The Six Five Pod | EP 295: Vibe Coding, AI Infrastructure, and the Future of the App Economy

The handpicked topics for this week are:

1. MWC Recap and the Rise of “Vibe Coding”: Experiments at Mobile World Congress highlighted how AI interfaces are lowering the barrier to building applications. Tools like Perplexity Computer enabled rapid prototyping of workforce tools, content systems, and market-modeling apps. While experimentation is easier than ever, production-grade systems still require security, accuracy, and operational discipline.
   
   
2. The Collapse of Traditional Development Gatekeeping: AI-driven interfaces are reshaping who can build software. Users without deep engineering backgrounds can now quickly generate functional applications. Pat & Dan explore how this shift could dramatically increase the volume of software development while changing the role of traditional developers.
   
   
3. AI Infrastructure and the Silicon Arms Race: AI infrastructure continues to evolve as companies compete to deliver efficient compute and networking at scale. Qualcomm is entering rack-scale inference with LPDDR-based architectures designed for efficiency, while Nvidia is investing heavily in optical networking through companies like Lumentum and Coherent to address power and scaling constraints.
   
   
4. Intel’s Push to Stay Competitive in Enterprise AI: Intel continues advancing its enterprise and carrier roadmap with technologies like Xeon 6 and the 18A process. While the company faces pressure in hyperscaler markets, it remains focused on maintaining relevance in enterprise and telecom infrastructure deployments.
   
   
5. Apple’s AI Infrastructure Challenge: Reports suggest Apple is evaluating Google Cloud to support infrastructure for future Siri capabilities. The hosts highlight the company’s ongoing challenges with internal AI infrastructure development and the broader competition for AI talent.
   
   
6. The Flip: Is AI Ending the App Economy? The weekly Flip debate takes on vibe coding. Will AI-driven development lead to an explosion of applications that disrupts traditional SaaS monetization models? Or will this shift simply upgrade the app economy with simple tools, while durable SaaS businesses focus on unique data, strong governance, and trusted platforms.
   
   
7. AI Capex Continues to Drive Semiconductor Growth: Broadcom and Marvell continue benefiting from the rapid expansion of AI infrastructure. Demand for networking, connectivity, and high-performance silicon reflects the ongoing global buildout of AI compute capacity.
   
   
8. Cybersecurity and AI Disruption Questions: CrowdStrike delivered strong financial results but faced investor questions about how AI could reshape the cybersecurity landscape. The discussion highlights how AI will both disrupt and reinforce security platforms.
   
   
9. Capex Pressure and AI Investment Cycles: Amazon faced stock pressure related to the scale of its infrastructure investment, yet its Trainium chip strategy and expanding partnerships, including work with OpenAI, reinforce its long-term AI ambitions.
   
   

For a deeper dive into each topic, please click on the links below. Subscribe to our channel so you never miss an episode.

Listen to the audio:

IMPORTANT: Note to Uploader: Libsyn AUDIO EMBED - added by uploader.

Disclaimer: The Six Five Pod is for information and entertainment purposes only. Over the course of this webcast, we may talk about companies that are publicly traded and reference share prices, but nothing discussed should be taken as investment advice. We are not investment advisors.

‍

Daniel Newman:
Yo, yo, yo, yo, yo, yo, yo. We're back. We're in the chair. We are back from overseas.

Long trip. Not sure it was worth it. What's your readout, man? We hit MWC, my 12th year, Pat's 75th year. Been there so many times.

Patrick Moorhead: 

Yeah, I was there when we did Dial Tone.

Daniel Newman: 

Do you know how many chicken burgers we ate this week? Do you have any comprehension of how many we ate in the same hotel, sat at the same table?

Patrick Moorhead: It might have been eight or nine.

Daniel Newman: I ate like nine chicken burgers. And it's where they put the chicken patty with the egg on it and a little bacon. We did get to eat like 10 meals together though. That was fun.

Patrick Moorhead: I know it was crazy. Yeah, I'm remarkably recovered for an international trips my third and two months to you know, did did Davos did MWC did the UAE? Maybe I'm just getting used to this.

Daniel Newman: 

Well, you know, now that we've got bots doing all of our real work. Now, the theme of MWC for everybody out there that was Pat and Dan discovered Perplexity Computer. Pat comes into town and he's just showing off all these apps. He's basically like, I've built everything you've ever built on the trip, in the car. I did it on the plane. You know, I've spent years with like a dev team building like a platform. And, you know, Pat just mogged me on the way in and started showing me all the stuff he's built. And I'm like, you know, fuck this. I'm going to download it. So we're sitting at the table eating a chicken burger. And I download perplexity computer and I start hacking up, like I start building out like bottoms up semiconductor models using open source. And by the way, perplexity computer, because I use cloud a lot and code and AI, but like the graphical interfaces that computer create are just next level. They're so good, like right out the gate. I look at these things and I'm like, there is no differentiation anymore in interface. Like that has been broken by what perplexity can do in creating like user experiences. But what did we build? Like I built eight or nine apps. I built like workforce augmentation apps. I built content augmentation apps. I built model and evaluation apps. Like literally, and we were doing this while still running our daily schedule. We would like go to dinner. We'd sit down at our computers and we would just sit down. We're like giggling like school girls. No, we're like building stuff and like showing each other our computer.

Patrick Moorhead: 

No, no, exactly. And then I think one of us would go to the bathroom and then, you know, we'd kick off a new app that we would create. Completely silly. I mean, I built market models, analyst productivity tools, brand new website. You know, the team told me it's going to take like a year to do a new website. I vibe coded a really good one in about 30 minutes. Three new AI tools for external consumption, social media tool. I mean, it's just, it's just ridiculous. It's fun. I like, I feel like I'm creating again. It's kind of like, when LLMs first started or even when the internet first started, it truly is incredible.

Daniel Newman: 

We're so here or back or whatever we'd say, but yeah, we've seen the disintermediation of technical know-how where if you can really communicate what you want, you can build it, which is great. Let's be clear, we'll talk about this later in the show, but There is still some line in the sand between building something for your own use and like truly, fully opening an app to the world, okay? but the line's getting thinner. It's getting thinner. And like, and by the way, there's apps you can build for yourself, which you can pretty much stand up and use immediately. There's apps you can build for your team. Those can be pretty efficiently deployed because there's not a lot of like security and the concerns. Then there's apps where you're building like content or external facing content. And those need a little more proofing. Cause obviously, you know, we gotta make sure the AI is being, honest and not lying and giving in. I mean, because you and I got to admit, we saw some real nonsense data, you know, and some of the things like it looked good, but it wasn't accurate.

Patrick Moorhead: 

Well, the meme is so funny. The meme is just like, it's so true, like make no mistakes. I've added to all of my instructions, triple check for accuracy. And I even was playing with it and said, you know, if you don't get it right this time, I'm switching providers, like you're done. And it surprisingly did a much better job.

Daniel Newman:  

Yeah. Threat steward, Daniel. You're scolding your bots. These poor bots, man. I mean, you say they have no emotion, but I see them in the back room having a coffee and a smoke and crying. They're crying in the back room having a virtual coffee and a virtual smoke. All right, well, we could talk about this the whole show. In fact, both of us kind of were thinking about maybe not doing the show and just going back off and building. We will come back at some point and talk about this. But in all seriousness, a lot, a lot cooking, busy week about MWC. We're off to, you know, not next week, week after off to GTC, RSCA for you security nerds. That's coming up after that. I think I'm going to be there. Pat's going to be at HP up in New York. I'm going to be at Arm in California. Stuff's happening, Pat. But we've got a pretty thick docket to go through. We've got our decode for those of you that forgot how the show works. That's where we break down the kind of big news of the week. We do a flip where I mog Pat in an argument, in a fake argument. I was on the debate team. No, I wasn't. That's a lie.

Patrick Moorhead: 

I was on the chess team.

Daniel Newman: 

You think I was a nerd?

Daniel Newman: 

You think I have guns like I was on the debate team?

Patrick Moorhead: 

Come on.

Patrick Moorhead: 

You're like a freak of nature. Being a communicator is cool again. I don't know, man. I had a 4.0 in high school and I got these guns.

Daniel Newman: 

But listen, even if you didn't have a 4.0, you're at the point where it's so far in the past you could lie about it. There's no way anybody could check. You know, you're like Al, you ever watched Married with Children?

Patrick Moorhead: 

Yeah.

Daniel Newman: 

Remember Al Bundy and his football memories? That's you now. That's where you're at in your life. You know, when you talk about it, like I was a starting varsity, I benched 700 pounds. You know what? You might've, nobody's going to know. There was no picture. 

Patrick Moorhead: 

Hey Daniel, do we have a show? 

Daniel Newman: 

No, I don't want to do a show. I'm having fun. We should have fun. Let's, let's do a vibe coding in real time show. 

Patrick Moorhead: 

Okay. I'm game. 

Daniel Newman: 

See Pat, this by the way everybody is when Pat starts to get moody because I'm not going fast enough to the docket. Now he's thinking about the next app he wants to build. He's bored. I am pretty boring. 

Patrick Moorhead: 

Well, I'm bored, let's go. 

Daniel Newman: 

But then we're gonna talk about these topics and I'm gonna be bored. Okay, I'm just kidding. All right, let's go to the decode, Pat's being a little bitch. We can edit that if you want, I don't care, but I think it's funny. All right. So let's start with the decode, Pat. You know, I saw a great photo of CEO Cristiano Amon standing next to you. I wouldn't say like, they're not entering. It's kind of like a re-entry. They kind of made an announcement and MWC was a second chance to talk about Qualcomm entering rack scale AI inference race.

Patrick Moorhead: 

Yeah, so a lot of players here in AI infrastructure And the way that the market and I, you know, I don't know if you had called it too, but it was clear to me that we would see some heterogeneity kicking in, right? Which would mean you can't just have one piece of silicon that's good for everything. And one thing that's clear is that inference has certain characteristics that having a unique piece of silicon benefits from and training. And in fact, you know, the split between Even the inference workload is happening between pre-fill and decode. Qualcomm, best known for low power, high performance silicon in areas like automotive, IoT, and smartphones, and now PCs, threw its hat a while ago into the AI accelerator space. And what they did at Mobile World Congress for the first time is they showed the rack. Now, they were light on details as they're going to be waiting to give all of those during their financial analyst day. But they did talk about some interesting stuff. It has hexagon NPUs, LPDDR, and that's the key. It has a tremendous amount of memory that's accessible, right? And what that means is two things. First of all, LPDDR is less expensive than HBM. And you can pack a heck of a lot more directly on the card. And that's exactly what Qualcomm is doing. In fact, you can pack on here 768 gigs of LPDDR. They showed a 350 billion parameter model on single card. And by the way, you can get 43 terabytes of memory at the rack level. 140 kilowatt rack, so it's not absolutely insane. That's almost 10x lower than some of the new racks that we're hearing about. So yeah, I mean, forward progress. The company's got the deployments on the AI100. I've got the AI 200 which they showed at the show and the 250 is on deck for the future in the future that has a memory based computer architecture which we did most now is cool now that rock had an architecture that was memory based.

Daniel Newman: 

Yep, absolutely. I mean, look, inference, all the stuff we were nerding out about in the beginning, you know, the ability to do to work on, these are inference workloads. We're seeing it ushered in by the Blackwell era, but we're going to need more efficient and more of it. And so this is going to bring attention to these kinds of solutions. It's going to bring attention to whatever Jensen announces at GTC. It's going to bring attention to more CPU solutions. I mean, we are seeing And look, Qualcomm is going to have to fight for its right for market share. It's going to have the right to compete. And it's going to have to show, I think, you know, being low power, I think being I think, you know, the efficient memory that you mentioned and architecture. I think they'll win. I see them having some success in the Middle East. I see them having some success potentially with some NEO clouds and some tier two clouds. I think they're going to have a harder time breaking into the massive hyperscalers just because you're competing with multiples. But if they can get the numbers and economics right and provide the volume and capacity expansion. I mean, you did hear today that the CFO of Meta came out and said they are still capacity constrained, even with all that CapEx. What does that mean? It just means all the compute that can be built will be sold. So if Qualcomm can build something that's good quality, I think they can sell it. We'll actually have some stuff out. We're working on some market models to try to figure out what Qualcomm share will look like a few years out. So we may have more to share for everybody on that pretty soon.

Patrick Moorhead: 

I created a market model sitting on the toilet this morning. 

Daniel Newman: 

You're such a dick. 

Patrick Moorhead: 

It's a joke, Dan. 

Daniel Newman

No, no, it's not, this is my life Pat. It's not funny. It's not funny. Now I'm going to, what was that commercial, essentially the energy queue commercial? I'm just going to be riding cycles and powering data centers when I'm unemployed. So anyways. 

Patrick Moorhead: 

I don't know if that's funny, but it is. You're funny. It's just you kind of laugh. You're a funny guy.

Daniel Newman: 

Like a clown? Anyway.

Patrick Moorhead: 

I remember the first time I saw that movie, like, what a scary thing. Like, it was the tension in the movie theater, right? Back when we used to go to movie theaters was super thick, you know?

Daniel Newman:

 And for all of you, like, under, what, 35, 40 that have never seen Goodfellas, I think you should see that. It's actually still a good movie. Like, a lot of older movies feel slow. It's actually still good today. Good movie. Actually don't make them like that anymore. Now I sound like a boomer. 

All right. So let's talk about NVIDIA and it's $4 billion, $2 billion each goes to Lumentum and Coherent. You and I had the chance to actually talk to Jim Anderson, CEO of Coherent. I like a few things about this. One is that the signal it sends to the market and to investors as well is that optical is the next big thing. It's got a big constraint. It's coming. And, you know, I think it's, you know, whether it's with Feynman or something that comes after, I mean, NVIDIA is committed to kind of all optical. And it knows that it needs to get ahead of the supply chain. That's their biggest advantage, in my opinion, right now, beyond the technology itself, is that NVIDIA has more capacity. They bought ahead of every one of the trends. And as they accelerate the speed of production to market, they continue to put more pressure on their competition. And their competition can't keep up. They can't keep up on design. They can't keep up on build. They can't keep up on supply. packaging anything, because Jensen has the money. He's got the war chest to get ahead. And of course, he's doing all the right things going overseas. But the money spent with Coherent, by the way, everybody complains about NVIDIA's investments. NVIDIA's investing just so that people can buy their stuff back, which I've talked a lot about. But this is a case where they're buying, they're actually investing in a company to go build more so they can actually buy more. So I say kudos to Jim Anderson, Coherent, to Lumentum for locking these deals down, being the chosen companies to provide optical scale and basically getting a guarantee starting in what, 27, to see a pretty significant revenue ramp with NVIDIA. And it never hurts to have that company on your cap table.

Patrick Moorhead: 

Yeah, it was such a clean deal too. I mean, I think you already mentioned the non-circular way they did it, and it's essentially just buying something from somebody for a long term. It's also non-exclusive, right? A lot of the questions came up, for instance, when NVIDIA invested in OpenAI, was it exclusive type of arrangement? And I think that it's super clean. Yeah, hats off to Jim and also, you know, quite frankly, the folks over at Lamentum also who saw a deal. One thing I do want to bring up though is, and if you've paid attention to the Broadcom Call some of the takeaways could be that you have to do CPO because you have a power issue in some other part of the rack. Let's say that your switches consume an incredible amount of energy. And, you know, we've had this tension with cables like forever. and interfaces, and every time you think copper is dead for a certain implementation, somebody figures out a way to do that, a way to do PCIe, a way to power your cables, the way to change some sort of an encoding pattern on it. And copper is cheaper. So I think what we can say is that the world will continue to be a blend of advanced copper and optics. But the thing that's great is we're solving a lot of these technology problems, what seems like near, not near real time. I mean, my gosh, believe it or not, 15 years ago, you know, basically taking data over light, you know, that wasn't some sort of an internet, internet backbone was pretty much rocket scientists. And it was kind of like quantum, right? It's always five years out. But, you know, to the credit, the, the industry has gotten their act together on optical networking, made it more reliable, made it a lot less cost-reduced as well.

Daniel Newman: 

Absolutely. Let's talk, Pat, a little Clearwater Forest. You and I had some sit-downs with Intel. You took some cool pictures of yourself inside the chip. I couldn't figure out how to do it because I'm not very bright, but you did it. You looked great.

Patrick Moorhead: 

Yeah, what's up? Yeah, so we had a good chance to talk with Kevork Kechichian. Hopefully I said that right. It was the first time he had ever gotten on camera. And, you know, we didn't go in there and, you know, pin our ears back and ask, you know, gotcha questions. It was really a good conversation on what Intel is doing in the space. And, you know, we were at MWC, but we talked about that. But it was good to see, right, the first 18A server compute part, right? It has Clearwater Forest, Xeon 6+, 288 DarkMine E-Cores, 12 Intel 188 compute tiles stacked on three Intel 3 Active Base tiles. That's essentially for I.O. with the Fulvius Direct 3D and EMIB for Unpackaged Tile Interconnect. So I think it's good to see real technology going from a PowerPoint to something even more extraordinary like this. And I'm not going to mince words. I mean, Intel is very challenged in the server space, specifically in hyperscaler. AMD has like 75% market share there right now. Where Intel is still cleaning up is in the enterprise. And quite frankly, in the carrier space, they have a very high degree of market share as well.

Daniel Newman: 

Yeah, and you and I, you know, CPUs are cool. I'll keep saying it. It's a moment for CPUs that this agent flow is going to be big. You know, for Intel, the challenge is getting its capacity right, getting its supply chain right, getting, you know, obviously it's its own path. But I mean, I think this is an area of growth for them. It's funny, Pat, for a long period of time, the company kind of steered away from the AI CPU story because it was like they were kind of being picked on for it, like if that's your story. And now it's a chance to get loud. It's actually a chance. But they have to execute to be able to deliver, because I think part of why they're being quiet is they don't have enough supply. And so this ramp is going to be really important here. So keep your eye on that. So this story is creeping up again. 

All right, so this is an expanding story. Apple's getting some wins this week. It's got its new $599 MacBook. It's had a huge run on Mac minis for OpenClaw. Apparently their high powered notebooks are looking like AI war machines. And now that inference is all the rage, Apple's getting cool again. But what Apple still hasn't really figured out is it hasn't really stood up its own meaningful infrastructure yet. It looks like they're gonna be turning back to Google again to do that. So, you know, the deepening of partnership, the dependency on Google. I guess really the question here is, is this a moat for Apple or is this a risk for Apple? Meaning is the moat that Apple is not spending big on CapEx and they're actually gonna gain an advantage because they've got the user layer, the customer layer, or is the risk the fact that they basically are dependent and have to turn to other companies for everything they need right now? Because at this point, by the way, next gen Siri is still a pipe dream. Because of the last several attempts with Apple intelligence and everything with AI that they've done has been kind of. Underwhelming is that a fair term it's been you know. Um, and so I think it's interesting, you know, I, I keep talking, you know, my, my take is like Google's win streak is pretty impressive right now. They seem to be, you know, continuously being chosen as the, uh, AI cloud for the AI era. Um, and they're creating, uh, you know, a challenge for everybody to keep up. Apple utilizing Google, using them on search, using them potentially for more AI tools, but maybe now using their infrastructure for Apple's own AI tools. My read on it is, look, I think Apple made its bed, good or bad, that it's never going to be able to catch up on the infrastructure side. It's so far behind now, I don't know how they could do it. And also, their talent grade has been a problem. They haven't been able to keep the AI talent in-house. But I kind of in this economy where everybody's a little. bit risk off about CapEx. It's very interesting that if Apple can do what it does best, which is on design and delivering solutions that people love and use, does it actually matter? So I don't know, like I have my doubts about Siri, but I really still think Apple will remain the platform in terms of the product people use. And you know, Pat, I don't know if you see, we did see some next generation, the pendants and the cameras and the wearables. I want there to be a next generation platform because I'm tired of my neck crooked over looking at my phone all the time. But I'm still just not sure we're actually going to get there. I don't know. What do you think?

Patrick Moorhead: 

I mean, listen, I mean, Apple needed to be pragmatic. And, you know, you know, I think with the first announcement, I think maybe this got surface surfaced because the first deal was kind of the Apple personal AI that was running on its own ships, but it needed a cloud layer of that. And this would just make sense. I mean, they've got two or three choices. They're not going to make the CapEx themselves. They had a search deal with Google that was very much a cloud deal before, and this is an AI cloud deal, so I don't think it has changed much. The question is that how long can Apple go if cloud-based AI truly is the incredible value that it delivers to the user? How long can Apple stop doing that? I think it's going to be Yeah, this is a tough one, Daniel.

Daniel Newman: 

It was more interesting the second time I tried to do the topic than the first time.

Patrick Moorhead: 

All right. I went back, and we did talk about this in episode 290. OK.

Daniel Newman: 

Let's talk about something we're passionate about, Pat. You and I don't get to talk about things we're passionate about a lot. Perplexity computer sparks new AI operating system narrative. What is going on? Is perplexity back? Did you have like a monster tweet, by the way, just a monster on this topic?

Patrick Moorhead: 

Yeah, I did. I had like, I don't know. I don't know where it ended up.

undefined: 150,000, 120,000.

Patrick Moorhead: 

Oh, there it is. 130,000 views of it. So pretty much. you know, Perplexity was my number one AI app when it first came out. And probably a month ago, I just, I stopped using it because of the lack of value I thought it brought. You know, that month before I had used Comet to go and automate some stuff and then everybody blocked it. So, you know, low or no value, I was about ready to pull the plug. on it. And then, literally, this computer comes out. So, vibe coding is pretty much known by everybody who's anybody, right? You fire up codecs, or you bring up Cloud Code and you have it do its work. Now, the cool part about computer is that is a multi-vendor, multi-model, and it actually hits the models that it thinks are the best for the task. And obviously, for coding, it does Opus 4.6 right now. But you'll see Gemini that's really good, apparently, at going on and scraping data off the web. And that would make sense, right? Because of all the access that Google has and its robots don't get blocked. And there was a lot of controversy about that, by the way. before. And, and, you know, essentially, it's for dummies like me, who, who really don't know, you know, aren't comfortable with a CLI, but are very comfortable at, at telling my agent what I want. And this, this, this, for me has been the game changer. And you know what, Daniel, my, my strategy always was, and you and I had talked about this years ago, that I wasn't prepared to put the dollar investment into the infrastructure because I thought the tools would get good enough and soon enough. So listen, I have five to six apps that I've created personally about, not personally, I mean for the business. I haven't deployed any of them. And they're certainly not at scale. But what it does do, and I urge, who I think it's going to be best for, if you're an expert in what you do, tell it what you want it to do. And that's the big unlock for me. I don't have to have product management. I don't have to have IT to be able to help me prototype and even deploy. I just think that is an absolute game changer.

Daniel Newman: 

Yeah, I agree with that. Look, we kind of gave this topic ahead because you and I were both leaning into it, but we are watching a structural realignment, a disintermediation of something that has long been inaccessible to people like you and me. You know, there's been some low code stuff, but I mean, it was just garbage, like drag and drop website development. And I mean, it was like, you know, there was two schools. There were those that could code and those that could build. And then there were those that basically had to settle for the ugliest websites ever built. And now it's different. I mean, Pat, you can inspire it by things you like. You can inspire it by just communicating what you want. And every person on the planet has the opportunity to become exponentially more productive with these tools. Um, and I think that's going to be the opportunity. You know, I shared something this week, like, look, the era of. Of being gated by your technical aptitude has coming to an end. The era of being, uh, of creating opportunities based on your ability to envision the future is being opened up to everybody. And it's really exciting. Um, but it is going to create amazing efficiency. The period between efficiency and actual realized productivity is going to create some pain. It's gonna create some pain because, you know, Pat, you just built like 27 employees worth of work in like a weekend. I mean, and I'm not talking about like a week of work. I'm talking about years of work.

Patrick Moorhead: 

Yeah, it's absolutely insane, yeah.

Daniel Newman: 

And so, you know, the world is changing before our eyes. People's intellectual curiosity will be very important because I look at things that, you know, used to take, you know, and by the way, I'm sure you've had internal conversations, but like I'm having these conversations, I'm like, like something that like a report that would have taken to be built, like you would need like three people to work for like five days. I'm like, just drop all the spreadsheets, just drop all of them into computer or into in cloud and then talk to it. You don't need to organize them. You don't need to retabulate them. You don't need to, you know, put them into any sort of order or even into any sort of structured data. Just talk to it and it'll put it into order for you. And it will do it in minutes. It's amazing. I'm sorry, I'm crying. All right, I'm over. All right. That's good, man. I like that. Thanks. That's good.

Daniel Newman: 

So last thing we'll just touch on is Stripe released a preview of a new feature that could help AI startups solve the problem of passing through the underlying cost of AI model usage. Stripe is previewing a billing feature that helps AI startups track model usage and automatically mark up cost of LLMs, APIs, basically As we start to create agentic workflows and agents become our employees, right? I mean, how many, you know, Pat, more insights now is like 27,000 employees.

Patrick Moorhead: 

No, seriously.

Daniel Newman: 

I mean, it's 6,900 units.

Patrick Moorhead: 

It's crazy. By the way, what I love doing is having 10 running at the same time. And here's the cool part is when you've got persistent data coming in nonstop, it just brings it in automagically, right?

Daniel Newman: 

It does. Once you build the API or the data flow connectors, yeah, I mean, my CFO is like, I'm locked in through PipeDream through a bunch of APIs now, and I'm pulling the pervasive data out of Salesforce.

Patrick Moorhead: 

I got to build an ETL, I don't have one yet for NetSuite, but I've- I gotta tell you, when I looked at that PipeDream thing on a couple of times, I'm like, I don't know, should I give it access to this data? Yeah. Cause you can pull Google chat data.

Daniel Newman: 

You can, you know, I didn't give it like open, I'm doing it like item by item. Yeah. You know, but like the report is the reports that it's pulling for me is so good. I mean, So good. Anyways, I mean, look, this is going to be a problem that needs to be solved, you know, Stripe sensibly, given the fact that how you connect to payment mechanisms, a lot of the payments are remade on like credit card flows and stuff. It's a good way to potentially look at API utilization because these are basically your employees going to work. You can have some sort of management of it. Like, hey, how much credit can you run? Like how much over your monthly max plan are you at? Yeah. No, no.

Patrick Moorhead: 

No, I mean, within within five hours, I re-upped. I think you get like, funny, ironically, I was just on my credits page seeing, and I just bought $5,000 worth of credits. Just stop bugging me in code.

Daniel Newman: 

We're stopping work now. It's like having a contractor in your yard, coming to your door and being like, we're not going to stand up the next day. We're not pouring this concrete until you pay us. By the way, I know you've been berating your agents quite a bit and making them feel bad, that whole line of code about like, you know, if you screw up, I'm going to switch platforms. Have any quit yet? Have you actually seen any of any left due to your belligerent behavior?

Patrick Moorhead: 

Not yet, but we'll see. But how would I know? And it'll probably happen when I'm sleeping.

Daniel Newman: 

Yeah. Imagine if they unionize or something or start demanding like health care.

Patrick Moorhead: 

Yeah. That would be, that would be, that'd be tough.

Daniel Newman: 

All right. That's the, that's the decode everybody. I just, I just went there. So let's, uh, let's get into the flip is where we, uh, we choose a take, um, you know, on the latest industry news and we basically debate it and, and, you know, we'll flip a coin to see who's for today's, you know, Pat, we've, we've set this one up, but, You know, let's just let's just hit this one. Are we witnessing the end of the app economy? Is this new era of Opus and computer and open claw? Is this going to create the end of the app economy? Let's see who's for it.

Patrick Moorhead: 

Are you going to say something? It's you, dude. It's you. I know.

Daniel Newman: 

You usually say, hey, it's you. Oh, hey, it's you. I just thought we would go. I thought people were smart enough to know the thin face as opposed to the fat faced guy with the beard. We got to fix that. You all fixed up Pat and made him beautiful. You changed the whole introduction. You took all Fat Pat pictures out. Fat Dan is still in the queue. What's up with that, guys?

Daniel Newman:

I'm just kidding.

Daniel Newman: 

I'm not that sensitive. You're for it. The app economy is going to end apps. Sorry, the apps as we know it are over.

Patrick Moorhead: 

I mean, I am so sorry that this flip of the coin. I mean, this is the easiest one we can get. We have to look at the app economy in terms of the number of apps that are out there. The number of applications that I created in 40 years before Perplexity was one, okay? And it was in college where I had to do Fortran applications, or maybe it was two, because I had to do one in Pascal as well. So, if I apply that to today, if you look at the number of apps and just, you know, let's just say there's a hundred, right? Just to be easy. I think it's very simple to say that we're going to get to 10,000 very quickly. We've already seen spikes in the app store. on Apple. We're seeing GitHub repos just go absolutely bonkers with the amount of applications that are added there. We can't even currently count the number of applications that are completely being Completed you know with things like perplex perplexity computers and you know if if one control plane can coordinate browsing coding doc gen data analysis and tax execution. I think a lot of people are going to stop caring what the underlying app and completed work is. I'm not saying that enterprise SaaS goes away. This is not a SaaSpocalypse type of debate. But what I think it does say, if you can get access to your corporate data, you don't necessarily have to pay the value add for AI on top of that. If you can get an agentic orchestrator at the enterprise level, IBM's got a decent one. You can do that at the enterprise level. So I think it's just a foregone conclusion that we are witnessing the end of the app economy.

Daniel Newman: 

Maybe faster and they may be able to be developed, but the deployment is still going to be an issue. And look, if you and I can vibe code something in 10 minutes, then anybody can do it, which means there's no longer any real differentiation in what we're creating and there's no monetization opportunity. So the amount of differentiation and the level of upskilling and capabilities to utilize these tools correctly and find something truly novel that people are willing to pay for is much further away than most of us think. Sure, we can build cool things that are disruptive to the way things used to be. But in the future, this is all purely table stakes. Everything that we're building, nothing new, nothing novel, nothing anybody's going to pay for because they can all spend 200 bucks a month on perplexity computer and build it. But the net of it is that the app economy isn't going to be destroyed. It's going to be upgraded. And the work is going to be done to make applications better and incremental and differentiated. And we're going to finally see higher quality, higher value. And what we're actually getting is the next generation of productivity. But you're not going to vibe code something that's going to drive to billions of dollars. You're going to have to build it. You're going to have to differentiate it. You're going to have to create novelty in it. You're going to have to make it actionable. And you're going to have to make it scalable. And if it was that easy to do, everybody would be doing it right now. Everybody's building apps, but they're toys. They're not truly scalable, and the world's not going to pay for the crap we're all building in Perplexity Computer. So, no. No, no, no.

Patrick Moorhead: 

No, no, no, no, no, no, no.

Daniel Newman: 

Who won?

Patrick Moorhead: 

Did you win or you won? How many apps are you creating as we speak?

Daniel Newman: 

I'm working on tons of them. But like what I really do think is happening is like we're becoming productivity machines. But I actually do stand by my philosophy of like the novelty of anything we built that people are all going to pay for. Like, you know, because like if you build something super cool and you put it out on the App Store, you're like, this is amazing. I built this really amazing dashboard tool. But like, someone looks at it and is like, oh, I could do that too now. And then go on a computer and they can vibe code their thing. So what I'm really actually concerned about is more, it's like an app sprawl, is that everyone's gonna build super hyper-personalized apps at scale. But like, what's the monetization like of any of this? Like who pays for any of this stuff? Like what does it actually create in terms of economic value? when everything we've done has now been commoditized. Like just look at our industry, for instance, right? Like we jokingly, you joked about market models, right? Market models are truly novel. What was really useful about them was there was some gate, like you had to have some level of skill. You had to have some access to some unique data. You had to have, the belief that you actually understand a business or a market so that people would pay you for your development and consulting or access to these models. So you and I can go out, we can open source this, right? We can just open source and say, hey, I build this semi-model. We both did it, right? We actually both did this, we built one. And the point was is like, but there was nothing you and I did that was super unique, like that someone else with half a brain couldn't prompt the system to do if they're just paying attention. And so what happens is now is we get exponential volume of these kinds of tools out there, but like, then who do we pay? And you and I actually, I still think it comes full circle back in like this particular instance. So like people will still pay for the model that they think comes from someone that has unique insights, has data that no one else has, or has some layer of influence that their model will be more accepted widely into the market, even if you have your own version of it. Now, could you personally build a version of it that you use to trade on that you could make a shitload of money because you knew something or you had some insight. But if no one else cares about your model, the monetization is going to be very localized. So it's going to be super interesting. But the speed is just I mean, these apps are good. These apps are really good and they look really good. So I just don't know who pays for it. I don't know what the monetization looks like.

Patrick Moorhead: 

Yeah. And I think, um, at some point too, you know, there's a lot of, uh, free numbers out there at the top line. I wonder if those start to be, uh, reduced or just completely eliminated. Um, um, I'm super interested to see kind of, kind of where that goes. So for instance, you know, um, sell side, sell side research is, is pretty much available. You know, if, if you have a subscription or like, like what happens with that, I think that, uh, that could happen. Uh, but, uh, I don't think it will, you know, we saw the same thing in the internet, which was, Oh, I'm not putting any of my, any of my content on the internet. Right. I think you saw the newspapers, uh, do that and look where it got, got them. I think. first principles, AI will be so good, you can create it. And then you have to ask yourself, well, what's the differentiation at that point, Daniel? Like if everybody can do this, everybody's got access, I think it comes down to the people and maybe the data, right?

Daniel Newman: 

Yeah. I mean, people, data. And like I keep saying, like some perception of authority.

Patrick Moorhead: 

Yeah, I can't believe I can't believe I forgot about it. It's the people, the people actually people actually saying it, right?

Daniel Newman: 

Yeah. And like that, their, their, their opinion is valued. Like yours, your opinion is valued, my friend.

Patrick Moorhead: 

I think you might be in that ballpark, too. I don't know.

Daniel Newman: 

Who knows? Who knows?

Daniel Newman: 

All right, let's get to my favorite part. We'll talk about the markets. Bing, bing, bing, bing, bing, bing, bing, bing. All the little bulls and bulls and bulls. Let's just talk about bulls. F the bears, F the bears.

Daniel Newman: 

That sounds exactly like the New York Stock Exchange at the close.

Daniel Newman: 

Yeah.

Daniel Newman: 

Sounds like a little bell at my local bakery.

Daniel Newman: 

Bing, bing, bing, bing.

Daniel Newman:

 All right, well, busy week, we had a bunch of earnings. We had some broader market related news, Pat, but let's talk about the custom AI Silicon boom. We've got two companies in this space. You wanna start with Broadcom?

Patrick Moorhead: 

Yeah, Broadcom, what do we say like, you know, beat, beat, top line, bottom line. I mean, segment breakdown, semiconductors were up 52%, Right? If you look at the guide, crushing guide up 47% year on year, shares rose 5%, which you might be like, but wait a second, NVIDIA crushed it as well. And their stock went down, Broadcom crushed it and went up. Now, right after they brought out numbers, their stock went down. You get Hawk Tan on there. And what he did is he explicitly named gigawatt scale commitments. He brought a really good counterpoint to NVIDIA's $4 billion optical investment. I talked about that in an earlier segment, but I'll repeat it. The question people should be asking is, is NVIDIA adding CPO because they have a power problem on their switch? And if you used a Broadcom-based switch that is apparently lower power, you wouldn't actually have to do that, so you can have your cake and eat it too. You've got the low cost of power, but you've got the high performance as well. I think it's going to be a big debate coming up in AI networking technology with Broadcom and NVIDIA on both sides. The other thing that Hawk poured water on was this whole analyst reports about Meta's MTIA silicon program. I called this before. I said, you know, basically last week, it's not canceled. It's alive and well. And as you go into multi-year, multi-year I think it gets even more likely that there'll be a higher volume. You know, infrastructure software missed, but nobody cares. Nobody cares. 6.8 versus 6.99 on the consensus. And really, VMware Sorry, software right now is primarily an adder, a stabilizer, right, as opposed to something that's going to be this huge, you know, huge driver of revenue and growth.

Daniel Newman: 

Yeah, I mean, look, you hit it pretty well. I agree on the MTI. I immediately said that's just nonsense. That is not canceling anything. They are. augmenting for that immediate capacity need that they have, and they are using what they can get their hands on. Okay. People just love to suck up FUD and just share it liberally. And, you know, they're perpetually wrong. So, but yeah, I mean, Broadcom did the business that hundred billion dollars of 2027 AI revenue is juicy. I think the stat I gave was the Broadcom's profit in 26 is higher than Broadcom's revenue five years ago. And they have 68% EBITDA margin. Gotta love that kind of business. Yeah. So another company that's got, they got the God candle was Marvell. Shout out to Matt Murphy and Chris Koopmans and team over there. You know, they had about a three year where this period was flat for them. And this was a pivot as they were moving their business from diversified, more diversified to more data center centric. But a couple of big numbers, I think their data center growth expectations here is 40%. Interconnect business, 50%. They went from like 5 billion last year in this period to eight. uh for the year five billion you know full year to eight billion some massive growth Huge upside. People have been obsessing with this business based on its accelerator business, but its interconnect and its connectivity business is really where all the good juice is. It's not that their chips, their AI chips don't matter. It's just people over-rotated to that as important and underestimated. And I mean, again, did you get the confirmation with these various deals on optical? Like, you know, the portfolio at Marvell has been built to add all kinds of of attached content to these systems. And by the way, there are situations where you might have a Marvell Accelerator and Broadcom network content. There's situations where you might get a Broadcom Accelerator and Marvell attached content. Like people kind of want to make it one thing, it's not. And so my overall read on it is just, it was a really strong year. It's a bit of a coming out party and Marvell, kind of over the negativity was a bit overplayed. And clearly, this is a this company has been from the AI CapEx build out.

Patrick Moorhead: 

Yeah, one of the reasons that Marvell stock tanked suddenly, a while back was because you had things shady. I thought I thought they looked shady stuff going on in Taiwan. And I'm putting together an interesting analysis showing how Morgan Stanley, Fubon, and a couple other guys over in Asia, how they set an ALCHIP number based on what the CEO had told them. They've adjusted the guidance down many different times. It's laughable what happened here. If you compare even Alchip and how they've done, you compare that to how Marvell did. right? I mean, you know, Al chip just looks like an absolute, um, let me just put it this way. It doesn't, it doesn't look great and it doesn't tell a story. And some people got burned on the way down, but I think they're, they're getting, um, they're getting their money's worth if they held on the way back up. Listen, and on the XPU portion of Marvell, regardless of exactly what they do for the chip, they're going to have content, right? And there's three ways that they get in here. There's some IP, but let's say an IP block that gets hardened into a design. There's a chiplet that is also a potential. And then there is the back-end design and taking the supply chain into TSMC. So I'm convinced that regardless at AWS, whether it's the, whether it's Tranium or it's Maya, Marvell will have content here. So what is that set up, right? Scale up networking is the next battleground for Marvell. like the Broadcom competitive threat at 1.6 T is real, but manageable. You know, the TSMC supply chain positioning, I think is a hidden advantage in certain areas versus, you know, versus certain competitors. And I think, like you had said, right, this custom XPU attach is really an underappreciated growth vector out there. And I do still think this valuation disconnection persists if you just roll the numbers here. So congrats. I think the stock's up 16%. It's up 16% today. Congratulations.

Daniel Newman: 

On a pretty red day, by the way.

Patrick Moorhead: 

Yeah, congrats to Matt. And by the way, Matt went right after some of the stuff that I've addressed. I think you should all look at the transcript to get the Marvell point of view and not just mine.

Daniel Newman: 

I only need yours, to be clear. Don't lie. It's unbecoming. All right, a couple more. Let's rapid fire because we're coming to time here. CrowdStrike, Pat. CrowdStrike.

Patrick Moorhead: 

You're a cyber guy. No, no, I am now. AI is going to displace all of cyber. No, I mean, it is. Just wait. Anyways, I had a great conversation with CrowdStrike CEO at GTCDC. He's a good, he's a good dude. Right. So, you know, they had a double beat And, you know, stock declined 4% after hours trading. Guidance was in line. It was not above consensus, which I think really disappointed investors. And I think they're going to have to fight the slog of why AI isn't replacing them. I would say to a lesser degree than let's say, uh, an Adobe or a Salesforce or a workday or an SAP, but they're still going to have to, um, uh, fight the, uh, fight the battle. So, you know, okay. CrowdStrike proved to survive the outage, but the stock price is asking basically, can it run outrun the AI disruption wave? Maybe we'll recreate CrowdStrike inside of a computer.

Daniel Newman: 

Let's give it a shot. Let's give it a shot. Yeah, I mean, look, the numbers in the guide indicates that AI is not yet going to completely erode it. My big thesis on all this software stuff is just watch the RPO. Watch the RPO. Because if it's happening, you'll see it in an erosion of RPO, because multi-year contracts will show up and will continue to anchor quarters with these subscription businesses. But if the RPO starts to get pulled back, that means in the future, you're seeing defection, you're seeing attrition. And if that happens, that's when the risk comes in. Not saying anything specific to CrowdStrike, just more of a general comment about software companies and the possibility of their risk, because most of them have multi-year agreements with many, many of their customers. ServiceNow, another one, by the way, has a nice, I think it's a back about 20% after, you know, like a 50% bleed out. Like this was a, so it means they're going down 40%. But you know, it was great to see Bill McDermott step into the open market and buy the stock at 54% off the highs. always a great signal to send to your investors and to your community and to your employees about stability of the business. I know there was also a number of its employees canceled planned sales to hold on to their stock longer. Look, you know, this is leadership. I will commend, I think Lisa Sue's done some of this recently. One of my personal holdings, SoFi CEO, Anthony Noto bought a million of stock last week at the market price, not some RSU deal. I like what that signals. I like seeing CEOs step up and say, my company's undervalued, I'm gonna write a check with my own money. Not a buyback, it's not using the money, it's not a grant, it's not some sort of options plan, it's literally money they've earned, paid taxes on, and they're putting it back into the business.

Patrick Moorhead: 

Yeah, I mean, listen, when the CEO of buys $3 million of its own stock at a 54% discount, when every other SAS executive is like running for the hills, I think everybody should pay attention. Right. And I think Have there been circumstances that people have done this and gotten burned? Absolutely. But the thing I think people need to consider about ServiceNow is that it's not a traditional SaaS company. It works on workflows. It works on workflows that connect the back end to the front end, inside joke, Daniel Newman, and let's just take an onboarding of an employee, right? It works ITSM, sorry, it works from their own ITSM to Workday and SAP and it orchestrates that entire workflow. That's a classic one. And oh, by the way, it seems like that ServiceNow would have a lot of data of how it connects different applications and workflows that could make one of the better agents that you could put together. Because if we do believe that it's partially about the data and customers are looking to fast track the capability, it seems like that would be a good choice. All right, Pat, we got one more topic.

Daniel Newman: 

Amazon's CapEx, which by the way, I could replace it with any of these companies' names, maybe barring Google, is sending its stock to the worst month in years. Are they being penalized because of their big CapEx expense? Is that fair?

Patrick Moorhead: 

So I don't think it's fair, or I guess it depends on if you're a glass half full or glass half empty. I think it's more about the magnitude of it, right? They plan to double CapEx by the end of 2027. I think that's that's what got everybody going. I think the $12 billion Louisiana data center is just one example of this. And I think you're right. This pattern really is identical across Microsoft, Google, and Meta. So I think you absolutely nailed that. So, you know, do you believe in AI and do you believe in Amazon? I think what people are just missing is Amazon got no credit for Tranium and now it should. Right. I mean, everybody's darling is Anthropic right now. And guess what? Anthropic runs more on Tranium than it does on anything else. They just knocked out a brand new OpenAI deal, which, by the way, I think is differentiated from Microsoft. It's running core frontier models. They're the exclusive provider. of it and OpenAI is going to make their stuff run on top of Trainium. So I think it's a big, big deal here and the market is probably getting it wrong.

Daniel Newman: 

Yeah, I tend to agree. It's a, you know, weighing machine and a voting machine in time, it will be right. I mean, people worry because AWS's margins eroding hurts the fact that Amazon really doesn't make a lot of money from its core business. And AWS has been the golden goose for a long time. But I think this is also a necessity. So maybe it's getting repriced, but this is what the company has to do. And I think in time, as long as they can show they're monetizing it, They will be fine. Pat, we did it. We made it through another docket. I was boring in the beginning. I bored you. You fell asleep. I'm waking you back up with all this really interesting talk about vibe coding. That's all we needed to do. We appreciate all you out there that have stuck with us through this entire show and through all of my patantic behavior and Pat's amazing handsomeness. Hit subscribe, join us for all of our other content. Check us out each week. We're gonna be on the road. We're gonna be at events. We're gonna be at shows. We're gonna be busy, busy, busy. The next big one's gonna be us at GTC, but we'll be back with you again next week. So appreciate you tuning in. See you all later.

‍