprises and sovereign nations and, you know, Apple's a great example of this, private cloud compute plus confidential compute, which is a really big deal. And those are going to be more expensive tokens. That's just the reality of it. But they went out of their way to talk about this because Invidia has very good confidential compute at the GPU level. And obviously, if Intel has to, the CPU is some involved in this and some inference, it needs to be accomplished. So anyway, all that to say, they made a big deal of other people's silicon and infrastructure, not theirs. And I thought about this and I was actually on, if anybody's a fan of Ben Thompson, I was on Ben Thompson's podcast the other day, and we talked about it this. And he was pretty adamant like, I don't think they need to do any custom silicon in the cloud anymore. And I was like, well, maybe, I mean, we know Apple likes to control these things. So maybe not right away or maybe they don't need to. Maybe they just let others do confidential compute. But regardless, I just thought it was interesting. They specifically named other people silicon in video, in particular, under the guise of being credible. I felt they did this to prove credibility. It's secure. You can trust us. We're running on Invidia's private cloud compute and we're using Intel. So it's like dual factor private cloud compute or confidential compute, a term I just made up. Anyway, you see what I'm saying? It's just interesting that they named this. Again, that they felt the need to specifically call that out. Some background for those of you who maybe have broader social life than some of us, Apple hates Invidia. Apple hated Invidia back to the dawn of time for, when I think the feelings mutual. Well, there was bad blood. I'm going back to like, like, I don't remember when, like a really long time. Like back to Steve Jobs first 10 year maybe. No, maybe not that long. Maybe I'm act time. It's decades. Apple needed something from Invidia. Invidia didn't deliver on time. Back when Invidia was, you know, still basically a startup. And Apple has was so mad. You know, the way Apple can get mad at us, the supplier, they were that mad at Invidia. And of course, Invidia, you know, took it personally. And so there's bad blood between these companies for a long time. And now, everybody's come to their senses and where they realize they're better off working together. But it's a pretty, it's a pretty monumental change. Yeah. Anyway, I agreed. Yeah, I thought that was super interesting. Turning of the tides. So we'll see, right? Again, I could see them continuing to flush this out with Invidia. People were like, is that why Google wanted to get infrastructure from SpaceX? Because they need, they got, they need more GPUs now that Apple is there. And I will say, like, it's interesting. This Siri beta, you know, you have to kind of get on a wait list when you're this early like we are. A lot of people have been on that since, since Tuesday. They're not rolling it out very quickly. And I do think a lot of that has to do with their monitoring the infrastructure impact to inference a lot of a lot of this at scale because almost everything that while it happens on, on device, at least the beginning queries happen there. It's going to private cloud. So they're inferencing a ton in, in the cloud, to pull off these features. So, so we'll see. But anyway, interesting, interesting developments will see how, how long these relationships, relationships hold. But Invidia's got another, another logo to show up and say, they power X, Y and Z. My, my money is on Apple continuing to use Invidia, but also at some point down the road does its own accelerator for inference. Yeah. I would, you know, I've got real money on that. Yeah. I'd lean that way too, but it's interesting regardless. And I don't think it's, you know, imminent year one or two, but it would make sense at some point. But yeah, they still got, but they still got a hosted summer like that's the thing they still need to put it in somebody else's Dave center because they're probably not building data centers, the scale in which they need to serve, you know, a billion and a half unique people and two and a half billion devices at scale. Oh, no way. Like I actually have the as of their last financials, their their cap X is minimal relative to what their beers are. So we'll know we'll have pretty good warning. But again, I have no advanced knowledge of any of this like I'm just completely guessing just based on why people do custom accelerators. I think Apple is a great candidate for it and eventually they will come around to it and we'll know because their cap X numbers will shoot up through the roof. Yeah, that's true. Okay. All right. Good dub, dub. We'll talk more about that as it as we try these things out. I'm sure. Another event happened this week. I went to Nebius's inflection conference, which really wasn't a news event. It was more, it's just might take on it. It was more of like a coming out party to a wider audience. I think a lot of people have been familiar with this with this company, but we haven't all sat in a room and listened to their executives till their story, talk about how they're positioned, talk about how they think about serving customers, the software platform. And so that's kind of my take from this event and they all executives, I talked to kind of positioned it that same way. Like, we just want people to really hear our story kind of broadly for the first time. And I thought it was good. We had some good conversations with both their co-founders, Arkadian Roman and a few other executives afterwards on kind of where they're seeing. I challenge them on some of the narratives that I've heard about them, right? That they're good on infrastructure, build racks, but don't own the power and land to the same degree. They obviously believe they're deeply going into power and land over time in the new data centers there. That is going to be 75% cell phone at some point in the roadmap. So, anyway, I thought it was good. I did last week a hyperscalers versus NeoCloud SWOT where I basically just looked at everything that we've been hearing from customers of all, let's just say six platforms. I didn't put Oracle in there, but so the hyperscalers and I/NNebius and CoreWeave and just stacked up their software skills against each other and infrastructure skills. And it's interesting. I did walk away from that thinking Nebius was the closest ish to a hyperscaler than CoreWeave and Iron and I still came from this event thinking the same. They're not a hyperscaler. They don't have everything. In fact, I do firmly believe that most customers will remain multi-cloud, meaning run some workloads out of hyperscaler and then run some workloads at a NeoCloud. But I think they have more of that software stack than others in my opinion to serve more workloads of that enterprise versus what they need in multi-cloud. So I thought that was positive, right? It helped reinforce some of the things research I'd been doing, but it was a good coming out party in my opinion for Nebius. Yeah, I think it was interesting. I was sorry that I missed it. I was still in China, but I really wanted to attend it. And I think one of the big questions hanging over all the NeoClouds is to what extent are they real companies and to what extent are they just landlords or short-term arbitrage in electricity and GPU availability. And you go down the list of whatever 200 plus NeoClouds that are out there and only a handful really look like they have sustainable long-term business. And Nebius is high on that list. And CoreWeave both have taken great pains lately to say that they're building long-term businesses that they want to be public clouds like AWS or Azure. Both of them seem to have internal capabilities that would allow that. We can debate, I'm sure we will debate between the two who's better positioned, but those two really stand out in the crowd. And not for nothing, Nebius has been a tier two hyper-scaler for years. They used to be Yandex. Not a lot of people here use Yandex, but it was a legitimate, technically very capable. They have some very, very smart engineers. So they've been in this business for a long time. The question is they've obviously not pivoted very hard to AI. And for the time being, both CoreWeave and Nebius' revenue is dominated by those sort of real estate landlord-type businesses. And I think what Nebius here is trying to do is by holding a conference in San Francisco where the developer community is very deep. They're trying to market their sort of non-AI hyper-scale businesses to the developer software community. And I think that's sort of the goal of the event is to let people in the software world know how you can use us for more than just $20 billion hyper-scale deals. We have a real suite of software offerings. Yeah. To their credit, they had a lot of developers there. I mean, I probably talked to a dozen of developers who were there either existing customers or people who were there starting to talk with their teams about how to use them. There's also this very interesting other thread. They have a team dedicated to fleshing out how they'll serve the business. And that story wasn't told here yet and is not, but it's bustling. And I actually think that's really interesting because, again, if I am a robotics company, I probably don't need to worry about my AWS software workloads and my ERP systems the same way. I need to serve my robots at a zero latency as possible. So I think it's interesting to think about the NeoClouds and them in particular as a physical AI infrastructure play also when that market starts to emerge. The other thing that I came away with was I started probing some of their technical folks on, tell me about these racks you build. And it's interesting because their logic was they think they could build a more efficient rack than a typical OEM machine. I don't know if this is true. I'm just giving you kind of the way they think about it, right? The way they organize the networking, the way they run the wiring or the way they handle the power. This is the Russian engineering, coming to play here. They really believe they could design these things better for their power needs, for efficiency, for latency. I didn't get details of how, but that's kind of like this is what we think about this is why we do this. We are finding efficiencies to build and then bringing video components into it. So I thought that was actually very interesting. Again, people who are out there, the retailers, like loving Nebias will be like, they build their own racks. That's why they're amazing. I wanted to know why. And so I don't know if anybody got why, but that's interesting to me how they think about it. And I do think there's some truth to that, but I actually liked that as a strategy for them. What other was I mentioned, things we've talked about before. Are you starting to hear demand for CPUs? And they said yes. They are preparing to scale not just Vera, but Intel CPUs as well. And they look at multiple vendors, but they're starting to become a provider for dedicated CPU racks also, for inference. So it's early, but knowing that they're going to be there made me a little bit more optimistic than like, no, we don't see the need for CPUs. So that was positive. Go ahead. You want to say something? Well, it's very interesting because on the last earnings call, three weeks ago, they got asked that question by myself, included several times and gave us a slightly different answer. Yeah. So that's I don't know if anything's changed, but regardless, they were very clear. They will deploy Vera CPUs. We did ask, are you looking at other vendors? Right now, they're obviously largely Nvidia. I imagine they stay that way since being an early access and priority supplier for Nvidia is one of the tenants they hang their value proposition on. So I would expect it not to change a lot, but they're scaling. I think they said four gigawatts line of sight. I don't know the exact timeline for all of that because power contracts are weird. But okay, here's my point. I don't have lengthy conversations with CoreWeave. I don't have lengthy conversations with Iron. I have lengthy conversations with the hyperscalers. And I always come back to like, I want these people to prove to me in my conversations that they also have technical shops because that's relevant in AI data centers and infrastructure. And I walked away with that impression from Nevious. And to me, that's a bar, that's a filter I use. And I came away feeling pretty good that they actually have some technical shops. And to me, that is worth something, right, competitively in the analysis. - Okay. - Maybe it is. - Okay. Any other topics from this week, 'cause those were the two biggest ones, top of mind to me. - Oh yeah, you were in China. - I was in China and like-- - Educate us. We're robots everywhere, Jay. We're robots walking down the street, shaking your hand, handing out pamphlets, doing a song and dance for the store at the end of the corner. - No, no. - Okay. - They have robots to do delivery of room service and hotels, but they've had that since COVID. - Okay, yeah. - And they're not humanoid robots, they're all tracked, wheeled things. I didn't see a ton of robots in the streets. I saw a lot of things. I was there for, I was tied back for like three days and I was in China for 10 days. And we could just as well call this segment, Jay's fever dream trip through China. I saw, I was in six cities. I saw, I don't know, 20 companies. And it's China. Like I would say, overall things there are pretty good. I think one of the big takeaways I came back with was the entity law, the entity list, US restrictions are stupid. They're failing. They're absolutely failing. - Okay. - When the entity list first started five, six years ago as ZTE and then Huawei and has expanded ever since, there was a lot of fear in China. Like just like, I had a few people describe it as panic. I don't think it was that bad, but there was definitely a lot of fear. Everyone was really, really frightened of ending up on the entity list because it seemed so arbitrary. And so everybody's companies that were really, really worried about it. And today, nobody wants to get added to the entity list, but nobody's afraid of it in the same way. We've had a few high profile companies who were at near, near death, like YMTC because they were on the entity list. And they've muddled through it and now seem to be thriving. Here, I actually bought some YMTC memory. Five bigs. - Nice. - Yeah, around 12, baby. - I, I, right. And so the entity list has no longer having the psychological impact it has. I met with multiple companies on the entity list. They were very happy to talk about it and lament the fact that they were on it, but they didn't really get too worried about it. Just like the more you dig into it, the crazier it gets. Like, there is one company I spoke to who was on the entity list, partly for their ties to the PLA and partly because they were accused of participating in something pretty, the pretty bad, just accused. We don't know if this is actually true. So they're on the entity list. They've been on it for a long time. And I was, I met with them. I've met them a few times and they had, I was talking about a few things and we got to the subject of where they're getting their components from. And I had assumed like many companies in China they're moving towards domestic parts. Right, that's a big trend in China moving towards domestic suppliers. And they said, no, not really. We still get 90% of our critical components from a list of US companies. And I'm not gonna name them either, but it's all companies we've talked about before and you and I have heard of major US tech companies across every part of the industry. And so US companies cannot buy from this company, but US companies can't sell to that company. And I just like, I don't, at that point, I just don't understand what the entity list is there for. It's just, it's like, and then like as I was getting on the plane, there was a new bunch of additions dropped to the entity list. - I haven't seen this. - Oh, it's just like, it makes no sense. Like it's just like, it's pretty much every large Chinese company is getting added to the list now and there's no clear rhyme or reason. So that's my rant. We need to rethink how we're doing the end end end. So also on top of all this, like the entity list is supposed to be enforced by the BIS, the Bureau of Information Security, whatever standards, Department of Commerce, BIS. How many BIS inspectors are there in China, do you think? You wanna guess a number? - I mean, I don't, 'cause I don't know, but is it small? Is it like a handful? - It's five. - It's five. - Yeah, it's five, right? - I think one handful. - Yeah, it's five. And they're not allowed to basically do anything. - Geez. - Right? And so it's like what, you know, the whole thing is just like, needs to be rethought. Anyway, sorry, that's my rant. I did see, let me talk about something happier. I went to go meet with a company called Asprasif. Right? You will appreciate this. This is really interesting. Asprasif makes like Wi-Fi modules, right? And I know when I first signed up, I was like, oh, Wi-Fi modules. I've met so many Wi-Fi module companies over the years and they're also boring. Like, it's just a really tough business. Like, I've met dozens of these companies over the years, right? And, but I want to go see them, 'cause I'm curious, they're a fairly large company. And they have a nice office building out in Pudong. And they showed me, they sort of, they walk through their IR deck. And the first slide was like normal, like, here are all the products we do. Right? And I thought, okay, the next one slide is gonna be, like, we're gonna deep dive into each one of those projects, products, like, here's the one with Wi-Fi, and here's the one with Wi-Fi and Bluetooth, right? No. They put up the product slide for like two seconds and then immediately moved on to their slides about social media, right? This is all a presence on social media. And then they're like, this is our presence on Reddit. This is our presence on YouTube. This is our presence on GitHub, right? And we spent 30 minutes talking about all of their social media activity. And this is a chip company that knows how to market. All right, and I was like, yay, see, I thought this would interest you. It was really, really, it was fun to see, because, right, just remember, like, they have a huge number of YouTube visits, a huge number of Reddit posts, a huge number of GitHub commits for their open source projects, right? Just remember, in China, those are all blocked, right? You can't access Reddit from China. You can't access YouTube, and GitHub is very, very constrained, right? But they have this big presence on all those things, because of the work they've done to market to global users of their product. And I thought this was really fascinating, because what they really, what their main market is, IoT, right? And like, as you and I both know, and we've probably talked about before, IoT is just, it's a hard market because it is, everything is sub-scale, right? There's no, like, you can't get, you can't get to scale here, you can't, you know, you win a big customer, and it's like a million units at most, right? - Yeah. - It's always been the problem, both selling into it, but also for analysts covering it, it's just too hard to track. - Yeah. - And they have solved that problem by making this basically a long tail user-generated content business. They do all their marketing through user-generated content on YouTube and Reddit. And their whole thing is they have a common software set called ESP32, which is, you can buy books about it, you can buy USB to ESP232 kits on Amazon. And so they've solved that sort of fundamental marketing problem in IoT. And I just thought that was pretty, like, you know, I thought that was fun. Here's a chip company that actually knows about marketing. So if you wanna do, be 32. - I like that approach though, but seriously, like, that's smart because again, right? To your point, like, there are a lot of companies out here who are going to use those types of solutions for their IoT, not just the US vendor. - That's right, that's right. Like, you know, I know lots of US chip companies have sort of talked about software. I know one chip company who's, who I will not mention, but who CEO, one of the terms of his compactage was how many developers they get on the platform. And like, they, like, but they don't actually, none of them can actually market to that or sell to that. And here's this interesting company in China that has done a really good job of it. - Okay, I have a very important question for you for China. - Yeah. - Are there burritos yet in China? And I ask this because the last time, and this is a long time, almost 20 years now, that I was in China for 10 days. By day four, I needed a burrito, and they didn't exist. - All right, like you gotta remember, when I lived in China, we were really, really excited when the first Thai restaurant opened. - Okay. - Right, so I will say that like many things in China, the diversity of cuisine has improved remarkably over the last 20 years. Let me also, let me also phrase this appropriately by saying, I'm not gonna say that I'm not gonna say that, somebody who doesn't like the burritos I get in New York City, right? Or most parts of America, sort of. Grant, I'm not asking about the quality of it. Just asking if they exist. Yeah, there are Mexican restaurants in China. The quality of the burritos does not live up to my admittedly very high expectations. I'm with you. Yeah. They would not for me either. But to do J at that point when I was five days in, I would have taken anything. I literally would have taken any burrito other than duck and noodles and vegetables. As good as that was for four days, I loved it. But then I had a I had a threshold. Yeah, I had a lot of tremendous Chinese food while I was there. It's very good. But I will also say the only time I ever eat KFC or McDonald's is in China. For diversity. Yeah. Just because like, yeah, just something else. I hear you. I hear you. All right, everybody. That was our show. One more thing. On the subject of robots. I met a company called LeaderDrive, which makes actuators for robots. Yeah. And this is a company that started life as a machine tool company and still does machine tools. Like this is this is a business that is very much about cutting pieces of metal at very, very high specification. Nice. And there's this whole big debate now about how to do actuators for robots. You do strain drives, you do world planetary drives. All kinds of interesting debate taking place there. Lots of robots coming on stream. But if you add up the total capacity of LeaderDrive and the biggest competitor, HarmonicDrive from Japan, their total capacity is probably about two million units this year. Next year, it'll be about four million units. They're going to double, both of them are going to double. But every, every humanoid robot needs at least 20 actuators. Right. For each of the joints, multiple batteries. At minimum. That probably goes up. But yeah. Right. So you sit there and say, four million units divided by 20 is 200,000 robots next year. Right. That's, that's, yeah. But that's the global capacity of robots. And you know, my point being is we still have a long way to go. And you know, both these companies are doubling capacity this year and probably next year. But we still got a long way to go before we can start talking about millions of units of robots. A great, a great robots will be fun once we start talking about them more. Oh my God. I'm looking forward to it. So much fun stuff in there. But we've got so much to talk about still about AI and friend, what not for the time being. Okay. Anyway, thank you for listening. Everybody that was our show and we will talk to you next time. Thank you for listening. Have your agents click like and subscribe. Tell your robot to listen as well. All the views. Okay. Bye. [BLANK_AUDIO]