The Six Five Pod | EP 291: Davos to Abu Dhabi - Inference, Codex & the So-Called SaaSpocalypse

The handpicked topics for this week are:

1. Inside Abu Dhabi’s Full-Stack AI Play: From universities to healthcare to hyperscale infrastructure — Pat shares a firsthand perspective on how the UAE is quietly building an end-to-end AI ecosystem.
2. Optics, Cooling, and the Hidden AI Infrastructure Layer: Why companies like Coherent matter as much as GPUs — and how photonics, co-packaged optics, and rack-level cooling are becoming critical to scaling AI factories.
3. Inference Takes Center Stage: Microsoft’s Maia 200 shows real progress — and why hyperscalers are building custom silicon to boost capacity, economics, and control.
4. NVIDIA’s $2B CoreWeave Bet
   Circular finance or strategic genius? We unpack what NVIDIA’s latest investment signals about AI factories, cloud capacity, and long-term infrastructure buildout.
5. Codex vs. Claude: The Coding Wars Heat Up: OpenAI closes the gap fast — and developers start hopping between tools as AI coding becomes a moving target.
6. The “SaaSpocalypse” Narrative: Is software really dead? We separate market panic from reality — and explain why SaaS won’t disappear, but will never be valued the same again.
7. Cisco’s AI Summit Reality Check: From hype to execution: what stood out from Cisco’s AI Summit and why networking, security, and enterprise integration matter more than demos.
8. Are AI Agents Enterprise-Ready? The Flip Debates: real-world workflows vs. reliability, governance, and security — where agents work today, and where they still fall short.
9. Big Tech Earnings Whiplash: AWS, Google, Microsoft, Meta, NVIDIA, AMD, Palantir, and Coherent — massive CapEx, cloud acceleration, and what Wall Street is getting wrong about AI ROI.

‍

For a deeper dive into each topic, please click on the links above. Be sure to subscribe to The Six Five Pod so you never miss an episode.

‍

Listen to the audio here:

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 we may even reference that fact and their equity share price, but please do not take anything that we say as a recommendation about what you should do with your investment dollars. We are not investment advisors and we ask that you do not treat us as such.

‍

Daniel Newman:

It's been a minute. It's been a minute or two. This is Episode 291. Six Five is back.

‍

Pat, I'm out of practice. It's amazing. They say it only takes what a few, a few weeks to create a habit. It only takes a few days. It seems to get out of a habit, but wow. I know our audience has been clamoring. My, uh, my inbox is blowing up. People said, what happened to you guys? And I said, Patrick is over in the kingdom doing important guy stuff. And so left me no one to pod with. And they were like, you should do it alone. I'm like, I would never. do this alone. It would never be as fun to do this alone. It's such a horrible podcast. I mean, there's a chance it could be better if I didn't show up, but without you, it would be a total nightmare. So, but even on the road now, like the first good, almost more than half the first month of the year, you're going for some sort of land speed record.

‍

Patrick Moorhead:

I mean, listen, we’ve both been on the road, right? You and I came off CES, went to Davos. I came home, packed my bags, hit Seattle. I wasn't in the kingdom, I was in the UAE. So in the general vicinity, but yeah, it was a great trip and kind of understanding kind of how the ecosystem works. And it was well worth it. Did you stay in Dubai or did you go more than Dubai? I stayed in Abu Dhabi, right? So UAE is the country and you've got Abu Dhabi and Dubai. I did hit Dubai airport, but I didn't spend, you know, I spent nearly a week in Abu Dhabi going all the way up the tech stack, you know, all the way from kind of deep tech to universities, MBZ-UAI that creates the models. And then you've got the applications of the technology, huge in healthcare and genomics and near and dear. It's basically putting science to stuff that you and I are shotgunning with related to peptides and getting all this body work done. really turning that into a science. Then you've got the G42 network, which is a full stack of infrastructure. I got a, you know, essentially stood in front of the model, the same model that President Trump stood in front of, of the five-gigawatt project that G42 is building in Abu Dhabi. So yeah, really great experience there.

‍

Daniel Newman:

Well, that was the third big week of travel for the year. It's good to have you back. It's good to have you back stateside. I hope you enjoyed those 18-hour flights. No gyms on the plane. I hope you took enough peptides. Nowadays, you can take that peptide that actually exercises for you. I'm waiting for that one. No, it exists. It's SLUPP232, I believe is what it's called. It's available on market. Your body just exercising all the time. Gets you super lean and jacked. From what I hear, I don't know.

‍

Patrick Moorhead:

I'm going to call the DN. I'm going to modify my camera. It's bugging me a little bit. I thought it'd be called the DN200. The Daniel Newman, you know.

‍

Daniel Newman:

I'm here for you. It's been a year. I'm optimized. No, let me just say that I'm optimized. So buddy, we have a ton to cover because we're going to try to do a little bit of like the bonus episode. We're going to go back in time a little bit. A lot of important stuff happened. Yeah, I mean, you know, it feels a little late to really talk about Davos, but that was pretty cool. There was a lot going on there. But man, it's been a whirlwind. Markets been in a whirlwind. We've had a crypto meltdown. We've had AI is going to revolutionize every industry. It's going to eat software. It's going to destroy services. And then we also have, well, does it even work? Yes. And by the way, we might even have a little flip to talk about how well are the agents? Are they are they agents? Are they actually collaborators? Are they helping us? When will they be helpful? But a lot going on the chip space. We had another big tech earnings wave. We had a total market meltdown. What else do we have happened? I won't mention the broader macro news or other news and leaks in the market. I don't want our pod to get censored, but all kinds of stories that broke over the last few weeks, Pat. Yeah. So let's get after it a little bit. We're going to run through the decode, as everyone knows, and then we're going to do our little flip, and then we'll get into bulls and bears. We have a long list for the decode. So I'm going to, without further ado, I'm going to get this party started. All right, man, we've got, it looks like at least five things on the decode. We're going to run through, we're going to talk a little bit about some new inference chips. We're going to talk about NVIDIA investing in CoreWeave, we'll hit OpenAI Codex, we'll talk SAASpocalypse, and then a quick breakdown of the Cisco AI Summit, which was my only day out on the road while you were gone for nine straight. But let's start, Pat, talking about Maia 200. So Signal 65 has done some work and broken this down. And there'll be more Signal 65 for those of you that don't remember is Pat and I's love child. It's our technology lab where we test performance and measures of merit of new technologies. There's a lot to say there. Microsoft has been heavily criticized for being a little late to the infrastructure party. It sits behind, of course, Google with its TPU, Amazon with its third generation coming on fourth of Tranium. This is the second pass of Microsoft's AI chip and it's powering some interest is doing inference powering some some leading models including those from open AI. And of course, to be clear when I say criticism criticism for not building their own but have done incredibly well in terms of building out the infrastructure for everyone else. Pat, you know, this is another story of expected progress as I see it, like Microsoft, all these companies are going to follow suit. They're all going to build their own infrastructure. They're all kind of on slightly different schedules. It does take a few generations. It seems to really harden and get these chips right. It's good news for Microsoft as, you know, capacity is not unlimited for, you know, compute. So of course, Microsoft is going to go down this path. But I think the most interesting thing that keeps happening is they keep trying to make a race out of this, a competitive situation where it's, hey, it's Maia 200 versus Grace Blackwell versus TPU. And in my opinion, this is all about every and all. and Microsoft needing to basically access as much capacity as it can. Building its own gives it the chance to optimize and harden a few workloads on its own platform that it can benefit from some better economics. But the whole it's Maia or NVIDIA argument to me is still silly. It's still kind of stupid. It's not at that space just yet. But I will say that our initial numbers and findings is Maia 200 is a pretty nice leap forward from the first generation.

‍

Patrick Moorhead:

Yeah, I mean, the OG of hyperscaler silicon is AWS. And then you've got Google with TPU and a whole lot more. And, you know, Microsoft was a little late to the party. And First version, I didn't expect it to be, you know, swinging around the room. I mean, nobody hits with their first chip. Nobody does. And second, the second Maia 200 was definitely better. Some pretty awesome 4-bit performance, you know, when put in certain circumstances. There were 30% price performance numbers that even Satya put out. And there's two ways to look at this. And I like to look at this long term. If it can reduce cost significantly, it can put Microsoft in the driver's seat, and it can architect a solution that works best on its software, I think you absolutely have something. And if you're optimizing it for internal workloads, OpenAI, and let's say running Claude on it, you have something. And an API that's external for enterprise is a whole different ballgame. And I think that is kind of the long tail here after this gigantic training build gets put into place. AWS has put a lot of effort into Bedrock and other types of services to essentially hide the complexity of its accelerators. And the good news is that Microsoft has enabled more kind of industry standard like VLLM access, but we'll see. I mean, It's, you know, it's kind of a, um, uh, I don't even know what you call it, you know, love Jensen, um, but compete with Jensen at the same time.

‍

Daniel Newman:

They all have to add capacity in there. You know, I always say it's two things. It's it's adding capacity and it's economics. There's just going to be certain parts of their stacks that they want to optimize and they want to create the best economics and margin. And they also need to hedge a little bit. No one wants to be 100 percent beholden to a single company. We see how sensitive supply chains are. So, yeah, speaking of Nvidia. You know, it's investing 2 billion into a different cloud. You could argue that it's a cloud that's 100% committed to its hardware. It could be considered a cloud that it has a heavy investment interest impact. You could also say this is a cloud that this is another circular finance. What's your read?

‍

Patrick Moorhead:

Is it circular financing? Yes, this directly relates to NVIDIA investing two billion into CoreWeave. And, you know, we talk a lot about what customer zero means, but but if you're not running what your customer zero actually does, then you get somebody else to be your rabbit. Right. Everybody needs a rabbit. And CoreWeave is NVIDIA's rabbit for their software, for the latest chips. And, you know, first with Grace Blackwell and GB300 first with Vera CPUs. And you heard that right. We heard Jensen at CES the investor day talk a lot about CPUs as an actual standalone business. And people were wondering what that was. And I know Arm has talked a little bit about it as well through the context of NVIDIA. But yeah, I think the comment was there's over 10 customers that are going to be buying services that run on top of the CPU itself. And we've seen when it comes to agents, when it comes to reinforcement learning, the CPU is very much an important piece of this pie. And also, you know, we talked about it at the phone at lunch, right? I mean, you need not shades of gray, and you need black and white, you need higher precision. CPU is the way to go. And there's a lot of elements of this code creation that actually runs on top of a a CPU. One part of this that got kind of glossed over is partnership targets over a gigawatt of AI factory by 2030. That is absolutely a gigantic amount of investment here. And markets saw it, they liked it, and they boosted CoreWeave stock.

‍

Daniel Newman:

Yeah, absolutely. Not much to add here. I just keep saying, look, smart corporate venture has always been about investing in companies that are going to basically leverage and benefit from the skills, tools, capabilities, software, hardware, tools, technologies that you have. That's a company that's 100% committed to building on NVIDIA. So it makes sense that NVIDIA continues to back it. There's some concerns and questions about its financial model, about its balance sheet, about its depreciation, about how it's doing its business. really dependent. But if the AI build out is seven or eight or nine years, like Jensen's saying, then it's going to have a long time to grow before those those particular issues are going to come home to roost.

‍

Patrick Moorhead:

For sure.

‍

Daniel Newman:

Let's talk about coding.

‍

Patrick Moorhead:

Okay, so you know, we're big coders, aren't we?

‍

Daniel Newman:

You and I are app builders. Also, I also did add to my LinkedIn profile full-stack engineer. I built an app this weekend, a real app. Wow, we should build an app to figure out what our best moments ever are in our podcast. And then we could create a video loop of those moments. Anyways, so OpenAI fell way behind, I think, you know, it was sort of almost felt like two weeks ago that Cloud Code had taken over the world and that, you know, while there was little bits of OpenAI and a little bit of grok for coding that Dario and the Anthropic team had run away. and that the rest of these companies are being left in the dust. Well, you know, just a week, two weeks later, Codex 5.3 is announced. I think it actually happened when I was at Cisco Live. He was on stage and we were talking about Cisco AI Summit. Very interesting. I mean, this is very much a GUI driven coding environment. It's a little different than cloud code, which focuses on terminal app, you know, developing. So you still need a little bit of understanding of what code environments look like. And that seemed to do really well with the coders. But I saw something today that actually showed in a very short period of time, Codex has, it's not completely, but close the gap. I saw a chart that showed like Codex closing the gap meaningfully on Cloud Code, which is really interesting to me, which to me kind of is a read that we're in this new era where people are willing to bounce from thing to thing. And we do this with models, right? And developers are doing it with their new AI development environments. And they're literally willing to go from cursor to Cloud, to Codex, back to Cloud, just like we're bouncing between Gemini and everything else. And like, it's odd because in the historics, like tribalism has been super important to tech. I'm a tribal, this cloud person, I'm a tribal, this infra person, I'm a tribal, this kind of software. It seems like it's going to be completely different in an environment where you are basically out the box, a full stack engineer and able to build. You know, I think a lot of it is really going to be about You know being able to call the right functions and these apps will talk about we talk about agents but the agents with these different tools having you know good ways to build applications that are able to basically successfully API into other things that's really how all this works and build like really highly useful tools that scale beyond kind of the prototypes that. You know, I'm mostly building prototypes right now. I'd say what I'm building isn't really ready for prime time, but they can really follow rules, rails that can connect to security. They can manage the infrastructure. A lot going on, Pat, but look, it seems like it's a really fast moving thing. And as soon as 5.3 codex lands, I expect Claude will update something or Grock will update something. I heard Musk followed the cursor guy this week, and now there's rumors that Musk is going to try to buy cursor. So it's changing fast.

‍

Patrick Moorhead:

It is, you know, the competition is good and this classic moving target, right? Remember when we were comparing LLMs, right? And then we were comparing agentic capabilities. And then now we're kind of getting in this this vertical application, right? Because this is to help developers basically code, have newbies like you and I create complete code and not even really understand the code underneath. And this is GPT 5.3. I wouldn't say they're reacting to Anthropic, but Cloud Code was Anthropic's sweet spot. It was really the vertical that they excelled in. So you'd expect them to be good at this. Meanwhile, almost nobody is talking about Google's version of this. Nobody's saying, oh, no, I forget. Gosh, is the name Synergy? It's terrible. I don't remember the name. I've never used it. I should know this. I do use Google Enterprise Agents. So. Us too.

‍

Daniel Newman:

Us too. Well, it's moving fast. It's changing fast. Speaking of moving fast, I mean, you know, we got Multbot. We got, you know, instant code to build your own apps. And then apparently, um, SAS was over this week. It's it last week.

‍

Patrick Moorhead:

Yeah. Yeah. We'll talk about, we'll, we're going to debate the, uh, the reality of this. So I'll just stick to the facts on this one, but, but essentially, um, you know, Anthropic’s, you know, latest model basically said, uh, software's dead, right. Any equations that you have. uh throw them off and and essentially uh you know sas is down 25 percent and i since i think since the the beginning um of of the year a lot of debates on replacement versus augmentation a lot of discussions i even got in with chamath polyhapitia on the realities of enterprise software. I was happy that the Chamath even responded to me because he doesn't follow me. I don't know how he even how he even saw that. So he doesn't follow you. I know. What's the guy's problem? Can you believe it? Maybe it's because he's a billionaire and I'm not a billionaire. I mean, I'm off by a big factor, buddy.

‍

Daniel Newman:

Okay, you're close.

‍

Patrick Moorhead:

Yeah. The funniest part was Renee Haas called the panic micro heart heist hysteria. I thought that was pretty funny. And, you know, the debate was also about AI, you know, real threat not being AI, right? It's your CRM, it's unbundling, right? Features migrating into co-pilots, bundles and data platform. It's, you know, it's the classic, give me perfect data and look at what I can do. And this is by the way, just to be, this was Anthropic Clot Opus 4.6 that we're talking about here.

‍

Daniel Newman:

Well, I mean, look, like you said, facts, I'll make it two things. SaaS is not dead. It will probably continue to grow. Every report that came out, we'll talk more about this with some of the earnings, actually continues to be strong. Business fundamentals are there. There's two reasons it's not going to get replaced by AI, but it will end up being more of a collaboration with AI. One is the rules and rails of business require some sort of structure. And that is still much more well handled inside these enterprise software tools. But the other side of this is there will be a forever rewriting of software companies in the market. Their valuations will never be what they were. They will never enjoy the same forward multiples. But remember, Nvidia trades at just over 20 times forward. Microsoft and Google are just under 30 at most times, 20 something to 30. Some of these were 50 and 60. So you know what I said? Welcome to the party, guys. You're going to have to fight and compete like everybody else does. But the actual structural business, I don't think, is actually in any real danger. The rerating means they're going to have to do more, grow more, be better, perform better, and continue to do so for a long time to maybe get back to some of those valuations. But the fact that people keep saying it's going to just get eliminated is really, really underestimating how deeply entrenched a lot of these softwares are.

‍

Patrick Moorhead:

I think there's going to be a lot of consolidation, particularly people who don't have suites. I think kind of the workday and Salesforce looks pretty interesting. I think it'll happen. There's less risk to ERP. There is risk to ERP, don't get me wrong, but are you gonna risk somebody vibe coding your piece of software that runs your entire manufacturing and logistics and supply chain?

‍

Daniel Newman:

Transactions, the money flow. No, but by the way, I do think like, I don't know if you saw like CNBC's Deidre Bosa made like a copy of Monday, like .com, like Sunday morning. Like, I think those businesses might be fucked. I really do. I think like the things that has no underlying data, Pat, Like they have no meaningful data moat and they're just sort of in GUI. Those businesses seem more replicable. Let Pico try, I bet you he could do it. Probably. He's really smart. Patrick Jr. is the, Patrick's the, Pat's younger, smarter, better looking version.

‍

Patrick Moorhead:

Very much, taller too.

‍

Daniel Newman:

Yeah, he's got all the goods.

‍

Patrick Moorhead:

He really does. Anyway, all right.

‍

Daniel Newman:

This isn't just a show about Pico, back to us, the most important characters in our movie. All right, we got one more topic. We'll hit this one quick.

‍

Patrick Moorhead:

It's not always about me, but you know.

‍

Daniel Newman:

We haven't used that we haven't used that as much lately.

‍

Patrick Moorhead:

We've sort of lost our I'm gonna go back and watch that series again. Yeah, that was when you were Fat Pat was that more Fat Pat era thing that you said or yeah, by the way, I got I got my license today And I got it renewed and thank goodness got a new picture taken. I was so fat in my picture embarrassing You're gross. So gross. It's like Jabba the Hutt, you know? We're just kidding. We're not. I was a fat pig. I know. I know. But listen, I call myself a fat pig.

‍

Daniel Newman:

I mean, you can sell. Don't call yourself that. You can call yourself Fat Pat. I'm fine with that. All right, Fat Pat. I mean, Skinny Pat, whichever one we're on. Very quickly, Cisco did have its 26 AI Summit and it was a barn burner. It had the who's who's of who's who. We had Andreessen. We had Sam Altman. We had Jensen Huang. You kind of go down the list. I mean, CEOs after CEOs, after VC, after founder, great day. And it was a really, really thoughtful. What I loved about it, just my TLDR, and I know you watched some of it from remote, I watched some of it in the room, was it was like a TED talk. It really was, I thought Jeetu and Chuck did a really nice job of hosting conversations and they weren't like product vomit. It was a very intimate room, maybe a couple hundred people max in the room and it was really thoughtful. So appreciate the event, Jeetu is a good friend of the Six Five, appreciate him having us and I'm sure you were bummed out to have missed it.

‍

Patrick Moorhead:

Totally, yeah. In fact, one of the guys I was supposed to meet over in KSA, the reason I didn't go to KSA was actually there, right, Tarek from Humane. was interesting, but no, it looked like a great event and I wish I were there. And I'm not just saying that folks, like this is something that you wanted to be there for. And I am really, if you think about Cisco, I mean, two years ago, they were just being pummeled by people saying, you know, they're going to be out of business because they're not part of the AI trade. And again, I know there's a difference between getting people on stage and doing file leadership, which Cisco has always been good at. In fact, they invented the concept of file leadership when Chambers was there. And doing it. And now, right, they've got three basic pots of goodness for AI. And they're connected very much so to not only the hyperscalers, but also the Neo clouds. as well, and they're plugged into the Middle East and all the growth that's happening there. So I think these guys are on fire. And you've got Jeetu, who, quite frankly, has completely changed the whole culture inside of engineering and the product delivery vehicle. You and I both got the chance to spend time with the management team there at Davos. And the teams never look better.

‍

Daniel Newman:

Well, they're executing well, and they're finding a key role. We have Enterprise AI as one of the trend lines for this year, and that's a company that stands to benefit across security, across networking. It's definitely moving, and it's finding its way back into compute after a bit of a hiatus, and that's going to be a happy part of the business to watch for sure. All right, Pat, it's been a while. Are you ready? It's been a while. I think I hear the debate since we've debated with anybody that debates as well as I do. I mean. Is there anybody you debate that debates that it's like King?

‍

Patrick Moorhead:

Everybody just agrees with me, Dan.

‍

Daniel Newman:

Sometimes you got to wonder why that might be. Why would everyone agree with you?

‍

Patrick Moorhead:

Some people say you're a genius. I'm just like, that is the that's silly. You know, you should if you were going to be scared of me, it would have been when I was in my 20s.

‍

Daniel Newman:

And I think you're scary over a keyboard 40s. You're progressively tougher the farther away you are from being physically in person.

‍

Patrick Moorhead:

Wait a second, don't like teenage boys, like, you know, anonymously, you know, shit post.

‍

Daniel Newman:

And yeah, for sure.

‍

Patrick Moorhead:

No, I mean, basically like a shit poster, but a shit talker.

‍

Daniel Newman:

No, I mean, like, you know, I've had you and I dear friends, besties, but like over the years, we don't agree on everything. Like we've had it out before. And like, Pat is a very reasonable guy face to face, but he's definitely can be a little he can hammer your ass over a text message. So I'm just saying that's like, you know, we all have our things. You are you are tough when when asynchronously, maybe that's not it's not further distance. You're just tough asynchronously, I think, is what I think it is, too.

‍

Patrick Moorhead:

There's the human element, right? It's it's the you know, it's easier to separate the humanity from the rage.

‍

Daniel Newman:

It's the guy, what do they call compartmentalizing? You're like in your moment, you're mad over here. And then you see me and I come up and I'm like, bestie, give me a hug.

‍

Patrick Moorhead:

And you're like, I can't be mad at you. Listen, you are the mature one in this relationship. I do try when it gets heated, you call me, right?

‍

Daniel Newman:

Appreciate usually. And I missed you. Like I didn't say that in the beginning, but I did miss you. Like, You know, three days away, like, you know, it's just like any relationship. I was like, oh, I can live without Pat. But after like five days, I was starting to just get genuinely sad.

‍

Patrick Moorhead:

And that was. Yeah, I was I was very lonely. Have you back past couple of days? Well.

‍

Daniel Newman:

You got a friend in me. I feel like that's a song or I think it was Taylor. I wasn't name the artist. He's a singer, something Taylor sang the song. You know, you've got a friend. I can't think of his name, but he's a very famous folk singer. Anyway. All right. James Taylor. Good God. That was terrible. James Taylor. Great, great voice, by the way.

‍

Patrick Moorhead:

It's good for a millennial. You're not even a boomer. Are you a boomer? Definitely not a boomer. Definitely the forgotten generation, you know, me and Michael Dell and Elon Musk, a list of these people, Brad Gerstner, David Sachs, we're all Gen X.

‍

Daniel Newman:

All right, well, I'm almost Gen X. I'm very Gen X at heart. All right, we've gone off the reservation. I'm gonna bring us back here. We are going to talk here today about whether or not AI agents, so basically, there's a question, okay? Are AI agents ready for the enterprise? Let's do the flip. Look, this one's going to be an easy win for me.

‍

Patrick Moorhead:

So you pick, they are ready.

‍

Daniel Newman:

I think they're ready. And there's no better way to know they're ready than when you're actually building something. And while I might be marching around town saying that I'm building apps, I am not really building apps. But I do have a team that's building the future of this particular industry. And the future is going to be agentic. In our business, it's the future. It's the future. Well, I named the company before this was even a thing. I knew it was going to be this. But, you know, bottom line is, is in an industry where you have so much constant signal and you've got to filter through the noise and you've got to immediately provide analysis like it's the perfect workflow for agents. Identify what's important, write about it, publish it, be able to make sure that the quality of the research is good and that it uses unique and Proprietary information by the way it does it all securely it manages the SEO all the stuff is agent workflows and you know we built this we tested this and it works now. What i'm gonna say is this anybody that says agentic AI isn't ready for the enterprise at scale. is probably right, but it's probably not the technology. It's the people. It's the people. They don't know how to use the tools. They don't know how to use the right software. They don't have the right data. They don't have the right workflows. And really, in the agent world, flows are everything, because an agent is really just taking the right tools and being able to pair that with a generative AI model and be able to take it through the tool progression to do a series of things in a way that it is designed. It's not that different than workflow automation, but it's just at the next level now with these AI these AI agents being able to make more decisions, have more autonomy, but again, all inside the rules and rails that you need to create. So the question is, is building enterprise AI, and by the way, this is whether you're using it through ServiceNow or Salesforce, or whether you're using it on something that you built at home. The question isn't whether or not any of these work, the question is, can they work at all? And the bottom line is, they do. And here's why I really think they work, and I'll break it into five things. The first thing is, models work well now. Everybody that keeps saying, oh, the hallucination. Look, if you haven't used an AI model in a year or two, you might still be able to point to hallucination. But if you're using anything newer than GPT-5, Claude 4.5, Sonnet 4.6, Opus, these models are incredibly good. You feed it enough unique data from your own business and it can do really, really remarkable things. The infrastructure layer, it's no longer immature. We have mature models, we have mature frameworks, whether that's crew AI, whether that's Ling Chan, Anthropic has built the API ecosystem. This stuff all exists and basically has come so far over the last 18 months. The companies that are really out in front with this are already showing real meaningful ROI. You've got companies like MercadoLibre, you've got companies like Amazon, you've got companies like Meta. They're using these in their own workflows and they are printing cash, Google with AI. And lastly, I'll just leave it at this. I think the risk is real, but it's there. But that's where partnering and collaborating with AI tools and real enterprise software. So that's whether it's partnering with a semantic language model like Anthropic plus ServiceNow or using OpenAI plus Salesforce. When you put these things together, they work really, really well. And the biggest thing I'll say is this, is really companies right now are split into only three groups. You got those that are building AI, those that are using AI, and those that are going out of business. The companies that are building AI already inherently are using it, and it's agentic workflows in as many parts of their business as possible. The companies that are using AI understand the importance of getting from here to AI, being embedded in their business. And the companies that are doing neither right now are quietly and humbly writing their death certificates. It's only a matter of time. Agentic AI is here today. It works today. The only excuse is not caring about the future well-being of your company.

‍

Patrick Moorhead:

Gosh, I'm pretty sure our producer forgot to cut you off.

‍

Daniel Newman:

How long am I supposed to have? I have no idea, but our producers are sleeping or hold on, you know, like, OK, I'd like to ask that the last thirty five seconds be strict from the record because Pat took a nap because I was going on so long and droning.

‍

Patrick Moorhead:

I'm wrong. Tell me why I'm wrong. Listen, you define what an agent is in your context. I'm going to define an agent in my context, which is horizontal agents that are not tied to a SaaS, because that's what I thought we were going to talk about. But that is what I'm going to talk about. So first and foremost, let's look at reliability. Do they do what they need to do? Agents on the whole are failing unpredictably. Error modes are not acceptable for many workflows, and my gosh, just for regulated workflows, it's just an absolute non-starter. I'm going to make up a number. Let's just say there's 100,000 workflows inside of the traditional enterprise. I'm not saying that there aren't 50 workflows. I'm just saying that the other 99,000 550 are not automated. Security is an issue as well. I mean, look at ClawdBot [aka Moltbot aka OpenClaw], look what happened. Unbridled over-permission agents are an absolute freaking nightmare. And actually, potentially bring in a new class of insider risk, right? It's a big enough risk that, you know, we kind of worry about most security issues being human driven. Now you have something that is just dumb and we'll do something like delete your hard drive or wire Daniel Newman twice as much money as the SOW or the contract. And the other thing is the integration tax, right? I mean, enterprises are messy. You know, most businesses are not, you know, born in the cloud, or I guess you'd say born in the AI at this point. They're messy. Building a stable tool connectors with data permissions is slow. I mean, heck, I've got more time using the Google agent framework number one, and number two, OpenAI, that thing gets Gmail wrong. It gets stuff inside of Workspace wrong. Half the time, it can't connect to its own calendar. I mean, that is just an absolute non-starter there. And then you've got the governance friction. You've got auditing, you've got compliance, you've got legal accountability that not only slows rollouts, but introduces enterprise risk. And the top three conversations that I'm having out there with real enterprises, cultural adoptions, right? They trust dashboards, not black box actions. And kind of slows the actual use of it. And I think finally kind of their skepticism on the benchmarks, the benchmarks are saying one thing like apex, right? They, you know, I mean, they're showing that these agents fail complex multi-step reasoning tasks outside of demo. So that's why for large enterprises, these horizontal agents, at least right now, are a non-starter.

‍

Daniel Newman:

That was a long, I don't know where the producers were. They should have cut you off about halfway through. Listen, there were no terms that said agents had to be BYOD, BYOA, build your own. But having said that, you make a great point because the whole software is eating AI, sorry, AI is eating software argument is really based on the idea that you'll just build your own every app. and the agents will replace the entire rails. I've said for a long time, though, that I think agents will sit on top of existing infra and software. But I do think over time, some agents will replace some software. Some agents will sit on top of existing software. Some agents will call into software. So I think it really does end up being a mix. I think what you're saying is true, that right now, doing your own, building your own from scratch, if you have a highly sensitive or regulated business, if you have significant compliance, if you have real security risks, I don't think anybody's launched anything yet that shows a lot of promise when all those things are incorporated that isn't going to be coupled. I do think if you're launching agents on top of a lot of these tools, they are market ready today.

‍

Patrick Moorhead:

Yeah, there's timing too, right? If you're a manufacturer and that is your core competency, you're going to put your own IP and your biggest investments in improving your capability of your ERP and your SCM. On the contrary, sales and marketing might not be as strategic and therefore you're going to pay Salesforce to, to do the AI magic because you just don't have the resources to go in and, and re redo that. I think the biggest threat for the software companies is going to be the lack of growth, right? Because anything that's not showing growth in there, um, like if you have, the fictional company I said, they're investing into ERP and SCM with their own stuff. They're going to freeze SAP or another company that it sits on top of, and they're going to build the value out on top of that. Why am I going to pay for Juul? Right. If that's the case, maybe it's agents talking to jewel. Right. Possibly. But yeah, I think growth is going to be the biggest challenge. Also, there's not, there's not one big pit of money just sitting there out there for it to use. And if I'm spending a big money on my own internal agents, I'm going to take it. I'm going to take it away from another area, but it's probably going to be enterprise SAS.

‍

Daniel Newman:

Absolutely. I think they're going to be calling APIs into those databases, like Satya said, in many cases, but I don't know that they're going to be that final semantic layer. I think the AI labs have done a really good job of showing what's possible there. Yeah. Um, all right, we've got about 38 earnings to cover, but, uh, in all serious, we're gonna do a bit of a rapid fire. Cause we got about 15 minutes and I think we literally have. I think eight earnings. So we're going to need to go pretty fast. Let's do a bit of a speed round here. Big couple of weeks, bulls and bears. All right. Market crashed. Market's back. Bitcoin zero. Bitcoin a billion. I don't know what's going on, but the big tech earnings, Pat, I mean, I even, I swear I was in California. I was getting ready. I turned on the TV and there's Pat Moorhead in a nice suit and tie from Abu Dhabi doing CNBC. You never miss a chance to be a superstar, my friend. I learned it from you. Dude, I am absolute media. There's a word I'm not going to say it. I've said every other cuss word. I won't say what I am here. But I really enjoy talking tech and talking stonks all the time. Let's just say we got the big three clouds. We'll talk a little bit of big three clouds. We've got Meta, a little bit of NVIDIA in here, Palantir, AMD, Coherent, I'm gonna start with Amazon. We really could say the same thing for the big three clouds as a whole. Amazing cloud results, growth looked tremendous, huge CapEx commits, scared the crap out of investors. Some of them are now raising debt rounds, selling a hundred year debt. Some of them are, you know, you know, issuing stocks, you know, basically, people are still in this AI is going to eat the world. But at the same time, companies can't spend too much on AI because it's going to hurt their cash flow. I was actually really impressed because AWS has really struggled to get back over like 20 percent. They got to 24. And I'm seeing reports now that they might get as high as 38% this year. So this was all about the early phase of AI missing the capacity build out, not hitting, but they really, I mean, it was a bit of a mixed result, but they hit the big numbers where they needed to hit them. The cloud number was a big one. They've got a great ads business that's generating a lot of profit for the company. Commerce is always a bit of the lag, but I love the fact that Amazon is an AI user and their company is going to benefit so much, not only from AI in the cloud, and AI in the commerce, but also AI in the physical real world at some point, they're gonna take a lot of costs out. So I'm optimistic on Amazon. We had really said that we liked it coming into the year, still like it coming into the year, but that 200 billion is a big number, but they're still playing catch up.

‍

Patrick Moorhead:

Yeah, I mean, if you look at their Anthropic build out, I mean, it is nothing short of just a miracle in the time that they did it. It's called Rainier. The size, the scale, a million, a million TPUs, sorry, a million traniums. I mean, absolutely gigantic. And Amazon is a conservative company. And they typically don't spin. They're very matter of fact, they're very right brain. And sometimes the market loves it. And sometimes the market, sometimes the market hates it. And I'm convinced that they're going to be a major player in what they do on the IS side. And even on, I mean, Bedrock is screaming. Bedrock takes the complexity away from creating a lot of these agentic operations, plus it masks the complexity underlying infrastructure. And I think for some people, that's exactly what they need. If you want to go deeper in the model, it's SageMaker. And I love what Dave Brown is doing and bringing these two worlds together and simplifying Bedrock and SageMaker. So, yeah, I mean, I am with you on this. I'm with you on this, Daniel.

‍

Daniel Newman:

So let's just bounce to Google, because, Pat, I don't know if you got much different to say than what I said. I mean, amazing, amazing cloud business. Amazing growth.

‍

Patrick Moorhead:

Yeah, it is amazing growth. AWS in whole numbers was still bigger on its growth and size matters. But this is Google's opportunity here on the cloud side. It's got this amazing advertising business that is feeding cash. Remember the meme that said AI was going to take down advertising and then Meta blew it away, continued to blow it away, and then Google came in and blew it away. We're always talking about ways to use AI that make a difference to the bottom line. And this is what Google is doing on the consumer side. And then what they've been able to do with Gemini and Gemini's enterprise and their agentic framework, and then which is a great parlay from the Vertex where they were the data, really doing some amazing stuff with data and analytics. It just makes sense.

‍

Daniel Newman:

Yeah, I mean, great results to your point, you know, whole numbers versus growth rates. But I think Google got it was like almost 48 percent, which is impressive. Turning a corner of growth definitely sits in the driver's seat in this AI cloud era. Again, market didn't like their big 180 billion cap, CapEx number. Everybody thinks they're going to make money without spending money. But what I like about Google is they have a core business that puts off a ton of cash. And so even Amazon, like where their commerce business does not put off the kind of cash like you know, the core Google business and search is not dead, folks. So let's tap on Microsoft, another one with- And Microsoft really kicked off the decline.

‍

Patrick Moorhead:

Yeah. Because they missed by like $500 million or so. No, no. It was a tiny miss on Azure. Yeah.

‍

Daniel Newman:

With what, a 500 and something billion dollar backlog. And what I actually said really freaked everybody out was the amount of backlog that was tied to OpenAI. And despite the fact that they have a backlog that's about the same size as Google and Amazon, if you take out the open AI business, it's just open AI has been such an oddly toxic relationship for so many companies right now. Oracle has been in an absolute downward spiral ever since announcing that backlog. And while it sounded really great, it just goes to show that there's some part of the market that just does not believe that open AI is gonna be able to meet its obligations long-term. I think this next funding round, by the way, Pat, is going to be really important. If OpenAI can raise $100 billion at like $850 billion and then take an IPO this year, it may take all the wind out of the bubble sale. Because even in the last few weeks, since the kind of AI will eat everything narrative has started, I'm seeing less bubble. People, I think, are starting to see the utility of it all. Now they're just questioning how much does it actually cost to get us to running. But Microsoft, it did well. I mean, all the attention was on that one part of the business. There was other parts that did pretty well, but I don't think people care. I just think right now it's all about that AI number, the Azure number. And so that's really right now what every quarter is being measured on.

‍

Patrick Moorhead:

I mean, $344 billion was not open AI. Right. I mean, isn't that completely. Massive. Right. Massive.

‍

Daniel Newman:

So quickly, Pat, you know, just touch on this one. NVIDIA, you know, and it's sort of rebounded pretty hard, but you know.

‍

Patrick Moorhead:

Yeah, it's definitely rebounding. I think it's up 8% in two days.

‍

Daniel Newman:

Yeah. I mean, do you think. I mean, because it went way out of its trading range. I mean, don't you think all this CapEx is really good news for Nvidia?

‍

Patrick Moorhead:

I mean, yes, I mean, you have to assume they're going to get 90% share of, of, of the capex of the GPU, which the GPU is the most expensive part, even compared to storage and power and facilities. So yeah, I mean, and then, but it's all just the, just the risk of the trade, what makes no sense. is people didn't sell off Microsoft because they didn't believe in the AI trade. They sold off Microsoft because it was tied to open AI. And then you would say, well, who does it go to? Does it go to a competitor? Does it go to Google? Anthropic through Google? Anthropic through AWS? You didn't see a corresponding spike. And I got on CNBC and talked about how the market's talking about both sides out of both sides of its mouth. It's like, I don't believe in AI, but I'm going to sell all these enterprise SaaS stocks because AI is going to be so powerful. Make it make sense. It doesn't.

‍

Daniel Newman:

So a speed round, four topics, four minutes. I'm going to start with Meta. Another great result. They actually went up against some of the selloff this week. I think people love that Meta is a company that's AI is actually being used to make money. They are going to spend a fortune on CapEx, but again, their CapEx is, they're showing direct impact to bottom line. He's slowing on the spending on the reality labs. It's all in on the AI supercomputing project. Pat, I mean, Meta is an absolute force. And one thing I'll just say about Meta is I think people under, it's one of those underestimate what you can do or overestimate what you can do in a year and underestimate what you can do in five. I think people overestimated the impact of the spending and the hiring they did last year, that it should have happened faster. But I think those impacts are going to come this year.

‍

Patrick Moorhead:

Yeah, I mean, it's metaverse stuff is still burning cash, but nobody cares. As long as AI and ads are growing and another company like Google making AI work and making a ton of money on advertising.

‍

Daniel Newman:

They are printing cash. Yep. All right. You talked Palantir. What's the quick run down there?

‍

Patrick Moorhead:

Real AI, real enterprise AI. We talk about it. One of the biggest questions, could they pivot to enterprise from like the US military and the Pentagon? Short answer is yes. Massive growth in the enterprise and a massive a guide and, you know, stock was up big time after hours and then it was muted the day after. And then it kind of got lumped in with the software trade, which was so dumb because what they're doing is AI. Right. So, yeah, I that was like indiscriminate selling.

‍

Daniel Newman:

You know, I guess I don't have time to dig into it, but my take was it was a crypto meltdown. People invested in crypto or invested in growth. When the crypto started getting liquidated, they had to sell better assets, better assets or Palantir, better assets or Microsoft. And I think it was a lot of covering to liquidation. That's my take. I'll never get full proof, but it seems like once crypto stabilizes, the stocks tend to come back pretty quickly. AMD, real quick. They had a huge decline. Again, numbers were okay. I mean, it was not really a numbers thing. I think this was another company that was sort of victim of this moment. The only thing about AMD is they're such a conservative company. You know this better than anybody, Pat. Jensen's talking about his $500 billion order book. for AI and Lisa's still not really coming out and just saying how big the Helios party is going to be just yet. She's talked a little bit about how big, you know, we see their data center business in the tens of billions in the future years, 30, 40, 50. And, you know, we think just because of demand and capacity, they have access to memory, they have access to compute, they're going to sell some serious Helios, their numbers are going to rise with this trend. But I just think the market's waiting to hear it. I think that's for sure.

‍

Patrick Moorhead:

Sell off was exactly what you said. They expected Helios type growth numbers and they don't have Helios. So you're getting a scale out solution versus NVIDIA scale up solution. We are where we are. A lot of execution to go. Elise was very confident. There were some rumors of it not being pushed out, but we squashed those, but it didn't make a difference.

‍

Daniel Newman:

All right, let's talk about laser beams. Let's just really quickly, Pat. Coherent’s, another big beneficiary of all this buildup.

‍

Patrick Moorhead:

Yeah, I mean, basically a quadruple beat. And, you know, I said it last quarter, in the quarter B before that, when CapEx goes up in AI, so does Coherent, because you've got to move that data quicker and quicker, and Coherent is the go-to solution for that.

‍

Daniel Newman:

Yeah, very impressive. I believe a deal in hand from Co-packaged Optics. Looks like NVIDIA. Looks like they booked something there. But this company is doing interesting stuff. Cooling inside of the racks. They're doing stuff for connecting data center to data center, rack to rack, inside the rack. There's a huge business to be done there. One of the big players in Co-packaged Optics. And frankly, Pat, they also do all kinds of interesting stuff like for nuclear fusion. I mean, there's a company that's got its hands in all these megatrends. So many different things. And, you know, deserves attention. And it's well run. We know Jim Anderson knows how to run a business, knows how to meet and beat consistently. And he's done that for a long time. Pat. Friggin' A, man. We did it. We did not even an hour, by the way. So, you know, I didn't know if we could pull this off. Kind of got two weeks into one, went really fast. But man, we're good. We're still good. We're still good. We still- I don't know, man.

‍

Patrick Moorhead:

I'm so jet-lagged. So I hope to improve on my performance coming up here.

‍

Daniel Newman:

All right. Well, I think you did okay. I give you a B plus. B plus. Thanks, man. Everybody there, you know, you lost the debate, but that's kind of a weekly in and out happens every week. Just kidding. Just kidding. Listen, everybody, we appreciate you being part of the Six Five community. Subscribe, you know, join us. Catch our episodes every week. Join our 100,000 plus fans and subscribers. And if you don't want to do that, follow us on X. Check us out there. Do both. Follow us on X. And by the way, send all your complaints to at Patrick Moorhead and all your praise to at Daniel Newman UV. I appreciate it so much. We appreciate it so much. We've got to go for this episode. We'll see you all back here again on The Six Five very soon. Bye. Bye.

‍