 Hello, and welcome to the Cube Pod episode 38. I'm John Furrier with Dave Vellante, my sidekick co-host of the Cube here for our weekly pod where we unpack all the latest trends, technology, hot takes, deep dives, breaking analysis, event coverage, all things enterprise and emerging tech. Obviously AI is continuing to dominate. Dave, great to see you. And we got a ton to unpack. I'm in Seattle actually right now in the Westin. Oh, I thought you were in Denver. Jess flew in from Denver this morning. I've been in supercomputing here to see Adam Sileski for the one-on-one pre-reinvent exclusive. I got an hour with him. I'm gonna pop and see Andy Jassy tonight. Check the hockey game out. Here is the cracker. Nice, who are they playing? Who are they playing? They're playing the New York Islanders. Oh, nice. Looking forward to two teams with not so good records, both under 500, but should be great to see Andy. I haven't seen him in a while since he took over the CEO jobs, would be great to connect. Dave, supercomputing in Denver was just there at the team. I was. The whole team coverage. You were in Boston doing your breaking analysis. It was just covering Microsoft Ignite and checking all that out. It's AI everything. Dave, we're announcing today on theCUBE that we're coming out with theCUBE chip, the 2,500 series. Everyone's got chips, Dave. Is it arm-based? No, we're going to start a fab plant. Billions of dollars. Were there people announcing chips? I didn't really, it wasn't plugged into the news at supercomputing. I was doing other stuff watching it at night, but were there chips announced at supercomputing? Well, everyone's has chips. So we cover the CEO of GROC, which has a chip that does language processing, LPU. They trademark that term. It does inferences at speeds. It's unbelievable. People who got chips, a networking chip. We got Broadcom there pushing the new Ethernet speeds, open X86 architectures, PCI-3, PCI-3, I mean six generation, just a ton of chips around all the enterprise data center world. It's a chip game. And Microsoft announced two new chips, Maya and Cobalt, which are like, I guess you could call it like, Maya's a accelerator, and then Cobalt's a cloud chip. You could call it like Tranium and Ferentia, Tranium and Ferentia and Graviton. They'll hate that comparison. But here's my question, John. So these are all arm-based or most of them are arm-based chips. Does Amazon's lead, because Amazon, I was just researching this because I want to put it in my breaking analysis. Amazon announced Graviton in 2018, I think that actually began shipping, in Ferentia in 2019, Tranium in 2021. So they got a big lead, but is that lead sustainable? Is it like no compression algorithm for experience or is there just so much innovation around these chip designs? And you got TSM, FAB, although they're capacity limited, that you'll be able to compress that kind of market advantage. Do you have a sense of that? You know, that's a great question. And my first initial response would be just from a traditional competitive standpoint trajectory is that they have a great lead and that lead's gonna give them experience, as you said. However, the AI workloads are more complex than traditional workloads, especially traditional cloud workload. I hate to even use that term. I mean, I think that's the first one I've ever said that the cloud is a traditional workload. But we're seeing customers, look at the complexity of the AI deployments right now and they're different. And so I think that's the only wild card. I'm gonna ask Adam Sileski that when I sit down with them around how their advances in the silicon are gonna be. My guess is they're probably on top of it. They do have an advantage. The other thing that's gonna be interesting that came out of Supercomputing that will probably play into reinvent is the toughest challenges around networking, Dave, because speed is gonna come into the processor side. So do you optimize for GPU, CPUs or TPUs? Or do you want more compute and GPU processing power? Or do you optimize for networking? This is a really big discussion. I think that the general consensus on the bare metal data center, kind of like the supercomputing world of high performance computing, they want more GPUs because they can compete at scale with the hyperscalers. Amazon and Azure will probably wanna get better with networking because they got the complex scale on their side. So it's an interesting perspective, depending on what view you're looking at that problem. So if you're in the cloud, you got scale. If you're going to be coming up from the chip layer and building these next gen super clouds, which are either tier two clouds but have this super cloud power dynamic, you're going to come from the chip level and essentially create a white label cloud-like environment in a data center. And it's not just the traditional networking that we're talking about here. You're talking about the interconnectivity within the complex. So you got all these alternative processors. You got CPUs, you got GPUs, you got NPUs, you got accelerators. You have all these other supporting components, controllers and they got to talk to each other, right? Used to all go through for decades. We've all gone through the the x86 and all the memory management, everything went through the x86. That's getting completely blown away. So that internet working becomes really important. That's why, you know, Melanox, the acquisition that Nvidia made, the Melanox is so important within Finneban. Of course, everybody's trying to shift the entire industry to Ethernet because, you know, Jensen's gouging everybody because he's got the competitive advantage. Well, what's the Ethernet's coming up to 400, 800, you know, gig, Dave. So like that's faster, right? That's getting faster. So this is interesting. I was talking to the Broadcom people about this at the Dell sponsored effort we had. And the Broadcom people who are telling me that the Ethernet ecosystem is democratizing the networking because the x86 standard and Chaz Trembly, who was on the cube this morning, I couldn't do the interviews on the plane to come for the meetings with Amazon, what's saying on the cube directly, their investment of multi years, many, many, many years of x86 ecosystem is gonna not try to replace the GPU. It's gonna be for the interconnect around it. So the question that came out of Supercomputing was how two things came out of Supercomputing 23. Can I get my hands on some GPUs? Number one, and everyone's fighting for that. And number two, how do I build around it with chips and interconnect and networking? So you're starting to see kind of like, like the old motherboards on a PC or a server. You got a processor, you got a bunch of stuff around it. The cloud scale is essentially like the same thing. So we've got an edge deployment. That's gonna be much different than say an on-premise core deployment or a cloud native public cloud. So if you're running cloud operations, then it should operate together, but they're just different environments. So you're gonna need standards to handle some of these plug and play components. I think that's where x86 shines not as a replacement for the GPU power. And Infiniband is at risk because if Ethan has a standard, then is that really providing the kind of benefits there? So that's gonna be the big question. And everyone's like, oh yeah, Infiniband is dead. I mean, Infiniband is not dead. That's bullshit, okay? I mean, it's under attack, but it's not dead. No, I know, but it's not as relevant. That's like saying Nvidia is dead. I mean, Nvidia is doing Ethernet too. Well, it comes back down to, is it a requirement as a linchpin of a value? And I would speculate and say that networking, standard networking with Ethernet, that's gonna be pushing higher and higher speeds. I would say that's Broadcom's agenda as an alternative and this doesn't make Infiniband that valuable if the platform is- Well, it definitely attacks Infiniband because Nvidia's got the market locked up. But to me, the other key point though that you're making is that, again, I go back to, X86 used to have the monopoly on everything. They owned everything, they controlled everything and that's changing. We're shifting from a CPU centric world to a connect centric world. And that interconnectivity is gonna increasingly become important for performance and cost and reliability because you're moving all this data around internally and because you're seeing just the processor complexes are getting so incredibly complicated and so that's where, I think that is Broadcom's play and then all these other alternative chips, I think what they're doing is they're customizing it for their purpose built use case, which to me ties into the Gen AI power law. You're gonna have all these things- The power law that you have that we have is matches exactly what's happening in the cloud scale. You have specialty clouds emerging as we talk about with SuperCloud and had two people come on theCUBE and validate SuperCloud and they said, I know this one guy, he's like a legend, he worked at both, he worked at Son back in the old days, Renu, Ramon, he's seen multiple ways of innovation. And he goes, I finally realized what you guys meant by SuperCloud because he saw it at supercomputing is that as the chip, do you integrate up from the chip level and create this kind of second tier, bare metal, disaggregated memory fabric with LLMs as the abstraction layer, foundation models as the abstraction layer to essentially run all the infrastructure completely frictionless and transparent. So he's seeing this, seeing the second wave and an example we use was CoreWeaver, a company it's in the weeds, CoreWeaver is basically enabled by NVIDIA. Yeah, it's a GPU cloud, right? It's a GPU cloud, but it's genius. That's the NVIDIA cloud that they announced at their event that they didn't even mention AWS. So NVIDIA is playing both sides here because they're in the clouds and they're in other areas. So this was a big announcement at Ignite, Jensen was on stage, Jensen's everywhere, of course, but he's on stage at Microsoft Ignite. And basically they were emphasizing the size of these NVIDIA supercomputer clusters. I got my notes here. It's not Azure infrastructure, it's not this Azure boost, which is like their Nitro. This is NVIDIA systems infrastructure. It's got a thin layer of Azure software to help open AI train and run their models. And but basically they're outsourcing the infrastructure, the GPU infrastructure to NVIDIA. Which I'm starting from wrong, but AWS was sort of fighting that, right? And Amazon and Microsoft's embracing it. Do I have that right? NVIDIA's DGX cloud, they're calling it, enables a new market, the super cloud market, Dave. It's already happening. And this has came out of nowhere. It's like left field. So CoreWeave is a great example. It's got 2.3 billion in debt financing from an asset management firms. And all they do is focus on the AI market in its hardware. And they sell GPU based virtual machines, which is perfectly suited for AI workloads. That's what they're doing. And they have other competitors like Paperspace, Lambda Labs and a few others. But the specialism around that's working. Dell announced the AI server, Broadcom again, standing right next to holding hands. They did, they made that public. I wasn't sure if that was public because I was just down there a couple of weeks ago. They would tell you whatever. If it's NDA, it's not anymore. Who's they talked about on theCUBE today? So this new Silicon. So I was bringing this up as on theCUBE as kind of this new, I didn't say is a cold war emerging, but you have the Silicon players and the hardware players versus the cloud, okay? The battle for AI supremacy was the theme of super cloud five, the week after Thanksgiving, during the week of reinvent, we're having an event in studio in Palo Alto is an interesting kind of a cold war dynamic because the suppliers of Silicon have the hyperscalers as their customers and the hyperscalers set the agenda because they have the scale and they're buying power. They got financial leverage. However though, the Silicon players have great leverage too in the sense that they're making a lot of money. So as the Silicon players look at the clouds built in their own Silicon like AWS, there's this emerging kind of game going on. So this tier two clouds, as we call them, super clouds are emerging. You got NVIDIA and the semiconductor companies have the scale and leverage to do a similar cloud with bare metal and some of these hosting providers were seeing emerge out of the HPC world. So the HPC world could be bursting competitors in the marketplace with this fabulous concept of chips kind of going cloud-fabulous, if you will. And if this explodes, then you're just going to create- But aren't I correct? Wasn't Amazon sort of resisting NVIDIA? Didn't NVIDIA want to do this on the AWS cloud and Amazon thought it? I don't have any data. I don't have any data on that. I may be confused. I thought somebody told me that, that NVIDIA want to do it. And then here we see Microsoft really embracing that. I don't know. All I know is that when NVIDIA launched DGX cloud and the Grace Hopper stuff, the killer, next gen stuff, they did not mention AWS. Azure was on stage, not AWS. So NVIDIA is part of Microsoft. I think it's because AWS was saying, not in our cloud, we're the cloud. And so now you see all these alternative clouds emerging. Now they don't have the AWS scale, but they got GPUs. So the Renew could get them in the queue. I brought them on because he's been sub-stacking this concept. He was in that generation like us. He's about our age. Remember the white boxes, Dave, in the 90s? The ODMs, yeah, right. ODMs, why did they do that? Because they can build similar functionality at no brand on it and people were building large-scale stuff, hosting environments and whatnot, clients, server integrations. Right, and that was one of those, your margin is my opportunity, thanks. Yeah, ISPs were emerging at that time, if you remember. So his point is that movie's playing out again with the cloud, where these NVIDIA DGX enabled market because they just leveraging their NVIDIA software and their cloud platform with that, using that bully pulpit of the GPU leverage and enabling this bare metal data center redefine concepts as clouds, specialty clouds. So your power law maps directly to where the AI market's going directly with specialty AI clouds. So, and that's where the LLMs and the foundation models come in, Dave, because the speculation at Supercomputing, which will probably come in to reinvent, is that the foundation models will be the abstraction layer that feeds the interface of AI. If you look at the Microsoft announcements this week, everything's being repositioned as an interface change. Oh, it changed the developers, co-pilots, being as- Co-pilots everywhere. Co-pilots are going to be the interface through technology. So if the interface has changed, what's feeding the interface? And what's coming out of this is that, that abstraction layer is going to be the power law of the foundation models, which then also means you're going to need to have infrastructure, which is trained data and inference. And it was very clear coming out of HPC, Supercomputing event, that even though training costs a lot, Dave, it's inferences where the value extraction is. So we had a conversation with the Grock CEO, and we were saying, you know, training is like a sandbox. You got to do it. You spend money on it. You set it up. The inference happens over time. And the inference is the killer app because that's where the value extraction is, and that's where the iteration happens. That's where the AI gets better and more personable and making them use as more productive. So you nailed this two pods ago when banging the same drum. The productivity gains will be the outcome that will be part of this new application shift. If no one's productive with your app, it's dead, period. That's the new AI, right? If it's not making it productive, it's not going to last at all. Now it could be productive under the covers. It could be productive in the stack. So the tech stack is merging. And this is where the Broadcoms and the Dells and the HPEs are going to win because they have their eyes on this HPC convergence with AI. And that's going to create a whole new ecosystem. So you just threw out about with eight topics, but let me go back to what you're saying about the co-pilots and the interface. I got a quote from Satya at his keynote. He said, the way to think about this co-pilot, co-pilot will be the new UI that helps us gain access to the world's knowledge and your organization's knowledge. But most importantly, it's your agent that helps you act on that knowledge. So the reason why that's so important is because they're talking about this Microsoft graph. So it's all your apps, all your services and the infrastructure that supports them. And so what they've done with this graph is they've made all those things coherent. So your co-pilot now has access to all that knowledge. There's a semantic layer that makes all those different elements coherent. And so that allows the co-pilot to actually act. It's a system of agency. And the reason why this is so important is because Microsoft has all this productivity software sitting on top of Azure and they're feeding it with compute, storage, networking and platform as a service and database. And that's a massive flywheel from Microsoft because they have this captive business that feeds Azure, it's like feeding the beast. So this is huge and it's a new era. It's the co-pilot era that we're entering. Yeah, I mean, that's absolutely right. And you nailed it when we were having that debate about productivity. And so, the interface is the new changeover. And by the way, Microsoft's trying to land grab that. And I'll throw out a debate topic right now, but I think Microsoft is trying to stall the AI momentum. They're trying to slow down AI. I think Microsoft is quietly creating a campaign to slow down the momentum of AI. And you're gonna laugh at that. So I know you're gonna just throw up and- I need to think about it. If you look at their event, you're like, what are you talking about? They're pumping AI all over the place. Right, so defend that statement. Okay, so here's my conspiracy theory on this. Microsoft has a short-term advantage with AI right now. They have all the apps. Clearly. Okay, everything that they're doing is about Microsoft. Microsoft products, teams, co-pilot. They even changed Bing, the surgeon that nobody uses to co-pilot. Why? Because co-pilot is, the co-pilot sounds cooler than Bing. Bing has failed to adopt. Okay, so, and they have applications that got installed base. And our premise on the pod many times, we've been saying this, that the enterprise, forget to put the consumers up. That's part of the side. The enterprise value is making AI, to making your legacy stuff, your current stuff better, and then creating net new advantages. That's the playbook that Microsoft needs to do to take advantage of the next level, because they can use their current base to get the momentum, and they are doing that. But what the best part about AI right now is that it's so good on coding, that I could actually create an alternative to Microsoft's products the way they are now. But so if Microsoft can get through the chasm here, make this hurdle jump to PowerPoint, Word, basic products like Teams, their products aren't really good from an AI perspective. But they're making them better now. But what's going to prevent an alternative to be created there? That is a hell of, that's a, wow, the way your mind works. So basically what you're saying is that all these co-pilots could potentially create an unintended consequence of making it so easy to create software that's competitive to Microsoft's estate. Microsoft's current estate is inimitable as they get more AI to their shit. That's a wild theory. That's a wild theory. And so I gotta listen to that. It's wild, it's legit because Amazon doesn't have apps. So as everyone goes into re-invent, they're gonna judge Amazon and say, hey, AWS, you're not as good as Microsoft. But on paper, Microsoft actually sucks right now, but they're gonna make their products more AI enabled so better than that'll create a barrier to entry for a startup to replicate an integrated co-pilot high data model, and they fix Azure, they get Azure up and running. Microsoft's executing a competitive strategy, in my opinion, that's pretty awesome. And if I'm Microsoft, I'm saying our current pre-AI products could all be replicated with AI within a year. Okay, but so wait. So on paper, Microsoft's kicking ass. Yeah, with announcements with co-pilots, they're investing, they're changing old products. And just their business performance, their business performance is off the charts. But so you got all these, so here's the co-pilots and I'm probably missing some. You got GitHub, which is for pro, the GitHub co-pilot, which is for professional developers. You got Studio, which is for end users. You got this power platform that's for citizen devs. You got the 365 co-pilot, which is for end users. You got the search co-pilot, they renamed Bing. They announced all these Azure ops co-pilots, like security co-pilot. So they're just like pushing co-pilots everywhere. And because they have this data fabric, this data element that's coherent now, they are in a like really strong position. And to your point, AWS doesn't have that up the stack of software. So Google, who's got workspaces and Microsoft who have these productivity software capabilities have an advantage because it's direct productivity hit. So it goes from, so developers, you can say developing software, but then that's going to make everybody a developer and it's going to drive end user productivity. I would bet that Microsoft's going to get there because of its huge install base. People, it's just so easy for people to say, yeah, turn on our co-pilot for $30 per user per month. Well, I mean, they have an install base that just adding another charge. Everyone loves open AI and they put 10 billion and they're going to make a hundred billion on it roughly more, more. And they're changing their products. My point is Microsoft's incentive is to slow everybody else down while they change their product for the better and make it a little bit sticky or a little inimitable. So that's hard to copy, right? So that's to me. Because their products, they do suck. And they suck in the sense that they're just too damn complicated. You try to, you know, the way you used to be able to just intuit Excel, you know, and even PowerPoint, there's just so many buttons. So if you can talk, I mean, if it actually really works, which it seems to when I've seen it, if you can talk to the tool and have it do what you want it to do, that's super powerful. But you remember early, early AWS reinvent, you, Jerry Chen and I were on the cube. And we asked Jerry, we were riffing at the time. Do you think Microsoft's going to go up stack? And Jerry said, you know, he was very articulate. He said, you know, I don't think so. I think there's strategies to enable developers to build applications that will compete with the SaaS vendors. Now, in many respects, Amazon's done that. They've got a lot of ISVs, but it certainly hasn't disrupted the sales forces and the service nows and the snowflakes and the Mongos. I mean, they've become partners. And in a way, I guess they are developing on AWS, but it certainly hasn't affected Microsoft. So the personal productivity piece seems to be the bastion of Microsoft. And they've got a pretty strong mode. I mean, not even Google workspaces. I mean, we use it, but it's deficient when compared to Microsoft tools. I mean, Google's sheets, that doesn't even add correctly. They get so many bugs in it. That was your rant last week. I mean, but your point is that Microsoft has an advantage. And this is why AI is interesting to me because they're recognizing probably in the smoky rooms that they talk strategy that if they don't jump on AI and make their stuff kind of more aligned with the expectations for the interface as Satya's pointing out, which he's right, every interface has changed. So the thing about this wave is, no matter what area you do, whether it's improving code assistance or learning exploration, education, finance, every application expectation will have a new interface to it, a search-like interface, a personalization interface. Things will be personal. AI will make things better because the abstraction that's feeding that is data in the form of foundation models. And I think our power law research shows specifically. And what's interesting, Dave, is that coming out of supercomputing, the data center world's transforming because supercomputing, HPC is now the new cloud-like capability for off-cloud, meaning data centers. So that distribution is following the power law of the models. So if you look at the specialty clouds, they're almost mapping to our power law on the foundation models. So foundation models create foundation clouds. And under the coverage, you'll see like all kinds of hardware configurations, GPU this, interconnects, faster networking, glue layers, which will probably be coded by AI itself, right? So this is like the mind-blowing aspect of the infrastructure. So I was obviously not in the road this week for a change. And I was watching the Ignite presentations. And so is Sarbjit Johal, a CUBE collective friend and George Gilbert, the CUBE research analyst. And George made an interesting observation. He said, you've never seen a transition that where you've got this massive demand for software, sort of this accelerated software demand. And you have software that makes it easier to build software in the form of co-pilots. And it's like you have this double whammy that to your point, you know, could have a unintended consequence of disrupting Microsoft. Well, his point is exactly what I'm, he's getting exactly the same kind of, where I'm going with it, which is the demand's high and the coding is becoming the commodity. So if the products aren't going to be evolving, like Microsoft's doing a great job of trying to do that fast and you can see them. So they're incentive to slow them, the entrepreneurs down because they're going to maintain their position. So of course the big guys, we've been saying to the CUBE with all this regulation bullshit, like they're going to try to slow down the entrepreneurial side. So I think you're going to see that grow. Now, here's the wild card in all this. I think it doesn't matter because one of the things that I'm observing, Dave, is the market's getting bigger. It's not a zero sum game. It's not like it's IT budgets fighting for each other. It's trillions of dollars of new value. Okay. And look, if you look at the GPU spending, okay, throw away the sustainability problem that it has, but if you look at the demand for GPUs, people are buying these things up like by the truckload if they can get their hands on them. And you know what they're doing? They're not even thinking about the TCO. They're spending millions and millions of dollars on GPUs just to get them. And then they're like, wait, I got them. Now what do I do with them? So the next question is what do you do with them? And that's the question people are like asking right now. Okay, I got some data. Do I do in a large language model and do inference? Do I bring in small language models and have them interact with each other through APIs? This is the open book question that's not yet answered. So you're going to see a massive collision. This battle for AI supremacy that's going on right now absolutely has got high stakes. But the issue is it's not a zero sum game. No one loses unless they don't have a good product. So the market's getting bigger money-wise. So I think that's going to make the entrepreneurial opportunity great. And that's what's coming out of all these CUBE conversations and these reports we're doing is that it's what we don't see that's going to happen. How do you know what the next app's going to be if the interface has changed and whoever controls the model abstraction layer can create value? So Microsoft certainly will dominate on their high end but an entrepreneur could come out and create something just as good maybe. So that's kind of my angle on that. So you're going to ask, you won't get invited back I guess if you do, but are you going to ask Adam if his cloud is legacy? It might kick you out if you do that. No, I might ask him what I used that word earlier in the pod and I said traditional workloads. I hear that a lot. I hear quantum conversations I hear with AI when it's going to be in production with traditional workloads. And what I was saying earlier and what's coming out of these AI infrastructure conversations is the complexity required if you're taking a traditional approach, buying servers, loading RAM and memory in there, SSDs, interconnect systems and putting that all together. If you don't have standards, it doesn't really work. So like it's complex. So do you run inference at scale? What are those clusters look like? How do you stand them up? So there's a big nerd conversation going on around that. So because it's complex, it's going to be hard to solve. So I think Amazon could be viewed as a traditional workload, traditional cloud. That's the question I'm going to ask. You guys in the traditional cloud last year we talked about, I talked about, we talked about next-gen cloud. And he's like, no, they're ISVs. Well, I think we took the cake on that one, Dave. I think we proved our point with SuperCloud that the ISV model, certainly with the AI trend coming is you're either an ISV building software apps with AI or you're a platform. So AI is very platform, system-specific. And I'm going to ask him that. I'll ask him about his AI position perception-wise. Does, do they think they're behind? And that's going to be an easy one he'll probably reflect the, you know, we're in the first three steps of a marathon. But I'm expecting an answer on that. But I'm not sure he's not going to be... Oh, that's the narrative right now. I mean that they're behind, that they're sort of legacy cloud. And I think, I don't know, John. I think the, look, I've said this a number of times. They've got the last word this year and I'd be shocked if they don't put forth like a really impressive performance. I mean, in July, they sort of cobbled together their little AI day to try to keep some momentum. Hey, you know, kind of, we're relevant too. And, but they announced a bunch of stuff that was, you know, pre-announced. And so they've just GA'd bedrock. I got to believe that it's just convenient for developers who are in AWS already. They just announced today a new, they announced that they just announced today. They're not doing press releases on these. They're releasing them all on LinkedIn. Matt Wood and Swami just posted introducing Party Rock, Amazon's new AI playground that lets you experiment, learn and build apps with Genovi in minutes. Powered by bedrock. Bitsy just dropped a post. If there's classic fits, there's nothing new here today except he's moving to WordPress. Sometimes I love the guy, but that's like a, I'm a dinosaur, I'm moving to WordPress. Which version? New and improved platform andomics. That's, that's good. Anyway, there's no, no fun stuff. I don't get the joke, why WordPress? I don't know. He said, and this is a new design. I'm also moving to the WordPress mothership. I've run this blog, I've run this blog using WordPress. Oh, so he's moving to WordPress.com. He's used some other hosting service. He's probably using Rackspace. I don't know. I considered substack, but that was too cool for me. No, I was just kidding. But WordPress can do everything substack can. Well, I guess, but substacks like a community, right? I guess WordPress.com is too, so. Yeah, WordPress.com is not as strong as hosting. No, God, substacks like got all the momentum. Yeah, substacks got, it's really strong over WordPress. He must have had a deal in there. Yeah, maybe. So what else, who impressed you at Supercomput? Were there any standouts? I mean, who was there? HP was there, Dell was there, Vast was there, WECA was there. I was very impressed by Seamus over at Dell. Seamus Jones, he got me an invite. He recommended I go to the Dell HPC community event. That was the day before it started on Monday, Monday afternoon. And it wasn't a Dell event, it was an industry event, but I think they were sponsoring it. And Armando went up there and gave a talk and they had all these customers coming and bowing all the HPC legends up there. Talking about Intel was there, you had Nvidia, it was a real multi-vendor and it was a really good sharing environment, but they laid out all the issues. And it was like a little bit of like the elite getting together out there and talking about it. And it was awesome. And I learned a lot. I learned that the HPC, well, I knew a lot of the HPC community in general, but I didn't know what was the mindset. There are all the issues. It's like classic data center folks, but they're totally pumped because this HPC community started when I graduated in college in 1998 from Northeastern, you know, young CS major. And when I worked at HP, we did a little HP, Eulah Packard was all over. They made machine, they made hardware and chips, but this community with the AI wave coming in absolutely vindicates all the years of hard work and grinding that they've been doing. Cause it's mostly been academic and big moonshot, large scale projects, you know, weather disasters, evaluating oil spills, you know, big supercomputer things that were, you know, pre-cloud where just build big iron, right? And with cloud, you got scale, that helped a lot. And now with this wave coming with the GPUs, you're starting to see HPC as a service. And the data governance models are going to be upside down. They discuss things like model management. How do you create reproducibility? What, what do you test? How do you measure model governance vertical approaches? So, you know, the theme there was that AI vindicates this computing paradigm at scale because that's the next gen level that the cloud takes everybody. So the next gen cloud has been going on for the past two, three years that we were reporting on now becomes high performance computing, but it's not just processors, it's GPUs, it's other chips. And so the semiconductors combined with software now combined with this foundational model AI, the generative AI, you now have a complete generational new era. Interface changed, abstraction layers are gonna change, how software is built is changing. AI hardware will dominate, AI will need more networking, and the hyperscalers are gonna continue to dominate the trend because they're the customer of all the semiconductors. So you'll see Amazon in my opinion and Azure and Google be the leaders in how they design their stuff. So to your earlier question, I think Amazon will have a good lead with Graviton and their chips. The question is how do they integrate that into their stack because they don't have apps except for call center, a few other things. And that's gonna come back down to their ecosystem. If Amazon can take their ecosystem today and enable their ecosystem to do what Microsoft's doing for itself as we talked about, make co-pilot in Word and Teams and all that Bing stuff, that's what their AI infusing their products. If Amazon can take their entire ecosystem to their partner network and make everyone AI superstars or super clouds, they're gonna win. That's a winning strategy because they're gonna say, we're gonna let the market create the apps faster and we'll let you, Microsoft, lean on your competitive strategy of owning the apps for the enterprise and let's see where it all falls. And like I said, no one loses because the market's growing. And then you got the hyperscalers and you got these NVIDIA clouds that are like in these data centers and then the edge, Dave, it's just a monster market. It's not gonna go down. And I guarantee you that the forecast will change significantly. So my takeaway from supercomputing is HPC and AI is real. There's gonna be an ecosystem of commercialization coming from it really fast because of the edge and on-premise needs. It'll run cloud operations but everything will change in the tech stack to support the interface. And that's what's coming out of some of the conversations with Dell and Broadcom and others. And dynamic infrastructure, fabrics, memory disaggregation, these are like nerd concepts that are gonna be scaling out. So it's a whole new ball game, distributed computing. That's the ultimate nerd fest, right? God, it was incredible. It was very strong. Hardware matters, AI accelerators and chips and innovation, you're gonna have AI, silicon platforms, silicon diversity, open connectivity, things like how the APIs connect in the data in and out, how fast does it move? How fast is the chips? Those kinds of thing is gonna be the real conversation. And then the cloud guys are gonna have to just make their stacks very strong and fast. Well, I mean, the action's in the cloud right now, but then you got IBM and Oracle and then you got every ISV in the world, ISV slash data platform slash, the Mongos, the Salesforce, the Snowflakes, the Databricks, into its name a software company, you can't name one that's not injecting AI into its platform. And then open AI just a couple of weeks ago or when it was a last week just made it much easier to do so. I think, I mean, Microsoft's all in on AI, right? So what'll be very interesting to find out from Amazon on this trip is that, and then reinvent is how all in are they, right? So, and are they even set up for it? They're gonna say, yes, of course, but back to my point about being traditional workloads. When things, if the new AI infrastructure emerges, and we said this, remember on two pods ago we brought the concept of, what if there's like a new kind of Linux model for AI or a new kind of AI system, a neural network, some AI infrastructure, call it new, some new thing that is architected with the piece parts of the traditional enterprise computing. What would it look like? I would have to support multiple datasets, real time, interaction and do the generative thing that we're seeing. And I think that's gonna be the question. Is there a new way and who moves over? Clearly Microsoft's clear in their event that they're going to the new way and they want to make everything, and then hide in the ball. Sathya Natal is saying, developer environment gain has changed forever. Direct quote. So, I mean, let's see, let's talk about how Amazon and the cloud changed the world. See, I would say it changed IT, right? There was initially a lot of lift and shift. There certainly was, there were new Greenfield applications, cloud native applications. It definitely changed, I mean, it completely changed IT. But how much did it change society? I mean, it definitely contributed. Things went faster, it was cheaper to do computing. So that obviously had an effect in society. You saw governments, you know, sort of push toward the cloud. But when you think about the impact. I mean, I think they have, I mean, if you look at the acceleration of SaaS apps, so if you consider the mobile generation, again, this is, you and I discussed this, I'm on, I think I'm back, I don't know if I was against it enough, but I think mobile was an inflection point because of the form of change, right? So, I think it turned out. And definitely data, but it was all, it was kind of under the covers. Does Airbnb, is Airbnb a society change? Yeah, and Uber. But would that happen without the cloud? Probably not. Probably not. Twitter wouldn't have been, Twitter never would have happened. Twitter started in the cloud. Social media definitely changed. Okay, so I'd buy that, but it was really felt, I mean, yes, it felt that through society, but it really felt it was because of the IT enablement. IT was a blocker and Amazon removed that block. Yes. With AI, what we're seeing with all these co-pilots is the system taking action for the users, the system of agency. And that's different. I mean, I, I mean- George Gilbert, and you talk a lot about system of record, system of intelligence. That stuff's relevant now more than ever because if you have data, and the killer app is personalization and real-time generative experiences based upon data and iterating through it, it's gonna be incredible. So I think, imagine having an interface. I was talking to the CEO, Jonathan Ross from GROC, who has that killer inference chip called the LPU, Language Processing Unit. The speed is incredible. We were talking about how, if you interact with a model after the third time, what if it made you better so that you were actually more creative in the third second in? The faster you can iterate through these initial prompts will change the learning trajectory or the productivity trajectory. And that's what he was pointing out. I'm like, wow, that's actually pretty right because think about that. What if it was so fast and so good where after two or three prompts, you were actually in a position where you felt comfortable and confident to actually do something? Whether that's a learning task, well, professional task versus being indifferent. Well, I'm not sure I can do that. That's the kind of creativity productivity that will rock with this kind of fast inference. That's kind of how chatGPT works, right? I mean, chatGPT gives you- It's still slow though, I mean, think speed. Okay, but I'm just saying, you prompt it and it's like meh, and then you prompt it again and like, oh, that's a little bit better. And you prompt it again, and then you're like, you'll pick out some ideas in there and say, okay, boom. And then that makes you better. Yeah, exactly my point, but it's minutes in now. Imagine making that seconds. Yeah, yeah, yes, yes, doing that in real time. Yes, that's I think that's why I like chatGPT because that's the format change. You know, yes, I had a guy in today who wrote this book, a restaurant in Jaffa, Mark Sorensen. He's an old friend, old EMC guy. And he wrote this book. First book he ever wrote, the thing is unfriking believable. It's a mix of like geopolitics with cybersecurity, with technology, because he's an engineer. So he knows, I mean, he was talking, you know, C++ and talking assembler and he's talking NOOP, you know, all this sort of cool stuff, PDP-11s. And so we took all through history and critical infrastructure and how exposed this. Anyway, he said it took him five years to write this book. Right? And he just kind of wrote half, you know, a couple of pages a day or not even page a day. And then get back, he said 12 iterations. So do you think that, you know, there's no doubt, right? We're going to be compressing the time it takes to create art, music, literature, plays, movies, scripts, right? Research notes, white papers. Exactly, that's just the case. This is why I think the conversation around training and inference is going to get better. And I think I'm now convinced. And Luis says from the CUBE alumni told me he was like, everyone's going to go gaga over training, but it's going to be inference, right? So that is going to be the big thing. And I think what's clearly coming out of this is that the training is going to be the big language models. Companies will have some sort of language model, some size, but that's like setting the setup. You train it and you kind of train it over time, but that's going to be the big upfront investment. The inference, Luis says it's from AcoML, by the way, folks listening. His own business is built on inference. Inference is the value extraction, Dave. And that's what the GROC CEO and I were talking about, that you get value as you iterate through the data. So inference, as you get bigger and better with AI, inference gets better. And then so we had nerd conversations on the CUBE that we're talking about, what's the infrastructure look like for that? And so training is like a big setup cost. And then the scale out servers, the server clusters are all inference servers. So imagine having large language models parked at the edge of every device. And the inference happens at the edge and there's not a lot of data moving around. So you basically co-locate LLMs in areas you need them or foundation. I'm looking at a piece that Floyer and I wrote in April of 2021, breaking analysis, Moore's law is not dead and AI is ready to explode. And the point is Moore's law, like we know it is dead, but it's actually accelerating because of AI. And that's exactly what happened. And then there's this chart that Floyer and I did and we were right, but we were wrong. It says, as AI matures, inference will dominate. And then we had the percent of spending today and then at 2030. And it just flips from 90, 10 to 1090 where all the actions in inference. And so I think the difference is we had it at 2030 and it's probably going to be more like 2026, 2025. Probably five years accelerated than what we thought because that's where all the action is going to be and these domain specific models, you know the power law of AI. Did you see any evidence that a couple of things that a lot of this stuff is going to happen on prem? You know, we're not seeing a lot of spend yet. We're seeing downloads of llama two on prem but not seeing any clear evidence yet. Did you see that at last week at super compute of that? And the other part B is, did you see any camel humps along the long tail? Remember the conversation we had with Intel? No, I think that there was not a lot of data to observe on that other than hallway conversations. The thing that super computing that was most relevant was thinking about how the non-cloud operators are going to be building up their infrastructure. In other words, the old data centers and the cloud service providers, the hosting, hosters from the old days, people with networking backgrounds, they're doing great. I mean, we had one Vulcan was a company, amazing company. They're so successful. They're one of Nvidia's biggest customers. They're cloud service, right? They specialize on GPU clouds. They're killing it. Their developer experience is great. The products good, it's less expensive. They're essentially an alternative to AWS. And that's what they say they are and they are good at it and they love it. And they were loud and proud about that. And so this whole show is about infrastructure scale, not so much AI models, but what they were all doing was saying that those models will be foundation models will be an abstraction layer feeding the interface. So they're under that cover, right? So it's like, you know, interface, abstraction layer of models and then infrastructure. Their conversations were like, how do I take a bare metal set of servers and create scale in a new way that solves the AI complexity? And what's coming out of that is composability of these fabrics, desegregating memories. You have memory pools. So they're essentially re-architecting the building blocks of traditional data center kind of thinking servers, networks, storage, glue layers of middleware. They're taking that construct and the chips that go in it and re-factoring it to handle large scale. They're basically rebuilding clouds that aren't clouds. They're like the edge or like a data center. There must have too been a lot of talk about just sustainability, energy consumption, liquid cooling, right? That had to be a big topic of conversation. One of the things Microsoft showed and they were making a big deal out of it is that, you know, this is a liquid cooled chip and we've set it in a traditional rack so that we can retrofit existing data centers and they were making a big deal out of that. I mean, everybody's going to do that. IBM is going to do it, you know, all the big computing manufacturers are going to do that. That's not like radical, but it is in the sense that it's going to allow these liquid cooled processors to be injected into existing data centers so they don't have to be completely retrofitted. So there's a nice bridge to accelerate AI injection because the demand is so high. I mean, I think, I mean, no one's talking about the sustainability problem, but it's a problem. I think that's going to be part of the results here. There was some really good talk about sustainability, putting data centers in different parts of the world. They think latency is not going to be that big of a problem. They rather focus on these more systems because if you take the extension of the cloud concept and saying, what am I going to do on the edge? Or as satellites coming, respecting Amazon, I'll sort of tweet from Jasmine today about satellites because it was started as a project at Amazon working backwards paper that someone wrote. It's now a product, it's about satellites. So the hyper edge or far edge conversation is going to be key. And it's a cultural shift that's happened. That's why the HPC event was big because it wasn't the yesterday's mindset, but they all think in scale and exoflops, right? So they're like, they get the scale, high performance computing, it's usually like the highest highest and supercomputing. That's going to be stable state stage. So the cultural shift is everyone wants a supercomputer. Basically you need that and you can get it. And so that's going to be the chip war. Okay, so JNAI is part of everything. The chip's got to get better and the cloud's got to move up the stack. And you got to have an ecosystem. You got to have the security and privacy. The dev experience has got to be modern. The edge is the opportunity. I mean, it's complex right now. It's very, very weird in the sense of like there's so much going on for the cloud. A couple of things. So Microsoft and NVIDIA, Jensen was on stage. They like within three months we put together in the cloud using NVIDIA capabilities and Azure. We put the world's fastest AI supercomputer and the number three supercomputer on the top whatever 100 list. So you know how those things are constantly changing. I don't know if that got any job. But they talked about energy. He said, in fact, in fact today we're one of the largest buyers of renewable energy around the globe. This is Satya. We have sourced over 19 gigawatts of renewable energy since 2013. I mean, kind of a hero metric by me just to put that in perspective it's the equivalent of the annual production of 10 Hoover dams. And we're working with producers to bring new energy from wind, solar, geothermal and nuclear fusion as well. And then he said, I'm really excited to share that we're on track to meet our target of generating 100% of the energy we use in our data centers from zero carbon sources by 2025. Now, I'm not really an ESG expert like Rob Stretzsche knows it well like Carolina Melanase is all over this. But that's like not that far away. Now, maybe that's not a big deal because you can just choose where you're sourcing. I don't know if that's like phase three. That's not, but phase three objective. The power is a huge problem. Power and energy. But yeah, these data center operators, I mean, they have no choice but to attack this problem. I mean, I'm gonna ask Amazon folks and some questions around when I say with Adam and around the cultural shift around how data is gonna work. Because remember, the Amazon's got a lot of data and they have sort of microservice like kind of all kinds of services. I wanna see if they're well positioned for it. But you asked me, what was my big takeaway from HPC? And I would say that it's not just HPC, it's HPC and all KubeCon, all the other events and our conversations that we are having here on the KubePod, and let's getting feedback from our audience, which by the way, thank you very much for the DMs and extra stuff is it's not about just the chips, Dave. And so like to use the analogy, the chips are like the heart, right? It pumps things. And networking's like the bloodstream, right? If you gotta get networking right, you gotta get the chips right, you gotta get everything right in an operating system. So if the heart's gotta be pumping, the blood's gotta flow through the veins, that's networking. So why I like this Dell-Broadcom relationship that we were talking about on the Kube, is Broadcom is working on the networking stuff. And we just talked about Infiniband earlier. So I think it's gonna be a combination of chips and networking. And if you look at networking, Amazon has an advantage with their cloud. So I'd be very curious to find out from Adam does he wanna optimize over the chips or the networking and getting it right, that balance. And if you get it wrong, you could be on the wrong side of history here. So... He's gonna say we're investing in both. We'll see. I'm gonna ask him that question. I mean, you know, you're a computer science guy. It's always about balance, right? And if you over rotate on one component of the system, you create imbalance. So it's always that struggle to just, you know, not have one part of the system create other bottlenecks that screw up the entire system, right? All right, well, Dave, I know it's late there and I had to get into Seattle. Yeah, I gotta go, I gotta do my breaking analysis. And the guys are like, oh, really? I appreciate it, I appreciate it, get it in. Well, I guess we'll have to skip the rant section. I guess my rant would be, we need more time to prep for re-invent. Miley, I was listening to Bloomberg today and I saw they were talking about the Microsoft chips and the Bloomberg hosts, she made a comment. She threw out an arm and she goes, yeah, these are arm based and I know arms behind technologically. And I was like, what? Yeah. No, they're not. You think Apple's behind technologically? Really? Oh yeah, Apple is also behind on AI, too, right? Yeah, well, all right, John, cool, I gotta go. All right, let's wrap, check us out. We're gonna be at re-invent. We're gonna have SuperCloud 5 special going on on the week of AWS's annual user commerce re-invent. Dave and I will be there with our team getting editorial content at re-invent. Savannah Peterson will be in studio with our team where people will come in in Silicon Valley where we're gonna have a live show called Battle for AI Supremacy, SuperCloud 5, Special Edition. We're gonna have so much content. We're gonna have people weighing in from our expert network, CUBE alumnus, friends, all focused on this evolution of the cloud with AI and how that's gonna change everything. And we're gonna compare the clouds, who's winning and who's not winning or who's it slower. And we're gonna break it all down with all the people in our community we're gonna unpack all the core issues. Of course, siliconangle.com is where all the content lands and all the video replays are on thecube.net. And the CUBE AI is getting better every day, loving that site and that might just be our future site, Dave. The CUBE AI, thecubeai.com, it's out of private beta. Anybody can access it, go check it out. And yeah, I'd love your feedback. And then don't forget Rob, Stretch A and Rebecca Knight are gonna be in Barcelona at HPE Discover. Yep, great stuff. Big European show. So, and I think our East Coast crew is high-fiving because they love going to Barcelona more than they like going to Las Vegas. I wonder why. All right, that's a wrap. All right, thanks you guys. We'll talk to you later, we'll see you at re-invent. All right.