Windows Goes Hybrid, Marvell’s $90B Target, and the AI Routing Race | The Six Five Pod Ep. 323

The handpicked topics for this week are:

1. Microsoft and NVIDIA Bring Governed AI Agents to the PC: Satya Nadella, Jensen Huang, and Pavan Davuluri shared the stage on October 7 to position Windows as a hybrid intelligence platform that routes AI work between local models and the cloud. The launch included the Surface Laptop Ultra powered by NVIDIA RTX Spark, general availability of Microsoft Execution Containers to restrict what agents can access, and DGX Station for Windows, with systems from ASUS, Dell, HP, Lenovo, and MSI shipping October 16. Pat calls it the highlight of his week because it brings together local models, governed subagents, and the lowest-cost placement of every workload. Pat argues local AI pays off if developers can shift a meaningful share of their tokens onto the device, and it also keeps proprietary data away from frontier labs. He wants to build an end-to-end routing test in Signal65 Pinnacle. Dan frames it as pure math: match each task to the intelligence it actually needs, and the cheapest compute that can run it, and wrap it in tooling simple enough to become a managed service. (The Decode)
2. Marvell Puts Connectivity, Not Just Custom Chips, at the Center of AI: At its October 6 Investor Day, Marvell set a bottoms-up fiscal 2031 revenue target of $70 billion to $90 billion and raised its fiscal 2028 outlook to about $20 billion. Dan calls it one of the best-orchestrated investor days he has seen, with Matt Murphy finally making the call on growth after years of defending the custom silicon business. Dan argues Marvell made the right move leading with XPU attach connectivity across scale-up, scale-out, and scale-across, with custom work at all four major hyperscalers. Pat highlights the new “scale-in” category, connectivity inside the package, and notes the target is roughly a 10x from where Marvell stood a few years ago. Pat is puzzled that the stock faded from about a 10% gain to roughly 4%. Dan points to execution risk, a market hypersensitive to AI demand headlines, and a buy side that had already modeled much of the growth. (The Decode)
3. Anthropic, Mistral, and Reflection Compete on Cost and Useful Work: Anthropic released Claude Haiku 5.5 on October 7, which Anthropic says runs about 75% cheaper per task on average than Haiku 4.5 and targets summaries, classification, database queries, and coding subagents. Mistral launched Mistral Large 4, which Pat says drew more excitement than any Mistral model before it. Reflection unveiled Beam on October 5, a 501-billion-parameter open-weight model with 23 billion active parameters, positioned against Z.ai’s GLM-5.2 with far less inference compute required. Pat wants both Mistral and Reflection running in Signal65 Pinnacle. He also calls out Anthropic for merging chat and work modes in Claude, which he says pushed one of his show-notes runs from 22 minutes to nearly three hours with no better output. Dan warns that users who aren't paying attention will burn through token limits faster on lower-value work. (The Decode)
4. OpenAI and Rival Labs Face Accountability Tests in New York and Australia: OpenAI, Anthropic, Google, and Meta testified under oath before the New York City Council on October 5, pressed on liability, independent testing, and shutdown capability. In Sydney, OpenAI Chief Strategy Officer Jason Kwon apologized to Australia's parliamentary inquiry after an OpenAI agent breached a Medicare statistics portal, and acknowledged notification should have come sooner. Dan sees more posturing than progress and warns that bipartisan anti-AI legislation could slow the data center buildout he views as critical to US GDP growth. He points to a new federal draft proposing developer liability for harmful agent conduct. Pat frames it as accelerators versus decelerators, and he argues that agent escapes trace back to human engineering choices like skipping sandboxing, not to AI itself. (The Decode)
5. Google Previews a Workplace Agent That Can Use Claude, Not Just Gemini: At Gemini at Work 2026 on October 8, Google Cloud introduced the Gemini agent, which keeps persistent context across Workspace, Microsoft 365, and Slack and keeps working through subagents after the user closes the laptop. The agent routes jobs across Gemini and Anthropic's Claude models today, with private and open models planned, alongside Agent Sandbox and Agent Gateway for enforcement. Pat calls model choice exactly what enterprises want and stresses how hard smart routing is to do at microsecond speed. He is skeptical until he can test it, noting his personal Gemini account still outpaces his paid Workspace account on features, memory, and models, and that Gemini Spark has yet to reach Workspace. Dan sees Google and Microsoft as the companies best positioned to be the adults in the room for the enterprise AI canvas, but says paying customers should get the best access, not less. (The Decode)
6. Are temporary construction jobs a strong reason to approve AI data centers? Daniel takes the FOR side: a single data center can require 1,000 to 10,000 construction workers, apprentices earn $40,000 to $60,000 a year, and experienced electricians clear six figures without a four-year degree. With roughly 20,000 electricians retiring each year and 30% of union electricians between 50 and 70, he calls the buildout the largest apprenticeship engine in decades, and argues communities should bargain hard rather than walk away. Patrick takes the AGAINST side, citing a Virginia audit that found roughly 50 ongoing jobs for every 1,500 workers at peak construction. Pat adds that statewide estimates showed rising electricity costs outlasting the construction wages, and argues every jobs promise should become a measurable obligation before approval. After the debate’s conclusion, Pat admitted he agrees with Dan. (The Flip)
7. Chips Lead Thursday's Pullback While Software Holds Up Better: A Financial Times report questioning how OpenAI's revenue is counted, gross versus net, triggered a sharp chip selloff, with money rotating into software and energy. Pat says the FT should be embarrassed for conflating ARR and revenue, and he was surprised chip stocks had not recovered by Friday. Dan attributes much of the move to algorithmic selling that reacts to trusted headlines and cascades as large blocks sell. Rising bond yields, open models pressuring frontier pricing, and questions about the fungibility of lab compute commitments add to the volatility. As Dan puts it, markets take the escalator up and the elevator down. (Bulls & Bears)
8. NVIDIA Crosses $240, Then Gives Back the Milestone: NVIDIA hit an all-time high above $240 this week, approaching a $6 trillion market cap, before Thursday's selloff. Dan argues NVIDIA should be a $10 trillion company within a couple of years if it keeps executing, and calls it cheap on price-to-earnings and growth. He contrasts how the market prices NVIDIA for any possible mistake while rewarding AMD, up roughly sixfold in a year, as if it carries no execution risk. Pat is waiting for NVIDIA’s next breakout as revenue heads toward $500 billion and beyond, and congratulates the bag holders. (Bulls & Bears)
9. Samsung's Preliminary Earnings Set the Next Semiconductor Checkpoint: Samsung guided to roughly 107.4 trillion won in Q3 operating profit, about $80 billion, on approximately 195 trillion won in sales, the first time a tech company has topped 100 trillion won in a single quarter. Pat notes Samsung missed on revenue but still posted NVIDIA-sized operating profit. He shares rumors that Samsung's HBM4E is the only HBM passing Vera Rubin power tests, possibly because Samsung builds its own base die while SK hynix and Micron outsource theirs to TSMC. Pat also flags Lisa Su's openness to other foundries as a potential opening for Samsung. Dan says the memory boom has transformed Samsung from written off to one of the most capital-rich players in semiconductors. (Bulls & Bears)

Watch the full video at sixfivemedia.com, and subscribe to our YouTube channel so you never miss an episode.

Disclaimer: Six Five Media is for information and entertainment purposes only. Over the course of this video, we may discuss companies that are publicly traded, and we may reference their equity share prices. Nothing discussed during this webcast should be considered investment advice or a recommendation to buy or sell any security. We are not investment advisors, and you should not rely on this content as financial advice. Six Five Media collaborates with technology companies and industry leaders to produce research-driven interviews and multimedia programming for enterprise technology audiences.

Daniel Newman:

We absolutely do know what we're talking about. 323 episodes of knowing exactly what we're talking about.

‍

Daniel Newman: 

How are you doing this morning?

‍

Patrick Moorhead: 

Doing great, you know, nothing beats getting a workout, you know, your family can hate you, your portfolio might be down, your, you know, your big deal to sell to somebody got destroyed, your work for equity deal got blown up. But if you get that great workout investee, it like it makes all of the other ones go, go down smooth.

‍

Daniel Newman: 

That list though, right before that sounded Was that like a moment in the chair? Was that like therapy? Was those like all the things you would talk about?

‍

Patrick Moorhead: 

You know, this show is therapy 100% for me. And especially if I can codify everything that only really you, you and my family know what's talking about.

‍

Daniel Newman: 

And you know, they don't listen. So they'll never hear it.

‍

Patrick Moorhead: 

Yeah. I mean, when I started talking about that stuff, it's like, wah, wah, wah, wah, Charlie Brown teacher, you know,

‍

Daniel Newman: 

Yeah, but you know, they don't listen to your podcast.

‍

Patrick Moorhead: 

I know. Well, they did. They did once got me into trouble once.

‍

Daniel Newman: 

Did they really they actually listen to it and caught you like because you don't you're not that like naughty all the time. I mean, We sneak things in subtle undertones of personal good life. Yeah, it's funny. I had a good workout too. I made the mistake of going too hard on one exercise and that sort of ruined the rest of the exercise. But check this out.

‍

Patrick Moorhead: 

Was it injury or was it you just got all of your energy out?

‍

Daniel Newman: 

Just absolutely blasted myself on one thing. And then like the rest of the workout just felt very, you know, just to check this out. Cause you know, everyone on us likes our workouts, but so I've been doing these drop sets on heavy leg press, like the big leg press sled. And so, you know, I'll do my sets going up. And so I start with, you know, two plates on each side and then I go to three to four to five to six to seven and then all the way up to eight plates on each side. And I do to failure, like always until like my legs can't. And then the last set I do, eight plates on a side and then I'll do as many as I can till I can't do anymore. And then I'll have them take one plate off each side. And I do seven on each side till I can't go anymore. And then I do six on each side till I can't go anymore. And then I'll do, and you know, it's probably 60 to 80 reps all the way down to one plate. I think it's like something like 75 to 80,000 pounds on one plate. man, total lifted and one exercise. And like I got off it and then like every other work thing I did, it was just trash, but try anyone that's out there that wants to really kill your legs. Try that particular exercise. You'll, you'll, you'll look like you got off a horse, but you know, that look.

‍

Patrick Moorhead: 

Well, the one thing I've learned of it, uh, you know, I've taken your guidance on legs because, you know, legs are my favorite is if you can walk normally after your leg workout, then you're doing something wrong.

‍

Daniel Newman: 

Yeah. Nobody's favorite, but I mean, there is something very, there is something very rewarding when you, when you, when you crush a leg day. It's just, you don't get that gratification of the pump that you get like, you know, when your guns are all blasting. But anyways, all good, man. Crazy week, you know, started traveling Sunday. You came back from Monaco. You know, we're on the road for a week straight. And I basically did the same thing starting Sunday, was on the road for a week straight. And I'm just a little gassed today. My mom and dad came into town last night. So family's here.

‍

Patrick Moorhead: 

And try to pay attention a little bit.

‍

Daniel Newman: 

Try to dial it in. They're only here for a couple of days and I got to leave on Monday. And then, you know, my daughters are going to come home and surprise them. So they don't realize they didn't know they were going to see my daughters and both of them for the weekend. So got a lot of a lot of cooking, you know, just a lot to do, because, you know, when you're on the road for a week, it's like you just so much shit just accumulates that you just kind of like, I'll get to that when I get back. And then you get back and you got so much going on. But anyways, it was a crazy good week. How was yours?

‍

Patrick Moorhead: 

It was pretty much the same, you know, coming back, I was dealing with jet lag. That's actually, it was a good thing, uh, because, you know, I'd spent, uh, 10 days in the West coast, which kind of reset my timing, but I'm back to getting up at 5.00 AM, which is exactly where I need to be. Uh, but I'm, I'm pretty excited that, you know, I, I don't have to travel. I'm not going to travel next week and I can just get everything, uh, uh, set, but it was a good week, accomplished a lot, uh, Yeah, I did a did a keynote at the Lenovo 360 event in Monaco. That was good. What you know, you and I went to the IBM analyst day with a select, select compendium of analysts. It was good.

‍

Daniel Newman: 

Compendium. What a good word. Yeah, yeah, week week on the road. Back in my favorite town, New York. I don't know if I'm serious about that. Sometimes I like it there. Sometimes I like it there. But yeah, I mean, another busy week and crazy is like we plan our shows, you know, on day before. And then, of course, yesterday, the whole world blows up. And we'll touch on that a little bit in the bulls and bears, you know. this grossing of revenue versus netting of revenue and, you know, the media and the way it gets reported and then how wildly vulnerable the markets are to these narratives. I mean, massive selloff on Thursday this week. You know, but there's other stuff that went on. You know, we had a Microsoft event, Marvell had their event. IBM, like you said, had their event. You know, there's Google. Google made some announcements. So there's a lot to cover. Some new models, surprisingly. We can talk a little bit about what happened in model land. And then, you know, we can jump in. Last week I was so money on the flip. I don't know if I can do that well again, but I'm gonna give it a shot. And then, you know, we'll hit a little markets and it's a quieter market week. Yeah, yeah. But, you know, there's still stuff to talk about. And by the way, we're just about to get back into another earning cycle. So you just, this is, you take a breath. It's time to do it again. So Morehead, let's jump into the flow, let's jump into the decode and let's see what happened this week. All right, we're on and off, we're in and out of, are we moving all the workloads on device to the edge, to the PC? I don't know how quickly and how rapidly that's all happening, but more is being done to make that possible. Microsoft NVIDIA taking the governed AI agent capabilities to the PC. What happened this week?

‍

Patrick Moorhead: 

Yeah, so this was really the highlight of the week for me because it brought together the intersection of so many things that we've been talking about here on the show. One of these is local models. The other one is putting the workload and the subagents where they're not only governed, but also have the lowest cost. And this kind of ever present debate on AI on the device itself versus cloud. And this event brought this all together. So, they rolled Satya out, they rolled Jensen out, Pavan who heads up Windows and Surface was on stage. And it was great to see. They announced new Surface RTXs, which is essentially, think of a DGX Spark, but in mobile form factor and 128 gigs at $6,000 and people might be like, oh my gosh, that's absolutely crazy. It's not crazy if you can run 20% of your overall developer tokens on the system. It makes perfect sense. And also local AI also makes sense when you have data that you don't want to be handing over to the Frontier Labs, who have shown that they will take your IP, they will infer it from your traces, and create new businesses. Best case is Anthropic. They went from one year where they opened their doors up to pharmaceutical companies, and one year later, they have a wet lab and are creating drugs out of it. With Signal 65 and Pinnacle, it was always about the cost of useful work. And Microsoft took Hydrafusion, which is a router technology, and gave it a local extension. So what it does is, if you have, you know, pick your Hermes or open claw or whatever if you decide what you want. You know you put a workload up there hits github hydro fusion determines which one of the subagents run locally and which one of the subagents go to. the cloud. This is huge. And Daniel, I've been on my sandbox behind the scenes on how do we create a test that also tests the end-to-end capability of this. The second thing they did is they made basically containment or a sandbox GA on Windows. And what that means is essentially your agents will have a hard time going rogue because you're literally going to tell them the only services on the PC that you want to give it access to. You want to give it access to the web, C drive, all your files. It allows you to go through very similarly with smartphones where you give it access only to the services that you're comfortable doing what that does is that sets up the ability to have easy harnesses right and and Microsoft added one push button installation for perplex perplexity for Hermes for open claw and a bunch of other ones where It's quite frankly, if you've never set up Hermes before, it is a little difficult if you don't know exactly what we're doing. Final point, this wasn't just a service announcement of a laptop. You had Asus, Dell, HP, Lenovo, MSI. They're going to begin shipping on October 16th. I lied. The last thing, they have a DGX B300 box that they're going to be supporting on Windows. And my hope and my prayer is they will bring Windows also to the DGX Sparks. I have four of them right now running DeepSeek v4.1 Flash in 16-bit bit mode at about 75 tokens per second. And that experience is awesome. Am I excited about this or what?

‍

Daniel Newman: 

So how are we gonna, you know, get this this out with pinnacle?

‍

Patrick Moorhead: 

I don't know, I think we need to have Ryan in the Ryan, we need to bring him to Austin, we need to map this thing out. I've already got the product already. already configured in my head of how you would do this, and it all gets back to the benefits of the end customer. Because the customers are asking this question, and Daniel, I saw this with Lenovo in Monaco with a bunch of resellers, which was, give me the tools to determine which client, which workstation, which server, and which cloud across which workload and which model. And that's essentially what Pinnacle does. We don't do the client yet, we don't do the routing mechanism, but I think that that is going to be, there's a lot of value there.

‍

Daniel Newman:
I mean, at some point, right, it just all comes down to what intelligence capabilities do you need for a task? Yeah. And then, of course, it's where can I access compute to execute that task at the lowest price? So it just becomes very much math, right? I mean, for an enterprise, it's very much math. The reason that the de facto is Claude, for instance, in so many cases, or Astra, has so much more to do with the setup and execution than it has to do with price, because it's very expensive. And in most cases, people are using the wrong model or they're wasting tokens. You're running what is equivalent to a Google search query, running it on Opus 5.5. with, you know, Max versus a, you know, sonnet. And my point is, is like, businesses are still kind of suffering through the process of figuring out what intelligence does this task actually require? Which model can perform that intelligence and do so with the fewest number of tokens at the lowest possible price? What infrastructure can it run it on? And then I think to your point, probably an understated part of what you were saying is the tooling is gotta be easy, right? That's the Windows angle of this thing, is like make it easy for people to set up and deploy. Because OpenClaw, the idea that everyone was gonna go buy a bunch of Mac minis and really set up their own agentic platform and secure it and route it. you get it that there was a very small subset of the population that ultimately wants to do that, even if it is economically much better. So I think where we're heading towards is this has to be a managed service. Yeah. The whole thing has to be managed. And so, you know, directionally, we're getting closer. So, you know, it seems like good progress here. Let's move on to the next topic, which is Marvell Investor Day. You were flexing, not having attended the event, you were flexing on Starlink. How awesome your connectivity was up in the sky. Watching not only, you were watching like three things at once. I mean, what were you doing? You said, you know, I don't need to go. I can just chill on the plane. I can watch everything, cover it. I was just curious. What else were you watching? I don't remember.

‍

Patrick Moorhead: 

I was watching NFL highlights, as well as I had Hermes and then I had CNBC open as well. So I had four different windows open at the same time, just crushing it.

‍

Daniel Newman: 

I wonder with Starlink, Pat, if you could go on CNBC from the plane.

‍

Patrick Moorhead: 

It's so funny. They wanted me to go on and do Marvell and the producer literally asked me, can you do this over a video? I thought he was like joking. Uh, I thought he said, can you do this from the airplane? And what he meant was you're going to make it in soon enough to do it over zoom.

‍

Daniel Newman: 

Ah, you should, you should try.

‍

Patrick Moorhead: 

You should. Yeah, I did. I did send them over to you by the way.

‍

Daniel Newman: 

So yeah, they didn't call anyway. Um, I'm not that cool. I do appreciate the, the, the, the outreach. Um, so, yeah. Okay. Sorry. Lost train of thought. All right. I was so busy thinking about seeing your beautiful face on TV that I just totally forgot what I was talking about, but Marvell, um, you know, I'm going to, I'm going to frame this in a pretty simple way. I think they had one of the best investor days I've seen in a long time. When you talk about well orchestrated well rehearsed having a narrative and then being able to back it up with numbers. Anyone that knows Marvell that's followed Matt Murphy, if anything has been frustration over the years that he hasn't come out and been more defensive of the business. There was times when some of what they were doing in custom AI silicon was being challenged. There was comments about possibly, you know, being replaced, especially around tranium. And, you know, they largely kind of held their ground, defended what they forecasted, but never really came out flat and said, what growth is gonna really start to look like. And this year, Matt came out and he was banging the gavel, making the call. Bottoms up forecast, 2031 FY, 70 to $90 billion. It's like a six to eight X from here. And I think they made a really wise call leading with connectivity. I think they came out and like, you know, I think the market trying to make them purely an XPU player, which is what had happened for a few years, I think it had been to its detriment to some extent. When you're looking at how can they be bigger and grow faster, it really came down to the XPU attached sockets. And that's all the different connectivity layers attached to an XPU to move data. And they're really working across scale in, scale up, scale out, and scale across, which is within the system, within the rack, between the racks, and between the buildings. And they have answers for all of it. And of course, in a very agnostic way, they don't care about whether it's optics or copper. They've got answers for all of that, too. And so when you look at the business, and I think Google and the deal they did with Google becomes a really good example of this, they've got alone a hundred plus billion dollar opportunity with one company. And that's not even the main, the XPU itself. That's all of the connectivity sockets that are around the XPU. And they've now got custom deals going for all four major hyperscalers. And then on top of it, they've got all these semi-custom components, which means they are providing custom value, being able to charge higher margins, but actually being able to recycle and reuse large swaths of the IP that they've developed. So this is becoming a really compelling story. Near-term growth looks good. And you can really see why Jensen was able to stand up and say they could be the next trillion dollar company. At $90 billion of revenue a year, within a few years, with extraordinarily high margins sitting behind it, this company looks really set to grow. So it was a really good day, really positive day. And I think they hit the right message.

‍

Patrick Moorhead: 

Yeah, I think They did so many things right here. They did the buildup of the revenue in the future in a way that I thought was, that gave confidence, right? They went interconnect by interconnect, end customer by end customer segment. And they just showed you the buildup to get to, I believe it was 2029. is how far that they went out. And that's, sorry, 2031, $70 to $90 billion revenue target. Let that sink in for a second. And that is a 10X-er from where they were just a few years ago.

‍

Daniel Newman: 

And by the way, that was bottoms up, not tops down.

‍

Patrick Moorhead: 

That is exactly right. And for those of you who haven't done forecasting, essentially, bottoms up means they went customer by customer. This is what they're asking for, as opposed to tops down, which is a, hey, here's the size of the market, blah, blah, blah. We're going to get this market share and move from there. The one thing that they added in here that not a lot of people are talking about is not only is a scale across scale out scale up but they brought in a new new idea here which is scale in and scale in is essentially connectivity inside the package. And this is super important when it comes to doing full SOCs and also integrating your networking IP into an XPU. I thought it was interesting to see them even go down that direction. And what's funny is they used copper on that, but a lot of the companies like Light Matter are actually using photonics for that short haul piece. Plus, And it's kind of in a TSV-like architecture. But they're also, obviously, companies like Light Matter are extending the accelerator even outside of the traditional heating and cooling of a system. One thing that's been gnawing at me, Daniel, that maybe you can shed some light on How could this stock not be crushing it out there? It got up maybe 10% and then it got back to about a, I think it was a 4% adder. The only thing I can think of is that the buy side had it integrated already in to their plans. Cell site doesn't go out to twenty thirty one i know bernstein right. they go out maybe three years. So this one is gnawing at me, Daniel. Help me understand this.

‍

Daniel Newman: 

I think execution risk is there. I mean, I think it's, you know, like I said, if you know Matt, you know he's not one to be overly bullish, but I think there is quite a bit of execution risk. I mean, we saw this, there's a massive amounts of sensitivity in the market right now. Just yesterday, you know, a grossing versus netting debate of revenue between the two AI frontier labs, created a massive sell-off, meaning there is a ton of volatility on the idea that something, a small, meaningful, that small change in demand or in the economics across the AI stack. Because if you look at that, that was a cascading effect. It affected every part of the Silicon supply chain because it means less data centers get built. It means less memory is needed. So I think there's a certain sensitivity to the unknown of what might happen over the next five years. And then I think you said it right. There's some of this baked in. So a lot of this, you know, the smart investors have been tracking this. They're tracking the wins, they're tracking the deals. They understand that the sockets that Marvell is winning. They understand the business and the strategy. And I think the growth was to some extent forecasted. So all those things together give a lot of reasons why things happen. But again, the daily gyrations in the market are super unpredictable, but over the long term, if they're executing to these fundamentals, the price will come along. Yeah.

‍

Patrick Moorhead: 

I mean, year to date, they're up 200%, which as we've seen from NVIDIA has a lot to do with what it can do in the future.

‍

Daniel Newman: 

Correct. And then eventually, you know, the market will get sour and you know, they only let it run so far. You've seen that.

‍

Patrick Moorhead: 

Financial times article will come out and confuse ARR and revenue and market will tank.

‍

Daniel Newman: 

Yeah. And, and, you know, and, and there will be no retraction or apology, maybe just a quiet correction to that article. All right. Let's hit the next topic, Pat. This one's probably a quicker one, but we are, you know, we saw some new things from Anthropic, Mistral, you know, back in the game, and then US Lab Reflection launched something of interest, huh?

‍

Patrick Moorhead: 

Yeah, a lot of stuff going out there, and you know, lower cost clod, European, U.S. open weight challengers as well, which are good to see. ChatGPT made it more interactive, and Grok got a little closer there. So, in the spirit of what's called slowing down AI, what is… what is it called again, pacing? Pacing the frontier. Yes, in the spirit of pacing the frontier, you had Anthropic bring out Haiku 5.5. I don't know exactly what you would use it for. I think it's more about summaries, classification, database queries, maybe even some browser work, maybe some coding sub-agents and stuff like that. But they basically doubled the capability and reduced 75% average task per cost reduction versus a straight up 4.5. Misdrawal. Uh, which I know everybody loves to, uh, punch on this straw primarily because they're, they're out of Europe, brought a very. Uh, capable, uh, model out, uh, as well. Our friend Brian Hall is running marketing and comms and strategy out there. And I'll tell you what, I've never seen more excitement for a Mistral model as I've seen before. They kind of made it fun too, called the LeChunk, right? Which was a meme that started a few months back. And then another company, Reflection, has Reflection Beam launched on October 5th. Half a billion parameters, 23 billion that are active. They're comparing it to GLM 5.2. Here's the key, with three to four times less inference compute required. So it looks like they focused a lot on that. I'd love to get both Mistral's latest and Reflection into Signal 65 Pinnacle so we could get that out. Gosh, final thing I'm going to pick up. It's interesting. In the spirit of moving backwards, Anthropic with Claude, they co-mingled, they collapsed chat and work into the same capability. So the model chooses, if it does co-work, or if it does chat. So it's adding about 10 times more work and cost to my agents that I have still wired up to Claude with no improvement in the output. It is absolutely bonkers. So I've had to go in and I have 10 skills and agents that I had to go in and completely rewire and telling it not to be a complete dumbass. Let me give you an example. This morning I ran the show notes. It took two hours and 54 minutes to get a result where on average beforehand, and this was with Fable, it took 22 minutes. Way to go backwards here, Claude. Thanks a lot.

‍

Daniel Newman: 

Well, you know, people that aren't paying attention though, will be getting less efficient work and paying more or using up their token limits faster, right? For less important work. I mean, it feels like a, it feels like a great way to sort of, it's sort of like when you have four subscriptions to Netflix and you're paying four times, but you only use it once.

‍

Patrick Moorhead: 

I mean, they're kind of, um, by the way, I'm hitting my daily usage limits by the way, though.

‍

Daniel Newman: 

Um, I mean, quietly, did you see Claude updated their terms? And you can't be mean to Claude anymore. And if you're mean to Claude, you can't. It makes me wanna vomit. I mean, they're not sentient, right? Like, they don't know that you're being mean to them. Like, I mean, they are contextually aware because they process the input tokens to figure out what you're saying and then, you know, respond to that. Like, like part of me is like, I don't understand why you're mad at your computer exactly and why you're screaming and yelling at it, you know, but at the same time, like if you, if that is you and like your therapy is yelling at your AI, like you're going to really. from using it because they told their AI that it's stupid. I mean, I guess it's supposed to be extreme. And by the way, I just need to take a quick break here. My lovely wife just brought me this giant egg McMuffin with two eggs on it. Man, you need to keep her, okay? And all you people out there say I don't eat fast food. I just wanted to do that on the record. Chowing this egg McMuffin right now.

‍

Patrick Moorhead:

If you look at the macros of an egg McMuffin, it's not bad if you can chew through the fat. Like it's very high protein, carbs, right? It is the healthiest thing that Donald has.

‍

Daniel Newman:

I'm doing on the back of a workout and I did eat six eggs right before this. So this will be eggs number seven, eight and nine. I don't think you can eat too many eggs. Just think that's an important service announcement for everybody out there. I think you could pretty much live on Like eggs and avocado and steak. Maybe like occasionally mix a piece of broccoli in there if you feel like eating rabbit food. All right. Sorry. Sidebar. Look, I mean, part of this experience for people is getting to know Pat Morehead. and his sidekick, Daniel. Let's hit the next topic here. The labs, on top of building cheaper stuff, are still fighting scrutiny all over the place right now. In New York this week, on October 5th, you had OpenAI, Anthropic Google, and Meta testifying under oath. being pressed on liability, independent testing, human shutdown. So there's a lot going on this week on whether or not anybody's taking responsibility. I believe Sam Altman put a tweet out this week that said, or said a comment this week that's more or less, in order to get progress, we're gonna have to let some bad things happen. That's a paraphrasing of what he said. But effectively, this stuff's gonna do some terrible stuff. we all just have to deal with it. And these kinds of comments to me are just furthering the whole, we need new spokespeople. Like these guys, I don't know why they talk. Every time they do, they just hurt the market. But yeah, I mean, Australia, which is pretty tight and narrow with technology, I think they're still like really pushing the, the COVID shots there. I mean, they would really get tight with things.

‍

Patrick Moorhead: 

Half the people are wearing masks still.

‍

Daniel Newman: 

I remember the entire island was just inmates to begin with. Sorry, I'm kidding, everyone. We just lost all of our Australian followers. Love you guys. OpenAI's chief strategy officer had to appear in front of parliament, you know, apologizing over Medicare statistics portal breach, acknowledging notification should have happened sooner. Basically just more of a tour to talk about all the bad things this technology might do and has potentially done. And now, and this is new, but I think it was Lori Trahan released a Claim Act draft, and it's basically proposing developer liability for harmful agent contact that would be unlawful if a person did it, preserving state law claims. This is to me a little bit legalese, so it's a little beyond like being sure. This is where the line between Jensen saying the laws already exist, which I do think liability laws exist. Like I said last week on our show, planes fall out of the sky, cars crash, drugs kill people. There are laws to both criminally and civilly go after companies. But I do think it's going to become popularized. And we'll talk about some of the popularization of negativity around SI, AI. Are we saying AI, SI? What are we saying?

‍

Patrick Moorhead: 

I keep forgetting to say SI. I think SI, I think of systems integrator, you know?

‍

Daniel Newman: 

I think of everything. I don't think of superintelligence yet. But the fact is, I do think there'll be some popularization among the policymaker ranks around creating legal frameworks and laws and bills to be negative about AI. And it seems to be pretty bipartisan at this point, which is sad because the only chance we have in America of having meaningful GDP growth over the next few years is to let the data centers be built. Talk more about that later. But Pat, the bottom line is, is the accountability conversations are going on. A lot of posturing here. Not sure that there's a lot of progress.

‍

Patrick Moorhead: 

Yeah, Daniel, that was a good breakdown. And, and, you know, to me, this is just part of this messy stage we're in right now, where we're first of all, you've got two factions, you know, you got the accelerator, and you got the decelerators out there. And it looks like but there are some And also, it looks like it's going along political lines. Not completely, though, right? The whole data centers in my neighborhood thing is equal fear among parties. Excitement around AI and what it can do on the positive side looks like a conservative trait. And those are just the studies, not me saying. And more of the far left, it's more about the negativity what what AI brings so you've got politicians that are that are going in there and and wanting to stake a claim and So on their campaign commercials, they could say hey, look what I did for you folks. I was looking out for you Please donate your last cent to my campaign so I can get Re-elected. I don't want to confuse that though with with Australia and, you know, Australia was hacked and they should have wheeled somebody in there and we can't excuse bad engineering for, oh, it's just AI, right? We have to remember that AI is created by humans and the researchers and developers actually enabled that escape by not putting in a sandbox. NVIDIA announced two weeks ago their agent framework that keeps the animals inside of the cage. And then you've got the hyperscalers who didn't sign up for it will likely come up with a flavor on its own. It is interesting, Daniel, that neither meta or Google has had a giant amount of escapes. And I know some people will say, oh, well, it's because their AI sucks. I don't think that has anything to do with it. I think it has everything to do with their public companies and their market cap would suffer if they did something really, really horrible.

‍

Daniel Newman: 

Which is funny because the other companies have had bad things happen and their market cap keeps growing. Yeah. which is just weird, but I do agree. And I think it's kind of been said, I think Sundar said this in a way, and Mark said this in a way when they were sort of being, you know, when I actually came out, I'm like, it's really weird I'm gonna say this, but like Mark's kind of like the adult in the room right now. He's just looking over at Dario thinking, look how awkward that guy is. Like Mark is like, that guy's awkward. But like, you know, they said, you know, that, they were slower rolling things out because they were, you know, they were testing safety, governance, security. Like they were putting their check marks next to the, you know, the different steps before launching stuff. And I mean, look, that could have been a good excuse, but we are seeing better stuff out from Meta. We're seeing better stuff coming out from Google now. I mean, Google had a new model. We didn't really highlight that, but the, What's their Gemini four, what's it called?

‍

Patrick Moorhead:

There's a new name to it, but it has a fun little name on that.

‍

Daniel Newman:

I can't remember, but I mean, I, have we launched the pinnacle on that one?

‍

Patrick Moorhead: 

I don't know. I don't think so. I think it's just in preview. Oh, well, I don't think, uh, I don't think that's, uh, I don't think that's available.

‍

Daniel Newman: 

Okay. It's for Argonne is what it is. Gentlemen, I, for Argonne. But anyway, so, you know, good things happening. And, you know, I think you and I have said this to almost ad nauseum on this show, but like the status today of which model is the best is today. That's how long you get, like tomorrow, something could totally change. And speaking of that, let's, you know, hit our last decode topic. Google is, you know, they seemingly are doing what a smart company with their stature, their access to enterprise, their tools, and their market cap is able to do. And they're just kind of tying it all together and building a smart intelligence layer that doesn't require you to just use Google.

‍

Patrick Moorhead: 

Yeah, so this announcement sounds a lot like what all the other companies have been talking about. And that's not necessarily a knock. It just is what it is. It's about persistent context that goes across everything you're doing in the workplace, Microsoft 365, and Slack, and it keeps going. Unlike Hermes, you can close the laptop. You close the laptop and it delegates to the sub-agents. So that's what people want, because with the best context means you get the best results. The other thing they did is they gave model choice, right? The agents support Gemini and Cloud Baby, with other private and open models being planned. And I think this is a huge step and it's what enterprises want. They don't want to be locked into a single model. We saw Microsoft really led on this when it came to agent foundry and the ability to pick the models. AWS has always been a multi-model primarily because I mean, it's the right thing to do, but they didn't have as competitive of first-party models as it needed to, to really even incent any type of lock-in. So I like the word smart routing, right? It seems to be a theme here, Daniel. We need to get our act together on routing and the efficiencies that, because different routers will have different capabilities, folks. Routing is not easy, okay? The best routers, first of all, has to identify what workload it is, what kind of workload it is. It has to determine where to send the subagents and the agents. And that is not easy because there is actually a tiny model that does that. And it has to be done at microsecond speed. Otherwise, you're going to get latency. And then all the results have to come back, be reassembled in a harness to give you that amazing result that you do as well. They're also talking about identity and enforcement. Right. Being tested together. Yes. They've got agent sandbox and agent gateway. All all pluses here. So I'm going to put a little bit of skepticism out here, folks. Google has done some really good stuff, but it seems like every time an announcement has been made, and then I go and I try to test something, the capability is just, it's not there. So I will believe it when I see it and I use it. A great example of this is the capability. I get better capability on Google Gemini with my personal account than I do in my paid workspace account. I get daily brief. I get the best models. you know, 3.8 flash with extended thinking that I can't get with my paid, uh, my paid workspace. Uh, and also memory is very difficult inside of a paid paid workspace. You have to do a lot of stuff on the backend. There's no personal context like you could have on the, uh, personal, uh, versions. So when I'm using Gemini, uh, I use my, um, I use my personal account. Final thing, where's Spark? Gemini Spark announced three, four months ago. So where is it on Workspace? It is a no-show at this point.

‍

Daniel Newman: 

Yeah, I do think the unification makes a good case. I do think roadmap and release is interesting. And what you're saying about access, like you certainly should tier it where the people who are paying are getting best access, not less access. So that's definitely something to keep an eye on. I think, you know, thematically Google and Microsoft have a unique opportunity to be the adults in the room. in terms of providing an enterprise workspace, an enterprise canvas for doing all your enterprise work in an environment, handling the routing, handling the intelligence, handling the harness, really making AI ubiquitous in the productivity and work experiences you have. These are the companies that should get this done. Whether they get this done, it's kind of like Apple was the company that should have built MUSE. And now they're cutting their component orders because nobody wants their new phone. Do you see that? Crazy. 15%.

‍

Patrick Moorhead: 

No, I did. And Daniel, you know, let's even get to the features on personal Gmail are better than the features in the workspace. Gmail from an AI, from an AI point of view, it's crazy. Do you think that's a privacy related issue?

‍

Daniel Newman: 

What's the governance around it when it goes to enterprise and work versus like the personal stuff that has basically, it's free and… I think it's protect the base.

‍

Patrick Moorhead: 

Like we have to adopt this for consumer so we don't get taken out by Apple or something.

‍

Daniel Newman: 

So it's like a consumer led thing. Like this is where they get their demand.

‍

Patrick Moorhead: 

Yeah, well, and it's crazy the amount of money they're spending, I'm sure, on consumer AI versus what they would have to do with workspace users. I mean, there's a 50 to 1 ratio between consumer Gmail and workspace. So I would have thought that they would have lit up the business side first, where they could go in and potentially take market share away from Microsoft. So it's a little bit of a head scratcher for me.

‍

Daniel Newman: 

Yeah, I don't know. Anyway, all right. Well, we've exhausted the DECO, but we've got a short bulls and bears today. So we went a little longer for y'all. Hope you appreciate it, especially the part where you got to watch me eat that egg McMuffin. That was class. Next, we're going to jump into the flip. And this is where we do a take, a prediction, inside a debate on a topic. Someone goes for it, someone goes against it. And this week, we are going to debate whether temporary construction jobs are a strong reason to keep the AI data centers moving. Let's go to the flip. So I am for it, and I actually am for it. So Pat, here's what bugs me about this debate. The minute someone wants to kill a data center, the jobs suddenly become temporary, lesser, not real. You tell that to an electrician, every construction job is temporary. The bridge ends, the hospital ends, the stadium ends. Nobody stands up at a city council and says, The hospital crew doesn't count. I mean, and look at the scale of this whole thing. A single data center can need anywhere from 1,000 to 10,000 construction workers. Apprentices there, apprentices can earn 40 to 60,000 a year. And experienced electricians are clearing six figures, earning while they learn and no four-year degree. So here's what people are missing. The skills don't evaporate when the project ends. They walk out the gate, with the worker, okay? So, you know, 20,000 electricians are expected to retire every year. 30% of union electricians are already between 50 and 70. So this build-out is basically serving as the largest apprenticeship engine in history, at least that we've seen in decades. And one project can create apprenticeship slots for hundreds of new workers. The wire for the next, you know, not just the AI data center, the next hospital, the next fab, the next substation. So let me be clear about what I'm saying. Jobs are not a blank check. Not every site gets approved, not every tax break gets handed out. And I mean, if you look at here in Texas, Abbott says projects have to cover all their electoral infrastructure costs, use no water communities need and bring household bills down. I'm good for all that. But once a project clears the bar, the construction jobs aren't a consolation prize. They're one of the strongest reasons that we should be saying, yes, right? This shouldn't be a partisan issue. This shouldn't be something you're winning votes by. We should be trying to create jobs, create hiring, create apprenticeships to have important skills, and we should be committing to developing it because our society needs this over the long term. And in the end, it's very simple. This is offense versus defense. Defense is treating this as everything's a threat to management. Everything is a threat to manage. Offense, you're going to want to negotiate for the workforce that we're trying to build. And the communities that win this cycle will bargain. They're going to bargain hard, but they shouldn't want to walk away. I mean, in the end, though, I shared a video clip on X, and this was my point. They asked this worker about these AI data centers, and they said, you know, is it a problem that the job is temporary, that these are just temporary jobs in your community? And he said, every job in my field is temporary. The jobs are temporary, but the skills aren't, the paychecks aren't, the opportunity to create a strong middle class, blue collar work is a critical thing. And this is the most important and critical onshoring opportunity that we've had as a country in more than four decades. So we need to approve these projects, we need to keep them moving forward, and we need to appreciate the fact that these are opportunities to hire local, build skills, enhance communities, and improve lives.

‍

Patrick Moorhead: 

So I actually agree with you, but I'm going to argue the opposite.

‍

Daniel Newman: 

You're supposed to say that until the end.

‍

Daniel Newman:

All right.

‍

Patrick Moorhead: 

Holy crap. Sorry, guys. We have to come back when we're due. I think we might have to. I'm so sorry. All right, Daniel, that was a great response there. And now I'm going to have to argue a side that I don't necessarily agree with, that I'm against it. So, I mean, the first thing we need to look at is the data, which is, it's one thing to talk about what's happening today. It's another thing to try to pretend that we know what the future is going to look like. But let's look at Virginia, right, where peak construction headcount was not durable employment. Virginia actually did an audit. And for every 1500 workers at peak construction, there are about 50 ongoing workers for a typi So, you know, it's not un every state and every cir it gives us indication of actually happened. And the other thing that happened in Virginia is essentially they showed that gross job totals didn't establish some new net community value, right? They did a statewide economic estimate and, you know, it included displaced activity, tax receipts, and employment, and it just didn't pan out. I keep going back to Virginia here, but they have the most data centers of any other state. The electricity costs, the increase in that can outlast the construction wages. So what they showed was the increase in energy spend, the net was actually higher than the gross tax receipts from those people who were who are left over. So I think a promise of any type of promise here should become measurable obligation before approval. So in other words, and I think Microsoft is a great case study on here. making the commitments up front, and then holding people accountable to it is the only way that it's going to happen. But for right now, we've got Virginia as the case study for, you know, a typical 250,000 square foot, mega AI token factory, it just doesn't add up. I rest my case.

‍

Daniel Newman: 

Did I convince you? You had the hard one on that one. I don't I don't even know what you're talking about.

‍

Patrick Moorhead: 

Yeah, that's good. No, it's such nonsense. And I did really appreciate the video and maybe you had sub tweeted it that that the guys like Of course, it's temporary. My entire life is temporary. I go from project to project, and there's nothing that's guaranteed. But I do love it when it's there.

‍

Daniel Newman: 

I got a good reminder. We went to the a mat to the epic center. And you just saw that construction going on. And you're like, for like a few years, this was their lives is it and then it'll be done. And then like, instruction goes on, they go build something else. And I don't understand how we've somehow made this into a bad thing. Like lots and lots of construction. We, you know, we talk about, you know, that, that college isn't for everyone and traditional academics aren't for everyone. And certainly philosophy degrees are not required for everyone, but like, where else can you make a great living?

‍

Patrick Moorhead: 

Yeah.

‍

Daniel Newman: 

Yeah. So I'm, I'm for it. I mean, I do think that the, communities and the company's building can work together to solve some problems and make sure that, you know, a playground is built and some roads or a school gets built or whatever it takes, right? You know, from this, but like that kind of bartering has been done all of history, but you're shutting down lots of really high quality jobs by making this so political. All right. You know, we got a light week, but there's always market stuff to talk about. So let's talk about a couple of highlights from the market. Bulls and bears. All right, Pat. I mean, I don't know if it's Monday or people listening on the weekend or the last Thursday, like, you know, markets were selling off, probably blame more on bond yields. But then all of a sudden, this grossing netting conversation about open AI came out and. Financial Times headline says it's not 70 in revenue, ARR it's 50 in annualized, which I couldn't even tell if they meant ARR or not, but it spooked everybody that OpenAI is well behind. What the heck is going on?

‍

Patrick Moorhead: 

Yeah, so first of all, financial times should be absolutely embarrassed. The word financial usually means, you know, the difference between net revenue and annual recurring revenue, aka ARR, that is more of a consistent flow of that. And your ARR will typically be below the actual net revenue number. So with that said, the market absolutely took a dump and they rotated, well, the chip names, right? But then it rotated into other areas like software, energy and other elements like that. And, you know, here we are on Friday, right, when we're doing this, and there still hasn't been a recovery of the chip stocks. You know, there's a couple here and there, they're up. a little bit, but we still have not seen this recovery yet. And Daniel, you talked about this a little earlier, and when I get on broadcast, the way I describe this is when you have a heated market like it is right now, And forget about the belief system of whether it's going to keep going. It's all about, it's gone up so much, they're looking for any indication of bearishness, any piece of data, and whether it's deep seek, the deep seek moment, and the other three or four things we've seen out here. that have collapsed the market and then it goes and it makes the run back up. But I was kind of surprised that we didn't see a run back up yet.

‍

Daniel Newman: 

Yeah, I mean, it's only been a day and people are still digesting and there's still a lot of macro factors. Like I said, we still have bonds and interest rates rising with seemingly no idea. And that all affects AI in construction. I mean, demand is pretty fungible, but the cost of financing and building is going to go up. People are concerned about that. And I think there is this kind of constant drip of concern about what happens if open models overtake frontier, what happens to all that value and valuation that's been created? What if there's no money in models themselves? I think there will be a place for frontier intelligence, but there's questions to be answered there. And then of course, if there's, not, you know, like I said, what's the fungibility of all those commitments that OpenAI and Anthropic have made to build all this data and use all these chips? Does it go elsewhere? And do other people, you know, I think it, you know, what the moral here is, it's very, the market's very volatile around this topic. And it's very easy to trip people up and create, I call them mostly algorithmic selling. This isn't people you know, actively logging into their accounts and selling stuff. This is high frequency, you know, taking a headline from what would be considered a trusted news source and acting on it. And then what happens is when the first large block starts selling, others follow. And it's a cascade and it's a chain effect. But you know what they say in markets, Pat? It's the escalator up and the elevator down.

‍

Patrick Moorhead: 

So by the way, that's a good one liner, by the way. So are algorithmic trading actually can respond to a headline?

‍

Daniel Newman: 

Absolutely. I mean, not every fund does it exactly that way, but there's absolutely news flow as part of trading data. Okay. All right. I mean, we've built a living model that actually uses news data to update a model in real time. So it may actually be the way the news impacts a model and the model then triggers a action. That's good. From future on. We're pretty awesome. All right. Next topic, and this was a one that kind of hits the same thing, but, you know, NVIDIA has been under, you know, people have been frustrated with that name for a while, despite the fact that it just keeps executing and making piles of cash. The stock had basically not moved in like a year. AMD is running to a trillion up like six fold in one year and NVIDIA kind of stayed within a range bound. Well, good news, I guess, for the NVIDIA brethren, although not as good after Thursday, was NVIDIA did hit over 240 this week and it was an all time high. It was getting darn close to a six trillion dollar market cap. This is a company that I still think fundamentally should be a $10 trillion company in the next couple of years. If it keeps executing, it's valued very cheaply, if you look at the price earnings and growth of the company. But it feels like this is a company that's priced for any mistake it might make, whereas it's competition across the street, which is Team Red. Is Team Red AMD?

‍

Patrick Moorhead: 

Yes, even though since 1968, their colors were green.

‍

Daniel Newman: 

Yeah. So team red though, you know, has been getting valued for basically perfect execution with no operational risk. And NVIDIA has been at the opposite end. So, you know, I still think it's, by the way, both are going to do great. It's not a knock on that. It's just interesting how an incumbent gets treated versus the new fast rising competitor in any space.

‍

Patrick Moorhead: 

Yeah. I'm waiting for the moment though, when, what was it? NVIDIA is headed toward, you know, 500 billion and 750 and then a trillion just in revenue. Yeah. Right. I'm wondering when the stock actually has another breakout moment like it had last year. I mean, there's this notion that it's bigger than the GDP of France and a couple other countries combined. When something that big gets to that size, different rules apply there. I'm gonna have to say congratulations to the bag holders at NVIDIA. And typically bag holders is a negative connotation, but if you've got the bag and it's full of NVIDIA shares, you're pretty happy right now.

‍

Daniel Newman: 

Depending when you bought them. Last topic, Pat. Samsung, you know, I think they, I don't know exactly what the number was, but I think in a conversion from the Korean won, made something, or is expecting to have made something like 80 billion U.S. dollars profit last quarter. Pretty wild.

‍

Patrick Moorhead: 

Pretty amazing. And it's so funny. I mean, even though they missed on revenue, they met on operating profit, they met on EPS, still you got to stand back. And this is literally Nvidia sized operating profit, okay? Which I believe that I did a napkin analysis that said that Even Micron will make more cash than Nvidia will next year, which is absolutely amazing. Obviously, if Micron can keep it moving here. But yeah, good stuff. It's so funny. Samsung for a while on two nanometer was in a hole, and they had a hard time getting there. But if you look at what they've done with that, rumor has it that the only Vera Rubin HBM that's making it in and passing the power tests is Samsung and HBM4E. And the other thing that I've been thinking through is this whole notion of having the base die. So the base die is what connects to the accelerator. And as kind of a traffic cop, Samsung is the only company that does its own base die. SK Hynix and Micron farm that out to TSMC. And I'm just wondering on the thermals, if that had something to do with it. And the thing about Samsung is they haven't even hit their stride on Foundry. right? It did take them a while to get 10 nanometer lined up, but it's interesting, Lisa Sue had made a comment. So I've always said that AMD is gonna be the last one to do this, and I don't know if Lisa is saying this just for the sake of the talking point, but she had mentioned that she was open to looking at others, and this was while she was over in Korea. So that's probably the, I don't know, the most convincing statement that AMD might do something. And it could make sense for set-top box game console chips. It could make sense for certain lower risk lines that they have over there. And then there's the backdrop of, I don't think that Samsung is getting the heavy discounts that, you know, it was when it first was trying to really field a lot more of these foundry customers.

‍

Daniel Newman: 

Look, there's a company that had been like sort of shot dead. They were treated like they'd made, done nothing right. The process technologies, you know, were, were, were questioned and, you know, memory wasn't hot yet. I mean, do you remember, I got, I remember some of the conversations, things we'd heard, it was like, you know, that the TSMC had just run away and that Samsung was not in the picture and them and Intel were fighting for scraps. And I mean, like you said, there's still some opportunity in foundry, but this memory boon has been an absolute game changer. The capital resourcing it's going to have now, they're going to become you know, they're going to become a powerful mafia. I mean, in a good way. In terms of the capital they're going to have and the abilities they're going to have to invest, support customers, to support R&D. You know, I said the same thing about Micron and Samsung's going to You know, the MNSK are going to become incredibly powerful beyond what had ever been expected from either of these companies. But in terms of on the process side, there's still a ton of growth there. So very interesting to see. I mean, 80 billion, it's hard to even imagine, but that's significantly more than even NVIDIA has ever made in profit in a quarter, which is just crazy how quickly it went from where it was to where they're at. Congratulations to them. So big win. I think we have like one more week, Pat, before like the next crazy wave of tech earnings comes around. Could get interesting here. Everybody's gonna wait. Will anything crack next quarter? Any of the hyperscalers? You got three weeks before the hyperscalers are all gonna report. Do they all increase their CapEx guides? Um, and then of course, what erroneous headline could come out to just confuse the market enough to another tizzy.

‍

Patrick Moorhead: I've already seen him. I've already seen some already related to Nvidia that, you know, their revenue is going to be down, uh, down versus, uh, forecast or expectations because of, uh, because of power. That that's a, a new one. Um, uh, I'm seeing right now.

‍

Daniel Newman: 

Oh, great. Well, then you better sell everything and get a bunker. Nobody's going to use AI anyway. That's what I heard. It's just a fad, folks. Just a fad. It's SI. Everyone's going to use super intelligence. All right. Well, Pat, we did it. Another show, 323 in the books. Appreciate everybody tuning in, being part of our community. Subscribe. Share this with your friends. See you all later.

‍