Sharp Tech with Ben Thompson - Generative AI and Hardware Upgrades, Google History and the Dark GPU Theory, Amazon’s Logistics Long Play

Episode Date: June 3, 2024

A question about generative AI and the hardware powering voice assistants, projecting the GPU future with the dot com bubble as context, and thoughts on Amazon’s capital expenditures and the state of AWS and Amazon Supply Chain. At the end: An Elon hater reviews the Cybertruck.

Hello and welcome back to another episode of Sharp Tech. I'm Andrew Sharp and on the other line, Ben Thompson. Ben, how you doing? Doing okay, Andrew. How are you? I'm doing well. I had a great little weekend here in D.C. We're really in the sweet spot here in D.C. How's the weather in Taiwan?

It is getting hot. There is not much of a sweet spot here. Not that there's a great one in D.C. either. But I'm glad you're enjoying it. Yeah, that's why I'm cherishing it because it's going to be. be absolutely disgusting in a matter of weeks, but I did my best to touch a lot of grass over the past few days with temperatures in a reasonable spot. But here we are back again to Pod. And this is going to be all mail today since we skipped 90% of the questions on the last show. So thank you to everyone

who has continued to send questions to email at sharptech.fm. And Ben, we're sticking with AI at the start here. Grant says, I've been surprised by how long it's taken for the incumbent voice assistant Triumvirate, Google Assistant Alexa and Siri, to reach mainstream news and analysis around LLMs. All three companies now seem to be openly admitting that each will get some kind of generative AI aspect added to it. Seems like a no-brainer. My question is about the tech. Is it purely back-end work for these companies to add generative AI to their assistant? or will we all end up needing to upgrade hardware at some point? My understanding has always been that the vast majority of the processing happens in the cloud,

as anytime my Wi-Fi is out, they are unusable, but maybe I'm missing something. What do you think, Ben? Yeah, I mean, the processing does happen in the cloud, so theoretically you could just drop in a better model. Need us to say, I think that's a little, that's pretty complicated for a few reasons. The first sort of obvious one is doing this sort of voice processing at scale is difficult and expensive. Like no one has truly rolled this out. We saw the demo from OpenAI of this sort of interactive voice capability.

And by interactive, I mean it actually is like responding to you. It's not a let me open the app, press a button, wait, record something. It gets transcribed, then sort of responded to, then transcribe back. Like that's not usable for a voice, like the whole idea of these voice assistance, these devices in your house is they need to have that sense of immediacy. You can't have just massive amounts of lag. So that's sort of number one. And so again, Open Eye hasn't shipped yet. Google's demoed it shipping sometime next year.

And this is just for sort of the apps on your phone. It's still sort of a step further to in devices always on all the time. So number one, just getting it to work. Number two, the cost, I'm sure, is going to be very large. And so all these are considerations as far as turning it on. So I don't think it's fair to say anyone is behind at this point, given that no one's actually truly shipped this capability. And when you say cost, you're talking about the cloud computing cost, right?

Yes. Yes. Yes, absolutely. And to sort of grants point, I don't think this would need a new device. The way it works generally, as I understand it, is the devices listen for a trigger word, which we won't repeat here. so that people's devices don't go off. Yeah.

But then after that, everything is streamed to the cloud. So the devices are fairly dumb. They're focused on. And that's why you can't change the words, willy-nilly, because they're programmed for a specific word, and then they start listening and sort of sending out of the cloud. So one of these companies, when you choose a name, you have to stick with that name for life at that point.

Well, and I think that Amazon app offers a couple options, but by and large, it's only one option or there's one or two options. So that's why, and that's why. it's not sort of a willy-nilly choice. The second thing is, and all these companies are in a little bit of a hard place, which is they have built up functionality around the current way that it works. Now, the way that it works is fairly deterministic. They sort of programmed a bunch of questions and answers.

Yes, you can go off script a little bit, but they have to sort of figure out what you're talking about so that they can sort of slot it in and give you the right answer. And that's why if you want to do things, like if I want to open the curtains in my house, I have to do it as like an incantation in like the certain words in the right order. I'm about to go back to the U.S. where I actually have this functionality. And every time I go back, it takes me like a couple of days to remember, wait, what's the right incantation to open my freaking curtains, right? Yeah. And so, but the point of that is all that functionality that's been done has to be redone.

And so, like, even though an L.M is obviously way more capable as far as holding a conversation, it's not by default more capable as far as actually doing the stuff people are using these devices for. Right. Now, you might sit here and say, look, the market opportunity for having an LM that is amazing and you can talk to is way larger. You should just scrap this and start over. But these companies can't just scrap and start over. People have built their houses around these devices, right? People have made investments around it. And, you know, it's sort of a, you know, I don't think anyone's going to come along and disrupt them.

But it is an example of how you can sort of get stuck in a certain technological dead end. And you have to, it's a lot more work. Now, can they do, you know, adjust LMs and build them up so that they work even better for opening my curtains? Absolutely. I'm sure they can. But that's going to take time. It's going to be pretty difficult. and there's going to be a lot of edge cases.

And I predict a lot of consumer frustration when they do roll out because, you know, they've tried to have these standards and anyone can build to them. And who knows how well all these various entities have actually built to them so that when you change something, this sort of fragile connection of apps and APIs and stuff doesn't sort of fall over. I expect it to go this direction. This is really interesting. Honestly, when Grant called it a no-brainer in his, email, I was nodding my head, but it sounds like there's a couple more layers of intrigue in terms of

how and when this actually gets rolled out. Yeah, I mean, it's probably a good example of, you know, my old axiom that most people just assume companies are stupid. And actually, if you stop and think, there's probably really good reasons why stuff hasn't happened yet. And yeah, I think it's going to be a pretty heavy lift to make it work for sure. Yeah. Well, and then on the Apple front, Mark German reported that Apple's planning to overhaul. Siri and that Siri will be powered by more advanced AI going forward. But the final line of his report was the real takeaway for me, which was Apple is betting that the new AI features for the iPhone, iPad, and Mac, as well as the Siri enhancements will encourage users to upgrade their devices. Many of the on-device AI capabilities will require an iPhone 15 Pro or later to work.

And that reminded me of our conversation on the last episode where I asked you whether Apple really needs a great AI story. And I didn't fully articulate why I think they don't, which is just that I don't know that consumers are at a point where they want AI that does anything more than retouch photos or create stupid emojis or whatever. So if Apple doesn't have anything transformative or groundbreaking to introduce at W. I don't think the market is going to penalize Apple as much as tech media might. But that being said, the idea of needing to upgrade hardware to access the full AI capabilities has always been the simplest way for Apple to win with AI. And it's at least a better selling point than last year's big titanium push around the new iPhone.

So that that piece of it makes sense. No, I've got to raise an objection. I love titanium for the iPhone. I still appreciate how much lighter it is. I don't notice a difference. I mean, to be honest with you. It is definitely lighter. There's no question about it.

I put this in category Ben's travel tips. I do always carry my old iPhone, like the old iPhone with me when I travel. Because if my iPhone stops working, it's a big problem. And then it's already has all my stuff on there. It's all ready to go. Yes, I go to Apple Store. a new one. This is just easier. I don't want to buy a new one necessarily. So I was carry one. Yeah. And every time, every time I'm getting ready to go and I grab that,

but check that iPhones in my bag, I'm like, wow, this thing's heavy. This is a freak. I can't believe we used to use this stuff. Barbaric. Fair enough. Agree to disagree on titanium. But but, but to this point, to this point, it is a very important one. And this is a point I've been assuming for a while and sort of predicted that a lot of Apple's AI features would require new hardware. Yeah. And that is the upside story for Apple from AI, that they have a selling point for new hardware. And I think it's legitimate, right? Like they, that they, their new devices with their new chips.

Now, again, the MacBook Air only has the M3, which is sort of the equivalent of the, what's in the iPhone 15 Pro. So that's just the cutoff is basically going to be any device bought since that iPhone and sort of going forward. But that's fine. Most Apple customers don't have the latest devices. This is a sort of a strategy credit where I think it's totally legitimate that you need to have a newer processor to do this on-device stuff. And also Apple will make a lot of money because they now have a selling point for getting a new device. So could they make it work on old ones? Maybe they'll probably get pilloried for it in the press.

But I think this is a legitimate sort of thing. And again, this is a distinction just to go back to Grant's question. our assumption around Apple's goals is that they want to do as much on device as possible, you know, as opposed to just pushing it all to the cloud. Now, this, in the case of new hardware,

for Siri would almost certainly require a new home pod, if that is one of the pieces of hardware you're considering. Okay. I can't remember what the hardware is on it. It's a, you know, I think it might be watch level hardware, at least on the minis,

maybe not on the big ones. I don't have it in front of me. But that, yeah, that is a distinction there where Apple wants to do as much on devices they can, hand some stuff off to the cloud. And, you know, and maybe that's the way it will work, like stuff that. And the other bit about here, just sorry, just to, there's a couple of things points in here. Mark German is talking about controlling individual app functions. This is the bit that, you know, that's interesting about Microsoft and this ties to the Apple thing. If you own the device where the apps are, you have way more information about how users use their devices.

And you have way more access to the way people do stuff by virtue of owning the device. And you could potentially do way more interesting things. So this is number one, something I expect, but number two, something that I think shouldn't be underrated. The extent to which, you know, like what are the challenges of building a new platform? We know, platform, platform, platform, is you need developers to, you know, adapt their apps to your platform. And, you know, that, that requires work on hands of the developer. But to the extent you can just leverage the GUI, leverage the apps, because your model understands all the interactions on a device.

And you can more easily gain capabilities by virtue, by basically acting like a virtual user because you control, you control the graphics layer. You understand how all this sort of stuff works. you can do some potentially really interesting stuff, and it's one of the reasons why Apple and Google and or Apple and Microsoft and Google to an extent on Android, probably more so on pixel devices than other ones, do have a potential leg up in actually leveraging AI getting certain platform effects for free where they can just make it work without developers having to do

a bunch of sort of building stuff out, tying stuff together. Now, developers have been doing Siri intense and stuff for a while, so maybe that's an angle as well, is something that is interesting to look forward to. Yeah, well, and it's a very coherent story to tell consumers as to why they need to upgrade. And the bar is pretty low as far as Siri functionality now. So maybe AI can improve things. The other question that came to mind, this idea of needing to upgrade hardware to access optimized AI,

it dovetails with a question that's sort of looming in the background as all these giant companies spend like crazy. and Nvidia becomes the most valuable company in human history. We're not there yet, but I'm sensing in the ether some impatience as to when exactly a lot of this spending is going to lead to profit. Do you have any takes on that question? Where is this impatience that coming from? Is it coming from the mind of Andrew Sharp? Is it coming from Charlie? I need a source.

No, from the outside, it's just a staggering amount of money to be spending every quarter on AI, hardware without a clear path to new products or new revenue streams. And I'm beginning to hear on podcasts and stuff like that. Like, this can't go on forever. So I just, it's a very general question. But have you sensed any of that? I mean, do you do you think that there's a point where there will hit a tipping point? Yeah. Well, there, you know, there's a lot of angst, I think, with last quarter results with all the sort of capex that was going on. There's been a few interesting dynamics of that capx spend that's been interesting to observe. Number one, Nvidia is basically gobbling at all.

AMD is forecasting that they will be demand constrained in the second half of the year, not supply constrained, which suggests that to the extent there is a hole in spending, particularly as Nvidia brings sort of new systems online, that hole is being filled by old Nvidia chips, not sort of competitors. And so the, yes, and there's massive amounts of expenditure. And there is a bit of a, it's to Nvidia's benefit for now. this sort of catch-22 where all these platforms feel they have no choice but to invest because this is the next big thing. But also it's not clear when and where that sort of revenue is materializing. Now you talk to them.

They're very optimistic. Microsoft will say co-pilot is ramping faster than any enterprise product ever, like the sort of uptake, all these sorts of things. You know, that's probably the clearest revenue generating product in the market right now. You know, chat GPT makes a few billion dollars a year. year, I believe. So there's, you know, there's, there's definitely products in the market, but it is not commensurate to the amount of spend. The amount of spend is on the assumption, there's more coming. And yeah, at some point that does need to materialize, you know,

one of the interesting things broadly in a point I've been making for a while is when you look back historically, we've been in these situations before, and the overbuild does tend to happen. And when it happens, it can be sort of very devastating. But, But it could be very beneficial, broadly speaking, because suddenly we have a ton of GPUs everywhere that could be used for inference. We keep talking about cost, cost, cost, cost, costs. Well, when it doesn't cost anything because they're already there and they just level off his playing field. Yeah. Yeah.

Then the possibilities sort of get much larger. Well, to that end, we got a question from Michael. He says, Ben and Andrew, one of Ben's key arguments surrounding Google's growth was their forward-thinking maneuver to purchase a significant quantity of dark fiber in the wake of the dot-com crash. Can you provide some more detail on the circumstances surrounding this? I was too young when the dot-com bubble burst to understand what was happening. The reason I ask is because I've been bouncing around a theory that I'll call the dark GPU theory. It argues that when the AI bubble burst, there will be an overcapacity of GPUs that will be

either dormant or underutilized. A forward-thinking smaller company could buy up and leverage these GPUs to great effect. Is this too simplistic a theory? One major hole I can see in this argument is that the major players in the AI space are so well capitalized that even in the event of a significant market downturn, they could still go ahead and purchase up all this capacity in a defensive maneuver to prevent competition from rising. What do you think, Ben? I love the idea of the AI bubble bursting and then creating a bunch of opportunities for startups. But his point at the end is worth considering why would

any of these companies allow startups to buy up GPUs on the cheap that way? And Google was the company. It's not like Google was a mom and pop shop when they bought up all the dark fiber. So how could you see this playing out? Yeah, well, number one, Michael, I'm going to give you massive credit for the name dark GPU theory. But number two, I'm not going to give you credit for the theory because I've been talking about this for like two years now. Must credit short tech. But yes, this is the bit sort of I was driving on. So when it goes back to the dot com bubble, I do recommend, I did an interview with Craig Moffitt, who covers the telecoms industry.

And we did talk about the dot com era a bit. I think that was last fall. We'll put a link in the show notes. But I thought that was a great interview. We got into some of the history of some of the drivers of this. And you basically had ruinous competition for an asset that the marginal cost was sort of going to go to zero. Like when you build out all this fiber, the long run, it's going to cost basically zero to transport it. So if you sketch that out and actually Craig Moffitt made his name by basically being a voice in the wilderness predicting like all these companies are going to go bankrupt, which is going to.

I have a tough spot to me. It's like I'm going to analyze this industry and my takeaways, you're all screwed. Yeah, not looking great for any of you. Yeah. And so, but, but yes. So, so the dot com bubble, it was two bubbles and people forget about this because number one was all the internet companies. Everyone sort of knows about that, you know, like the, the, the, the, the, the, the, the, the, the, the, the, the, you know, what was the shopping one, web van or whatever it might be. And, uh, and all those got crazy evaluations. They would IPO before they had any product or whatever. and they would skyrocket and then it sort of all blew up. That was one dot-com bubble.

That did have a big effect in Silicon Valley. To some effect, we're seeing a little bit of the same overhang. Actually, the last couple of years, there's kind of was a SaaS bubble that burst to a certain extent. And that's why there are, you know, like there has been a significant decrease in employment. There has been sort of a tough job market. It's not as bad as it was in the early 2000s, but it's, but it's, been, it hasn't been great, uh, for sure. And, and that, you know, there is some sort of comparison point there. The other bubble was the telecom bubble, which was you did have this, this fiber

coming online. You, you would have the breakup of, of the mama bells. You had these new entrance to the long distance market that were wanted to do sort of like voice over IP or they could like transfer voice via packets instead of like copper lines as it used to be. And you had this massive buildout of fiber sort of all over the U.S. in particular, all over the world. also, but particularly in the U.S. And it was like, if we lay this fiber, we can have this great cost structure and we can undercut everyone else. And the problem is everyone was doing that.

And you couldn't not do it because you'd be screwed. And so there was just this massive spend, WorldCom Quest, like these other companies that were just spending, spending, spending to build all this and then they all went bankrupt. So you had these massive amount of capital costs that you couldn't make money on because the prices fell through the floor. And so this definitely helped Google. It wasn't causal to Google's growth, to be clear. Google would have been a very large and accomplished company without it,

but they sort of IPOed in 2003. They had all this cash. They were already making lots of money by that point. And to help build out their worldwide infrastructure, which is the best in the world, there was all this fiber that they just bought up. And so they could have these data centers spread all over that were linked at the speed of light. and they didn't have to share the traffic or bandwidth or whatever. They weren't going over the public internet,

which could be slow and latency filled and all these sorts of things. And it was, they got it at like dirt cheap prices. So it was definitely. Stupid normie question. When you say help Google, it just helps Google what, serve search results instantly and just make for a better product?

Yeah. I mean, yeah. And also, you know, there's, you know, as Google has gotten larger and larger and the,

the computing has gotten more and more scaled and distributed. The extent, And even today, you see this with AI. Like one of the, one of the, and this gets to the Google's integration that I've been talking a lot about, they have built up their, their, their infrastructure so that they can actually do training across data centers. Now, I'm not sure to what extent they're actually leveraging that yet, they've been talking about this. But when you're talking about these astronomically large models, to the extent, like, one of the constraints is can you literally fit all the GPUs in one place? Because all these GPUs have to communicate with each other. Because everything, it's like this multivariate to the extreme calculation where the

GPUs are communicating with each other and they're like, and it's a serial process. It's sort of a great irony where this is a parallel sort of computing problem, but it's a serial process where you're going like step by step through these layers and everything has to be sort of in sync. And that's why Jensen Huang talks about systems as opposed to a single GPU. So that system can be like eight GPUs in one rack that are all operating as one. It could be an entire rack that in like their new Blackwell one. It's like copper wired, which is even better than sort of like like, you know, the best possible sort of communication.

And that's one unit. It could be an entire data center. And Google's talking. No, it could be like multiple data centers. Yeah. Okay. And you think about that because the real constraint in computing has been in all. always will be the speed of light, which is pretty crazy that that is actually an actionable thing, but that is critical.

Like, that's the absolute maximum rate you can communicate information. So when you're dealing, if you want to say have multiple data centers that are doing a parallel process that happens in a serial manner, the weight of sending information back and forth to sync everything together and go forward is a very, very hard problem. And Google is probably the only company that could do this. Like even like you talk about Microsoft or whatever building in the, you know, N plus three or whatever opening eye models, this is where the talk of like a gigawatt data center, you build this massive thing in the middle of the desert with a bunch of solar panels or something like that and batteries. Like that, that is why because of the sense you need to have it all in the same place. And so, anyhow, the long and short of it is that networking is a hard problem. It's a big deal.

It's coming back to the forefront. Oh, one of the interesting things about invidia lately is the degree to which one of their. brilliant acquisitions a few years ago. They bought Melanix, which is a networking company. For a long time, like the big winner of the dot-com era was actually Cisco, which provided the networking. They were the dot-com baby more than anyone. They went the highest.

And then plunged the furthest. We actually got for this middle period where networking was commoditized. You would actually just use commodity hardware. It was called software-defined networking. And the software would make the, like, just you could make whatever router you want because it was all software defined, as opposed to being like all hardware done. One of the interesting things about AI,

because the networking is so expensive, is we're seeing a return of dedicated networking hardware. That's a big thing that Nvidia is pushing, and they're pushing it not just for their proprietary technologies, like NVLink and, and the other one is escape, Infiniban, but also with Ethernet.

Like, they're doing a big push on Ethernet networking, which is why? Why is Nvidia, like, they have all this specialized networking? Well, because Nvidia, wants to make sure their chips will work everywhere. And data centers today are constrained by our, are on Ethernet. And now this is like crazy Ethernet, like much higher speed and capability that we have

in our houses. Okay. But. Yeah, you wrote about it. And I was like, Ethernet doesn't sound like the technology of the future. So it's good to know. No, absolutely. No, it's actually one of the oldest technologies.

And it's not going anywhere. It keeps sort of getting evolved. It's pretty interesting. There's different places to put computing and networking. Is it like within the network itself? is it on the edges. That's where Ethernet differs from something like Infineban. But there is some degree of sort of convergence happening there.

But the long and short of it is that networking is actually very difficult. It was kind of a, it was still interesting, but it wasn't that interesting for a long time. And now it's really interesting again. And, yeah, the whole thing about Google buying fiber was about networking. That was back when it was, you know. And so, in a year, the long and short of it is, yes. All those companies went out of business. All this fiber was available.

It really accelerated the fiber build out. Like the great outcome of this, the real thing that made the internet become a meaningful part of people's lives was broadband. It's a, this gets to the AI bit. Sorry, I'm kind of rambling here. There's like five gazillion threads here. Well, there's history and then the present and the future that we're trying to weave together with the answer to Michael here. Yeah, well, the analogy here is I talk all the time about this voice interaction. it's so different to just talk and get a response than it is to open an app, invoke the voice thing.

Like, and it sounds stupid, but it's actually a big deal. A perfect analogy is the difference in usage. And this goes back to our last podcast. When I told you, like, look, you're underrating how much potential usage there is here was the shift from dial up to broadband. Even if, like, actually, the big dream was to have a dedicated phone line so you could have have your your computer always online. That was slow. Dial-up was slow, but the difference in usage and sense of capability of your computer,

if your computer was always online versus if you had to go through the 90-second dance of like connecting to the internet. Thank you. I was. No, it's true. I was an early adopter, so I was happily using Dial-up as a sixth-grader. But I feel like more normal people and in enterprise settings, broadband sort of accelerated everything and integrated computing into basically everything everyone did. Certainly at work, but also at home, it just became more normal for everybody once that hit the scene.

So once everyone got broadband at home or in their offices or whatever, guess what they did? They used the internet way more. Yeah. All the time for everything. And so that's why all the dot com, the web company bubble, all the ideas were right. They didn't make sense in a world of dial-up. Right. In a world of broadband.

Totally different story. Well, not just that, but like anything. Like you're just, when you're just using the internet way more. And then the other piece of this, this goes back to my end of the beginning article, it wasn't just continuous connectivity that it was having the device always with you. Mobile was the other missing piece where you had broadband with you sort of all the time. but all this was essential to actually making the internet realize its potential. To tie this all back together with the GPU thing, yes, the big question, Michael, with the dark

GPU theory, this fits in with sort of the Carlo-Perez theory of technological revolutions, which is you have this crazy mania period, and then you have the crash. and that's actually essential to new technologies precisely because you need a period of insane, unjustified investment to get the necessary infrastructure in place so you can actually build viable things going forward. Again, the dot-com bubble is a perfect example. The telecom investment in broadband was insane.

Go back and read the Craig Moffitt interview. He'll talk through it. It was nuts. It made no economic sense. And yet you had all these factors driving these companies to incinerate billions and billions and billions of dollars that ultimately laid the infrastructure groundwork for the Internet and the sort of 20 years of moneymaking that followed.

In this case, the theory is, yes, you have all these drivers and all these companies that governed by very smart people that can't resist or ought not, or it's impossible to resist. If you're Microsoft, you can't not be invested. tons and tons of money in GPUs. If you're Amazon, you can't not be investing tons of money in GPUs. And if you're a startup in the space, of course you're going to spend all the money

your investors will give you on GPUs, right? Like the core weave or sort of whatever it might be, particularly when you have a, you know, you're sort of Nvidia as a partner and they're feeding you more GPUs than anyone. And so, yeah, the theory is that at some point, we realize that there's $75 billion or $75 trillion or whatever the number is billion, I guess, being spent on GPUs and $3 billion in revenue or whatever. It's some weird crazy.

I don't have if I had the units right. But and then suddenly you realize, crap, we're not making any money. And CorWeave goes bankrupt or all these companies have huge writeoffs and the stock plunges and the CEOs get fired and all the things that happened because you overbuilt. But the GPUs don't disappear. They're still there. And now suddenly the cost of inference is the same as the cost of compute. it was a critical once we the cost of compute which is basically treated as zero now it's not zero anyone who builds a cloud service spends a ton of time thinking about cost and optimization and

cloud spend and things on those lines but the product developers don't really think about that that's more like an ops problem to solve as opposed to a defining sort of product engineer design around these limitations, yeah. And so if we, the idea is we might get to that state with AI where what would it mean if, okay, we're just going to assume you can do this generation for free, then how would you design your products? What would you do differently? And, and that's, it's a meaningful difference.

And so, yeah, that is a potential we get there. Now, it's a little complicated because GPUs keep getting faster, keep getting better. is there a point where old GPUs, it doesn't matter because no one wants to use them. But you can also make the point that, look, broadband got faster, networking got faster. Like a T2 line was amazing in the late 90s and now it'd be like unusable sort of today. Well, and Vintyna is still selling its older model of GPUs, right? At least so far, the older models continue to be useful. Yeah.

So right now, I think they've mostly sold all their A series chips. Their H-1s are still selling great. there was a question, are they going to sort of have a hole because they announced the new chip, right? Is there going to be, you know, there's a, there's a very famous story in tech about the, um, uh, Osborne. The Osborne was this amazing computer that was actually really ahead of its time. It was you, I think you carried it like a briefcase almost. It wasn't quite a laptop. You had to like set it up, but it was somewhat portable.

And it was selling well. And then they told them, oh, yeah, the Osborne two is coming. And all the sales of the Osborne one dried up. And they went out of business. Yeah. And so it's called the Osborne effect, which is when you announce the next product, you kill your current product. And that did not happen in video. Like they announced Blackwell a few months ago.

It's coming in the fall. And it has not hurt their sales at all. It does appear to have hurt AMD sales, perhaps, or to, like, or I don't know whose sales is necessarily hurting. But their old chips are selling just fine. And that speaks to the fact that, like, the. It speaks to this demand environment. Like there is no, there's no waiting for it. And Jetson Woggs and the news call, I was like, look, if the goal is to make world changing

AI and if the difference is being first on the scene with a new model that gets all the attention or being second. And I was like, oh, yeah, like, like Anthropic versus Open AI, right? Anthropic comes out first before chat GPT, the world might look totally different. Like, it's a valid point. And so you have everyone with these incentives. have to be first. We have to be first. We have to be first. So no one can afford, we can't wait for Blackwell. We'll take H100s now. Oh, we'll take all the H200s that you have, the sort of new version with

more memory. And so it's working out great for Nvidia. It also is speaking to all the dynamics that did drive sort of that telecom bubble for sure. Well, everyone who's spending recklessly now should feel good about it. It's ultimately an investment in the future. Maybe someone else's future. Right. And so, no, it's actually, you know what? It is a, if you want to zoom out and get philosophical about it, it is a real sort of, you know, pat on the back, gold checkmark, whatever you want to say, for capitalism. Like people will get upset about companies.

Oh, companies don't invest. Don't do X, Y, Z. They're focused on quarterly earnings. Honestly, most companies should be focused on quarterly earnings. They're like they're they're filled with delusions of grandeur where management thinks they can invest the money better than their investors. They should just be returning money to shareholders. Like that is and let shareholders go and find new companies and new opportunities to invest. I've always been pretty skeptical of the like the the the critique that companies don't invest properly.

Companies don't deserve to live forever. Most companies that are old and established aren't innovative anymore. That money is there's a massive amount of money obliterated by management. pursuing something that is never going to materialize and give the money back to shareholders. In this case, though, it's a, that's not happening. No one is concerned about like these big companies. Oh, like, where's, where's all the worry warts here? Oh, I wish they would focus more on quarterly earnings.

They're spending too much on the future, right? Well, but shareholders are so fickle that that's, that's where it's like, all right, so how long does this last? Because right now they're rewarding spending, but eventually and maybe. within the next year or so, they'll start penalizing the same spending on AI. And so that's where putting the future in the hands of shareholders starts to feel kind of ridiculous just as a general proposition because they're so fickle as a rule. Last time I checked, Microsoft's stock is sky high, Google stocks is sky high, Nvidia stock is sky high. From what I can see, shareholders are investing in the future quite enthusiastically.

That's true for now. We'll see how long. Yeah, people are, people insufficiently trust the market. They're like, oh, I can sit here in D.C. And better decide how capital should be allocated. No, you can't. Like the reality is, is from where I'm sitting and even to your point, even if this goes wrong, the net effect is all these shareholders that got exuberant and all these executives that felt these incentives so they could not invest. The end result is a good thing for society, is a good thing for the future, is a good thing for innovation.

These GPUs are going to be built. They're going to get deployed. And yeah, someone like lose their shirt on it, but they're still going to be there. And someone's going to come along and scoop them up or the companies are going to figure out are going to price them super cheap at the marginal cost of electricity. And then we can have a lot of innovation happen. Guess what doesn't happen in a command and control economy? You don't get exuberance spending that gets you, like, the problem is you're on a local

maxima. And if you say, look, like, yeah, I can see this future of broadband everywhere. How do you actually ever get to the point of making the investments in broadband so that the applications come online that leverage that broadband? You need a way to cross the gap. And exuberism, yeah, it's a tough thing for shareholders in the meantime. And a lot of people, if this happens, will lose money on it. But why are we worried about that, broadly speaking? Why am I concerned about, you know, like the exuberism is part of the model.

It's something to be to applaud. Like so I'm serious. No, is exuberism is not a word. I'm struggling. Exuberance. Exuberance. Early in the morning where you are. Yes.

No, I hear you. And I actually have been excited about this theory since you introduced it 12 months ago. As long as I'm on the point, by the way, again, most shareholders critiquing management and wanting them to not invest are right. There's a lot of managers that should not be investing in the future because their companies stink. And they have some old model that worked for a long time and they should just be returning money to shareholders who can go put it in new companies.

So yeah, I generally, I don't have much time. I feel there's this is sort of like the, I want to be sophisticated and pro-capitalistic lack of trusting markets and command of control sense that this is how it manifests is like, oh, like shareholders are dumb, blah, blah, blah. Look at a company like Amazon. Amazon has for decades, you know, famously not made money. People were upset that shareholders kept giving them money and believing in their vision. And guess what?

They deserved it. And the shareholders got their return, right? Like, I think we, we, we, it's such a human drive just to not trust the uncertainty of the market, of the idea that it, like, it's going, like stuff will work itself out. and oh, I can figure it out. Like obviously this should be, should be better. No, you can't.

It's so much more complicated and difficult. Capital allocation is such an impossible problem. And failure is part of the model. There is this critique of someone I once saw who, he was like, oh, yeah, I worked in Silicon Valley. And, man, the company I worked for was so poorly run and they were wasting capital. I thought, man, like this whole right-wing argument that company, companies can spend money better than the government is really stupid.

What was dumb about that argument and what this person didn't understand is the company, I looked it up, the company this person worked for went bankrupt. That's the feature. The market at work. That's the point, right? Is if you're a private company and you're stupid, you go out of business. If you're the government and you're stupid, people get their taxes raised. You're not going out of business.

The whole point is the ease of failure. If that's a feature, not a bug. The fact this can all go wrong is a feature, not a bug. And people want this world where every investment decision is smart. Every sort of juicerow that comes along is like, you know, what a dumb startup. Like, oh, VC is stupid without appreciating that all advancements come with failures. If you like, if you want to increase the upside, you have to increase the chance of the downside. you don't get to look backwards and make perfect investment decisions for free.

And anyhow, this is a long- Not a deterministic podcast here. Did not know we are going to be getting into a spirited defense of the free market. But I do generally agree. I'm also reluctant to speak in absolutes because the two counter examples that come to mind are Apple basically being handcuffed by its quarterly earnings and the addiction to services revenue and some of the manufacturing and the PRC, which has been a huge part of their growth story, but also is a long-term risk. And then also the Boeing story and the conversations we had about the financialization of Boeing and prioritizing sort of quarterly earnings at the expense of long-term engineering. Well, sure, let's talk about it. Let's talk about it.

Yeah. Sorry for the mailbag. Apple, number one, what exactly should Apple be inventing that they're not? because of their supposed services addiction. Right? Like the phone is the perfect device. Like there's like what is not being done that sort of that sort of should be done. And even with services again, like there is a compelling argument for the way Apple does things. Now, what I prefer?

It just puts a giant target on their back. But the reason they've gone this far down the road is because they can't afford to take the hit unwinding all that revenue now. The reasons they've gone this for down the road is because they don't have obvious new revenue streams. Yeah. Like that's the bigger problem. And that's not a function of Apple being dumb. It's a function of the iPhone being the ultimate manifestation of personal computing. Apple is a personal computing devices.

They've produced the ultimate personal computer. And now what? Right now, no. Well, and the market is saying now what, though, aren't they? Because, I mean, they got this massive business, but they need to maintain the growth multiple. They do. and we'll see how well they do it. And that's why, you know,

and let's go to the Boeing. We'll come back to Apple. Okay. What's the issue with Boeing? The issue with Boeing is, number one, the massive consolidation in the U.S. airline industry, which was driven by the government, like particularly for defense needs,

first and foremost, where that famous last dinner, the Clinton administration was like, look, you guys are all going to consolidate or you're going to be going out of business. Not so smart in the long run. Also sort of driven by them. It's incredibly hard to enter this market.

because of all the regulation and difficulties around it. Now, again, I'm not going to get the debates about airline safety regulations, just saying there's lots of reasons why it's way harder to start an airplane company than it was 70 years ago. I would argue in that case, and this happens again and again, where people point to breakdowns in competition and you follow it down. It's like, you know, the meme with the goose quacking after the guy. It's like, what's the limiter? What's the limiter? It's often government regulation. It's the fact that you can't actually break into some of these areas. Now, again, there are also sort of quote unquote natural monopolies, things like the cable thing and the telephones were a good example. And you can make the case, that's the case for some of these platforms like the iPhone, right? The fact that because it is API locked in. And that's why, by the way, I have consistently said our antitrust law does apply and ought to be applied better.

to platforms where there is API level lock, and there isn't sort of an obvious sort of new entry. So yes, there can be sort of market failures in this way, but I think that those examples are overwrought and overstated to the extent they ignore where are the actual drivers of a lack of competition. But, yeah, but I mean, I guess if you want to pick one, the Boeing one is a good example, but no rule is absolute. Right.

There are exceptions that prove the rule. And I think that is that I would definitely grant that being one of them. Okay. And so as far as Michael's long-term dream is concerned, is there anything that would argue for one of the big companies not just buying up all the excess GPUs or overcapacity that exists a couple years down the line in that hypothetical? Yeah. Well, there is sort of a reality like as stuff moves on, like just it naturally closes new entrance. This sort of happened with cars for a long time. I remember, that was the analogy I used at the end of the beginning, was I talked about the car industry, had a ton of startups, a ton of innovation, all this stuff. Like there was hundreds of car companies that were started, maybe thousands. I can't remember.

I will link the article in the show notes. I have the number in there. Bill Bishop and I were talking about this. Like, if you go back to the dawn of the 20th century, there were so many different car companies. And that's now happening in China in the EV industry. And it'll be winnow down. That's one of the things that people complain about the Chinese auto thing. well, there's been a ton of Chinese automakers that have gone bankrupt over the last, like,

it's part of the process. It's actually been a pretty free market. Yeah, exactly. And so what happens is once it's all the innovation is sort of figured out, then you do get consolidation and market shift from differentiation to sort of superior production and cost structure. Yeah. Yeah, which is basically about scale. And I think there's an argument that to some extent, we're there as far as cloud infrastructure goes.

Right. To be a competitive cloud player, you need regions all over the world. You need all this sort of buildup. You need the sort of customer base. Can anyone really challenge the big three? Now, again, Nvidia is propping up to a certain extent core weave by virtue of giving them priority access to GPUs. And so you have a situation where Microsoft rents capacity from CoreWeave because Microsoft can't get enough GPUs on their own. So they need to get GPUs via CoreWeave. So CoreWee is basically taking a skim, which, which goes back to Nvidia, who is a big shareholder of Corleave. Yeah. So he who controls compute controls the world, man. I mean, Nvidia is pulling a lot of interesting stuff. Like they're doing a very effective job of capitalizing on their position in the market currently. Yeah.

There was a, there was actually a bit on the Dell earnings call last week where Dell like doesn't make it, like they have this huge revenue increase from selling AI server. but they have like no profit increase, which basically says the price of Dell getting AI chips to put in the servers is Nvidia takes all the revenue. And like Dell can't price them higher because someone else will undercut them because Nvidia has the choke point and they're sort of perfect competition around them.

And Nvidia is playing this to their own benefit to the maximum effect. My suspicion continues to be that the reason why DGX Cloud got off the ground, Nvidia is trying to build their own cloud without building their own cloud. DGX is actually in Amazon. It's in Google. It's in Microsoft. And Nvidia will say that's because, oh, we want to be close to the data. What I think is unspoken is the price of these clouds getting our GPUs when they desperately need them is they have to basically host our cloud for us.

We're going to be a cloud on top of a cloud. And yeah, I mean, like, I don't have a big objection to that. It's, it's, you know, I think that business is brass knuckled and that's fine. But, but yeah, there's a lot of that going on. And so maybe we'll get core weave out of this. Core weave is probably the most fragile entity if this whole thing sort of blows up. But we'll see. I will see what happens.

Right. And the cost for startups will come down regardless in that scenario, correct? Exactly. The whole point is there is cloud competition. All the cloud entities are trying to lock you in. Nvidia is trying to lock you in. Like they have, you know, these, these, these pre-built models and all this sort of infertural inference something or others, nims, I think they're called. That's basically only work on Nvidia GPUs.

The idea is, oh, we'll help you get started up as a company. Super easy. Get the model working. Oh, sorry, you're stuck on Nvidia forever. You know, the, and so, yeah, like, there's all a scrambling for place. But, yes, the idea is in the long run, if this happens, if there's a GPU crash, GPUs will be cheap everywhere. It's going to like, it'll be a commodity.

Okay. Well, in keeping with the theme, the unexpected theme of today's podcast, I want to jump to this question we got from Christoph, from Christoph, he says, a few weeks back, you highlighted Andy Jassy's comments comparing Amazon supply chain and AWS. One thing that I have consistently worried and wondered about is what the true return of these two CapEx hungry. Beasts actually is, and if they generated economic value added that is acceptable for the company's valuations. We constantly only fixate on the growth numbers or income statement margins, but don't really consider what data centers and supply chain business should return if the company should stop capital expenditure slash growth tomorrow. Single digit or low double digit

internal rate of return is simply not good enough considering the valuations. So Ben, What do you think? Well, one of the huge earthquakes when AWS numbers were finally revealed, I call it the AWS IPO. It was like the Google IPO. It was a massive deal. But it was a private just company showing off like, oh, breaking out of the other income statement. But it shook the whole industry. And part of what shook the whole industry is up to that point, everyone knew AWS is a big deal.

It was completely transforming the startup ecosystem. Now you could actually, it used to be you had to go raise money on an idea so you could buy servers to get started. Now you could actually build a company and you would raise money based on an initial prototype that was already running and could potentially already have customers. And so you had a drastic explosion, massive explosion in startups. And then all the crunch came at the Series 8. That's what you got the Series 8 crunch, which is, okay, that's everyone who actually has a product out there now trying to actually raise real money. And venture capital shifted from paying for equipment to paying for sales in many respects, like a Salesforce or customer acquisition on Facebook or whatever it might be.

So it was completely changed the startup ecosystem, AWS to generally. But everyone assumed that they were making no money on this because how could they be? Right. Like it's like, oh, well, that's Amazon. Your profit is my opportunity. Da, da, da, da, da. And it was like, well, that's why Amazon is in good shape. and why people will not challenge them

because these big companies like Microsoft and Google, they can't accept the lowered margin, right? Because they're software companies. Part of the earthquake of AWS is not just number one, how big it was, but how much money they were making. It was actually a profitable business. Now, back then, I want to say their margins were in the teens. Over time, that blew up into, I think, the 30s.

I think it's gone down a little bit then, but it turned out that cloud computing was way more profitable than anyone realized. And I think for Redmond, in particular, for Microsoft, it was like, okay, we have, we bid dilly-dally with Azure, but this is actually, maybe we should get serious. You make some business here, yeah.

And even the same thing with Google. I mean, like, where Google, you know, it took them years and years for GCP to be profitable. It's barely profitable now. But the reason it's tolerable is the sense that, look, okay, no, you're not going to get 90% margins. But given the revenue and volume and the natural sort of a, with your business where you already have, you know, data centers over the world.

Yeah, if we get 30% margins, that's pretty good. And like absolute dollars still matter more than sort of your, your, your, your, your, your, your, your, investors aren't dumb. If you're getting, you know, by and large as a rule, if you're a Shopify and you're going from 90% margins to weigh less because you want to build out in a WS or Amazon. Yeah. Yeah.

There's a lot of skepticism you could pull it off. But I think there's more grace given, you know, if it's like, you're not hurting your core business. This is additive. The revenue increase is so large. We will allow you to have sort of a lower margin profile over time. And that's been proven out, by the way. Microsoft, you know, this build out of like this huge run up in their stock price has come with this increase in sort of cloud computing that, you know, at least in theory would be sort of lower margin. I have to give the exact numbers.

I'm sort of talking off the cuff here. But so, number one, evidence suggests these cloud buildouts in the long run can be solid. And you think about it because there is a nice orthogonality, which is you have companies that find value in because it's operationalizing the costs instead of like, you know, doing fixed costs, they can flexible, they can scale up and down. And all these cloud companies work very hard to lock you in, right? You know, it turns out you can get way more efficient if you use redshift instead of like you have your own. Oracle installation, wherever it is. Well, now you're walked on Amazon, right? And they're all trying in every enterprise, like, oh, we need to be multi-cloud.

We need to be able to not get locked into one or the other. And then they get locked into a middleware provider that is like, oh, I'm now locked into data bricks, which is multi-cloud, but, you know, whatever. It's the eternal dance of like lock-in versus not, it's going on in tech for years and years and years. What of Oracle's selling points was Larry Ellison was he was pushing the fact that, oh, Actually, you don't want to be locked into IBM. You want to come with Oracle and you can easily get your data out.

It's MySQL or whatever white people. Of course, Oracle is seeking to lock you in. There's massive sorts of lock it. This dance has been going on forever. But the evidence is that there is money to be made in the cloud. Will that extend to GPUs? Right now, Nvidia is taking it all. Right.

succeed. They want there to be multi-suppliers. And I think it's all a reasonable bet. And there is a moat. Like, who's actually going to come along to the previous question and build sort of competitors there? I think it's a similar bet with Amazon.com sort of logistics arm. Yeah. Which is if we're so big. Explain Amazon supply chain. Explain what that is in case anyone's not familiar. Well, I mean, Amazon has built out basically most of the stuff that you order on Amazon is handled by Amazon from all the way to your doorstep. And they're actually shifting where they're sort of making money in a lot of suppliers are mad at this. They're making more and more money on their third party merchants. So this is an aggregate of play where consumers go to Amazon.com to buy stuff.

Now, you could be a third party supplier. you could set up a website and no one comes to my website. I have to be on Amazon.com. It's like, well, I mean, that's, you sound like a newspaper publisher right now. Like people can still go to your website. It's your job to acquire them. Oh, then I have to pay Facebook to acquire. Well, welcome to the real world.

And so there's a, and so you're on Amazon. You buy, you're not necessarily buying from Amazon. It says, you know, the little fine print next to the buy button. You're buying from XYZ sort of third party merchant. That third party merchant has to give Amazon a share of that person. They have to give Amazon an inventory fee or storage fee for storing stuff in their warehouses. And they have to like pay part of the delivery. So Amazon has built up all this infrastructure that you need to use as a third party merchant if you want to be on Amazon to get your stuff delivered to the customer's house in a day or two days.

And it's so consumer friendly. And it's consumers love it so much that you have no choice but to pay it. And meanwhile, everyone can be on Amazon. There's there's kind of like perfect competition, particularly if you're an undifferentated product. that Amazon can keep taking more and more of your margin, and that's where they sort of make the money. So Amazon, their differentiation is not just that they have the big supply chain and they're building it out and they're getting efficiencies from scale and all these sorts of things that make sense as far as the whole scale sort of argument, but they're combining that with owning customer acquisition. Customers go there naturally and so suppliers have to pay to get in front of customers. And oh, by the way, you probably have to buy ads.

I was going to say an extra 10 or 15% of your profit is going to ads, which is basically pure profit for Amazon. Yeah. So I think it's pretty compelling. I'd be like the fact of the matter is that Amazon. So you're not sweating the single digit or low double digit IRA if you're an investor? Well, I think I mean, I think Amazon.com has always been barely profitable. I don't think they've made any grand promises about being anything. But it's a function of how large the market can be, right?

Like the return, you know, I'm not here doing valuations, but there is a function of what's the growth over time, what's the margin, what's sort of the defensibility, and what's the absolute value, right? Amazon is an absolute value play. It's like, look, we're going to push so much through this system in the granting of time. I mean, that was my article about Amazon. It's called the Amazon tax. Like, buying Amazon is buying an option. on attacks of all economic activity. That's their goal.

Like, it's very large. It's very upsetting to me. It reminds me of my experience rooting against Luca Donchich against the Minnesota Timberwolves in the last series. I have never been a Luca fan. I was a diehard wolf fan going into the Western Conference finals. And I had, you know, I was toying with the idea that maybe Anthony Edwards is going to show up Luca Donchich on this massive stage.

Didn't happen. similarly, I think that Amazon is bad for American businesses because they use their scale to extract these massive fees that then limit the upward mobility of small and medium-sized businesses is very frustrating. But when I look at what they're doing with the logistics business and extending it out, similar to how by the end of that Wolf series, I had to just be like, well, holy crap, Luca's amazing. of this plan from Amazon, I mean, I don't love what they're doing.

I think there are bigger questions as to the net impact and whether this is positive, whether it will be positive. But purely from a strategic standpoint, I can't help but admire the elegance of what's happening here, where Amazon is doubling down on all of the logistics advantages that differentiate Amazon and building toward this future where they can just insert themselves into practically every e-commerce transaction in America because they're integrating all of the logistics prowess and just offering it to people for a price

and whether the sale happens on Amazon or not, long term, you can see that making sense for like massive swaths of the e-commerce economy. And I mean, we'll see. But I would guess that's going to be profitable for Amazon. I'm not worried about it. With all due respect to Christoph. I have two questions.

number one how many amazon dot com boxes are in your house right now probably like eight i don't we've become an omni channel family though i make a lot of trips to target we order online and then i go pick up from target so i'm trying to wean us off of amazon but amazon still does deliver convenience like no other number two do any of those boxes have a lucidasha's jersey in them absolutely not although i i don't know i may be rooting for lucca as the finals approach here But yeah, we'll see. You can't give up a Boston now. It's been a 10 year a bit.

I haven't made up my mind yet. But yes, any final thoughts on Amazon? We also got a funny Elon cyber truck email that I want to read at the end. No, I mean, I think all these questions are like the point Christoph raised about the long-term sort of margin profile of all these efforts. Is it like that was part of I was right about last week with the integration and, and, um, modularization sort of bit. You want integration. Like that's where you have defensible sort of margins.

You, it is different in the cloud versus Amazon.com and AWS are similar in their structural auto ways, but one's an aggregator. Like Amazon.com, because consumers go there to shop, that is the source of sort of all their leverage. AWS,

it's different. Consumers don't know they're on a SaaS website that is powered by AWS. right but there is also a different kind of lock-in right like if you're built on a w s it's really really hard to leave yeah um whereas you know theoretically i can just go to walmart dot com and shop or i can go to target dot com and shop and sort of whatever it might be okay all right well we'll see how that evolves um final note here from sam he says i enjoyed your comments on elon the other day i can't stand the stuff he says i think his talking points on politics and social issues are beyond dumb, and the dude is wildly unfunny and weird.

I have also loved hating on cyber truck for the last five years with its delays, design issues, and wacky claims by Elon. Remember when he said it would be able to be a boat? So the funny is I don't remember that, but I also completely believe it. Yeah. So it's plausible. There's no question. Anyway, he says, I have a cousin who's an Elon lover, and he was one of the first buyers

and has been patiently waiting for a long, long time. So when he finally got his cyber truck, he wanted to show it off to everybody. I reluctantly agreed to check it out. Holy crap. The thing is incredible. It is a car that gave me the same marvel that I feel with new Apple devices. The little things in it amazed me.

Its design is simple and sleek. Small things like air conditioning can be adjusted with a touch to put it at the proper angle. The trunk is satisfying to roll. up and has a camping attachment. And there are so many little things that I never could have imagined myself enjoying. But most impressive of all was the sound system. I've never heard such great speakers besides trying on my friend's high-end headphones.

I used AirPods pros and somehow I think cybertruck sounded better. And then we rode in it. Just to step in here. Compare the size of the speaker in your AirPods pro to the size of literally other speaker. Yeah. Size is a major driver of sound quality. But not to diminish the cyber truck, but I'm not surprised. Okay. Then we rode in it, he writes.

He floored it on an open road. And I hadn't felt G-Force like that since I went on mission space at Disney World as a kid. It was incredible. I do not think anybody needs a car that accelerates like that on a public road. But the sheer power required to move a three-ton obvious. is unimaginable and it was accomplished with a battery. I was blown away. I still think the cyber truck shouldn't exist. It seems perfectly designed to destroy any car or pedestrian it comes in contact with. But man, what a cool piece of hardware. It's a shame Elon is such a dork.

I include this Ben mainly because it confirmed my suspicions that the cyber truck, while looking absolutely ridiculous is probably pretty awesome to experience in person. I said this on Sharp China a couple weeks ago. I would never buy a cyber truck, but I would love to try a cyber truck for a day.

Did you have any thoughts on the note from Sam? I think the looks are underrated. I think it's going to, I think the absurdity is part of the appeal, to be sure. No, I mean, I think it's just a, it's a good email. I mean, I think what I would encourage, I mean, I don't know, you'd do whatever you want. Like, why it's probably not particularly useful or fruitful in the fullness of time to get wound up on the tweets of a particular individual as far as like your consumer hardware purchasing decisions. Yeah.

I don't know. I don't know. I don't know anything. I need to say that out loud. Yeah. I mean, it's just like, I don't know. I mean, who knows? The cyber truck has some really cool technology.

There's something about how it's like the way it's electrified is like completely different compared to cars previously or something on those lines. I don't know. I'm not a electric car sort of expert as I think we've established. But yeah, I think there is that is sort of the Elon conundrum. And there's probably a bit where the advice I would have to both sides is the same, which is in your whole life, if you're looking for like a person. perfect alignment of beliefs and products and outcomes with your own personal political preferences. You're going to live eternally disappointed. And it's like that's the whole Elon thing.

Like stuff he does, I think is really dumb, stuff that's objectionable. And also you can't like people will bend themselves over backwards to pretend like SpaceX is not a thing. Right. Or like Tesla is not an accomplishment. The great irony. The answer is both. Yeah. And the great irony of Elon buying Twitter is that Twitter more than any other.

tech platform, any media platform in history is responsible for this awful tendency where we just flatten people into good or bad and then sort of reverse engineer opinions based on that binary. And Elon himself is like exhibit A of what that phenomenon looks like in practice. But I applaud Sam. This is the change we're looking for here. Well, the one thing about Twitter, I will note X or whatever it is. I've been very surprised to see number one It does feel like a lot of people are trickling back to Twitter that left for a while

And number two One of the common responses I've heard is people are like the 4U tab is really good Like that's one of the reasons they are back And it's interesting the 4U tab is very funny because it's pretty basic Like I don't think there's anything like mad happening on. What it is is a testament to how incapable the previous Twitter management was. Like having a compelling algorithmic timeline should not be that difficult.

There's a bunch of on Twitter now that's really annoying. Like everything, you have to have an image. Anything with Winks is suppressed. It's all becoming more self-contained in a way that I find personally pretty annoying and harder to find interesting stuff that I've used it for. I will also note the 4U tab is way better. I do discover new people and new things in a way that I didn't always previously. And again, it speaks to the overall conundrum.

There's a lot to critique. It also, it's like what on earth was like all our critiques of Twitter management in the past. It wasn't just they couldn't make money. It's like they couldn't make basic additions to the product like a functional algorithmic timeline, which Facebook figured out like 15 years ago. And oh, by the way, they did it with a fifth of the people or whatever it might be. So, yeah, and there's like, oh, well, we're not making any money, blah, all which is true. Like, no, but like, no, we, we, we, we're, we've already established our, I guess these emails are funny just because everyone ultimately ends up in the same place, which is, it's complicated and don't get too worked up about in either direction because it's out of your control.

Yeah. But that is, you know, that remains the case with, with all his entities. Look, Elon, wildly unfunny and weird, but the cyber truck, pretty cool. that's a perfectly acceptable take. So I appreciate you, Sam, and helping move past the phase of culture where everyone is good or bad. And if they're bad, then everything they do is bad. I think it's healthy that we're all moving past that. But while we're talking about social media that annoys us, one thing that's very annoying to me is that I am a Twitter bull and I enjoy Twitter.

I'll freely admit that I enjoy Twitter. But a lot of the most thoughtful tech commentary happens on. on threads. And so I have to go to threads now every day to get good tech takes. Because on Twitter, it's all like influencer, hustler, like 12-part AI threads. And I'm not interested in that particular genre of tweet. But there's some good stuff on threads. And I don't know. There's also a lot of really annoying stuff on threads. And so I'm struggling with this new reality. Can I make an observation, Andrew. What?

Just to establish your point of view right now, you're longing for, wishing for a past in which all compelling sort of conversation was on one place. And now that there's competition, you don't like it very much. You find it annoying and frustrating. Well, I don't know how to reconcile that with my worldview and my view and my view on what the American economy should look like. I don't think Amazon should be taking 60% of every sale on the internet. You can ship it from Amazon. Oh, God. All right.

Well, on that note, Ben, I look forward to coming back later in the week. I hope one of us gets to try a cyber truck this summer. That'd be a fun episode. But for now, thank you for the email. Sam, thank you for all the other emails we got. We'll keep it rolling. I'm going to ask you for a finals pick on the next episode. So come prepared.

But until then, I hope you have a good couple of days. talk to you soon. Bye-bye.