 Welcome back everyone to SuperCloud 3, live here in Palo Alto. I'm John Furrier, Dave Vellante. This is theCUBE's SuperCloud 3, our third episode of our quarterly series of SuperCloud. Soon we'll have a physical event bringing everything together, hopefully in 2024. Of course, SuperCloud 4 is coming up in October. It's going to be all about AI and we have here our distinguished CUBE collective analyst, quasi-analyst on theCUBE, Howie Xu. Only kidding, he's really the SVP of engineering at Palo Alto Networks with AI and ML. Howie, you're practically an analyst here on theCUBE so many times. We love having you contribute your knowledge to our community, I want to say thank you and thanks for coming on SuperCloud 3. Very glad to be here. Hello everyone. Hello Dave, John. Great to see you. Yeah, and we've, the SuperCloud wave is going. You were there from the first one, two and three. So let's get into it. SuperCloud 3, are there any revisions to our earlier statements around what's changed because a lot's changed. Lama 2's out. So you have my security now as a co-pilot today with Microsoft, yesterday with Microsoft. The developer community in open source is booming since we last talked. It continues to rise. Last time we talked there was January and I actually mentioned the English language is going to be the programming language and then Andre Capati actually tweeted a couple of months later. So now it's actually understood by everyone. You got your, you're on the wave and it's great to have you here. Let's unpack the security AI story because if you look at the impact of AI mainly around these operational aspects. So security is very operational. The pace of play is high, defense has got to be the priority, attackers are coming in. AI now, you're seeing the augmentations with a big theme of SuperCloud 1 and 2 is that data driving everything. So with all this new data coming in, what are you seeing that's changing quickly in the architecture of this distributed computing next gen cloud, SuperCloud? Yeah, I think security just like any other industry is going to be data and the AI industry. Data is very important, as you said. I think another thing that's, I wanna say unique but for security industry, it's roughly $200 billion industry, half service and a half product, half service that means a lot of the work, human professional work and then there is a lot of the high end and then there is a lot of the repetitive work. I think AI is here to automate a lot of that and then there's always a number in the last about five or 10 years people throughout in the last five or 10 years which is always at a given time there are about four million or five million SOC or security professional openings unfilled. So that leaves a lot of the room for AI to come in. People talk about the job replacement or whatnot. I would say even before that, how to fill the void of that four million, five million, unfilled the jobs. So I think there's a lot of the human repetitive work. There's a huge potential to remove that. Dave, I loved what you said when ChatGPT first came out which you call AI heard around the world, multiple references to that because it did wake up the mainstream to see that kind of so-called magic. For insiders, we kind of knew this was coming around but I loved your take on ChatGPTs. It makes a smart person smarter and not so much smart person smart. And I think one of the trends we're hearing in SuperCloud Howie is that the creative intellectual capital of the human is augmented significantly with AI. So whether you're a security person or a data scientist or a developer or just an IT person, having the institutional knowledge of your job that intellectual capital is scaling with AI. This is very much not talked about much. It's mostly gloom and doom. Oh my God, AI is going to kill jobs. But in fact, SuperCloud is the next frontier of IT. It's going to actually, it should help people. It should be an intellectual scale. If they lean in, right? Yeah, I think the kind of the example I've been using in the last six months or so is that, imagine you suddenly are able to hire five very smart interns, right? They don't necessarily know a lot of domain knowledge but they are very smart interns. You can teach them to do things and then suddenly, you only need to pay those interns $20 a month. And what are you going to do? The funny thing is for most people in the industry, they don't know how to deal with it, right? If I give you five interns, are you really going to leverage them effectively? I think in the next year or two, we are going to either learn it or be forced to learn it or I think we are on that trajectory. So that's one. The other analogy I will always use is it's almost like people say, hey, what is the kind of the jobs going away or whatnot? The way I look at it is like a programmer, for instance. You always have lots of programmers. You always have tech lead, right? Managers, it's almost like in about three, five years, programmers are all upgraded, promoted to be tech leads and then managers because a lot of the lower level things will be done by not today yet, right? The co-pilot is good but now can do all of the coding. But I think in three or five years, a lot of the coding, it can be done by the co-pilots but you still need to understand architecture. You still need to understand from the business to the tech stack, right? So that job is not going away. And I think- You mean the systems thinking kind of the architectural? System thinking, architecture, right? Future proof, you know- That's not going away. That's not going away, right? Maybe 20, 30 years, but in the foreseeable future, I don't see that going away. I do see, and if you look at a code, GitHub co-pilot, what I see is it can do auto-completion of a functioning but it cannot do auto-completion of a file. I think we will see auto-completion of a file but a file is still not the entire software stack, right? Someone has to do that, yes. And I heard Walter Isaacson the other day pumping in his book for a musk. And he was saying, you know, all these jobs, everybody, you know, he was forever, machines have replaced humans in menial tasks and manual tasks and, you know, the blue collar, white collar people never complained and now it's changing, right? We're seeing the impact on white collar jobs and you're hearing all the complaints of writers and so forth. But to your point, it's going to create new opportunities that never existed. Now John, the other thing is we'd love to talk about competition on theCUBE. And you called the chat GPT your Netscape moment when you first saw the Netscape browser. And then Howie, what you said at the time in January was, hey, a hundred million get you into the game of building LLMs. And I think the numbers, you're proven right because the numbers just coming down further. And that's why we said, we didn't think that open AI was going to be able to maintain its first mover advantage. Now having, since then, open AI's- And that was what you said. That's what I said. You would disagree. Open AI's by far, you know, on people's mindshares, number one, Microsoft's right there, you know, AWS has dropped down a little bit, you know, even though everybody's been rising, but what's your current thinking? You're not walking that back. If anything, you're doubling down on it. What's your current thinking on what it takes to get into the game? The foundation model, right? In January, I said a hundred million would it be the ballpark. I stick to that. I think, you know, over the last few months you saw open AI raised a huge round, right? You know, and then they didn't say that explicitly, but people are saying that, oh, in the future, you know, open AI is going to train the model with a billion dollar budget, right? You know, I think people are saying that. So a hundred million dollars seem to be wrong. But on the other hand, you saw, you know, just a llama to release the yesterday, right? The, you know, the consensus, right? Of course they didn't tell people, you know, the consensus from the community is it's roughly 20 million dollar training. So I would say, you know, a billion dollar, yeah, you know, maybe, you know, GPT-5 would take a few hundred million dollars, but I think in the long run, it's not going to be. No, but I think you were right about that because your point was it's a lot less than people think. And so that was really your main point. You threw a number out a hundred million. It's actually now. And then your question was, does open AI become a first mover advantage? Right. And we said yes. And my main premise was scale. You said yes. I said yes. And my rationale was in light of even the cost side was the first mover had scale. If they get to scale, they get to skate velocity, but I'd also said open source would win. So at that time, we tell of the iPhone Android moment. Good analyst does. Hedges as bets. No, no, no, but it's a long to power law. So you have the primary model at the head of the long tail, but the mid tail and tail of open source, you get smaller models are emerging. Collectively as large. We had also said that it's not one model rules the world. It's not going to be one model. Yes. It's a power law. And you're going to see some proprietary big models up top that are hard to replicate with scale, but those trade offs, they have public information hallucinations. And then you're going to see a fat torso develop and then a long tail and open source and proprietary domain model. So the other thing that we debate, which I'd love to get how he starts on or is this a trend that is going to favor incumbents or disruptors? What do you think? Yeah. So I think that the analogy I would use, look at the cloud transformation, right? What happened the way is a lot of the traditional incumbent enterprise software didn't make it, either didn't make it or it didn't thrive during this cloud transformation. But a few small number of them actually made it and it thrived. That's the Microsoft Adobe of the world. So in terms of the percentage, in terms of the quantity of the incumbents that made it, cloud transformation, it's small. But the net impact of that small number of the incumbents that made it, like if you think about Microsoft Adobe, the value creation in terms of the market cap, in terms of the business value aggregate together, I would say they are probably combined is actually more than the rest of the, I don't know, five or 10 or even potentially 20 SaaS company combined, which is kind of in some ways I'm saying that small number of incumbents that are able to make it, the impact that they are going to make is just proportional. Because of their size. Because of the data, because of the size, the customer reach, we all know for enterprise software at least, right? Customer reach is a big deal. And so doesn't that make it more like the internet, which I would argue favored incumbents more than disruptors because of that advantage? We've debated this on the podcast. What do you think? Look, there are, again, I stick to what I just said, right? Small number of incumbents that are going to make it, in terms of the total value creation, you know, they are actually going to have a data advantage. But if you look at the total number of companies, the top 20, top 30 interesting players five, 10 years from now, I would say a good number of them are the startups today. I know one thing for sure. Everybody else is going to get disrupted, not withstanding the technology business, but all of us. Howie, thanks for making my point, by the way. I just won the podcast. We'll hit on podcast Friday. We'll see. We'll see. No, no, but the thesis went up, but he's right. In most inflection, and this is, again, the power dynamic, what he's mentioning, in all these inflection points, where you have clear visibility on at least a 10 to 20, 30-year stare of the future, and everyone agrees that this is happening, the abstraction gets built, and the simplification of the old world gets done, and incumbents don't thrive in that environment. They either get replaced, they become extinct, or transform, and the extent that they transform, they have leverage if they do it right. So, and we think, unlike the mainframe to Mini, and Mini to LAN, and PC, PC LAN, is that they're more agile now to adapt. So I expect Microsoft, they did that. Look at them. Look at Microsoft right now versus, say, AWS. We just heard Matt Garmin. They're in a great position, too. Amazon's not losing, but my condo's winning the PR game. But you guys are contributing yourselves. I beg to differ. So they are losing, in my opinion, at least for small sample of the customers I talk to, I'm aware of, they were or they are AWS shops, but now they are thinking about the Azure, right? I think, I don't think we can underestimate that trend. Yeah, and the $30 subscription is only going to accelerate Azure services. Okay, but there's an example of an incumbent. Okay, so I'm saying the incumbents are going to win, but you're saying AWS is ripe for disruption. Well, they are being, his point is right. They are being disrupted, but they still not the second cloud. They're still number one. They're still winning. It's just for the first time in their history, when, if you went back two years ago, and pulled an Amazon customer and said, hey, would you like to go to Azure? We'll pay you to move to Azure. They weren't moving, even if they were paid to. The incentives, and today that's changed. And his point is right on, and Microsoft is starting to chip away at the loyal Amazon customers to move over. By the way, go back six months ago, when China GPD first came out, all the first week, all of the energies were focusing on, it's going to replace the search engine. I was one of the few people actually stood out and say, no, the biggest change is actually the impact is on cloud. I think six months later, I mean, we're still talking about disrupting search, but it's going to be a little bit further away. I think the enterprise software is really proven to be, where most of the energy is. And we were with you on that, and we were first, and you were right, and that's not going to stop, by the way. I think you're going to see the gloves come off from Amazon. And by the way, the market of builders versus hosting is an interesting paradigm. Because right now, if you look at the adoption of, say AI, are we in a builder mode or are we in a hosting mode? So as things start going, where's that, I say hosting, you know, generically, meaning moving to the cloud. Obviously you host AI on the cloud or on Brem. When you build an app, you've got to host it somewhere. Yeah, the other thing, you know, a lot of people, I saw a lot of people talking about today, you know, hey, six months later after China GPD was out, how come there are not so many killer applications? Right? And I don't know about the consumer side, but at least for the business side, I disagree, because you know, I think there will be a lot, but enterprise software usually takes time to build, right? You know, it's not just split second, you have a next generation, it takes some time, you know, for the API to mature, for the software to sort of adapt to this new kind of the large language model being the platform, right, re-platform it. I think in six, 12, 18 months, there will be a lot of the killer applications entering in three, five years. You know, everyone will have, you know, every enterprise will have lots of the co-pilot-ish. Chat GPD is a killer application. But that's a good point, that's the thing on this one. People are saying that other than Chat GPD, what else? Well, hold on, this is a great point, because you can't just put a wrapper around Chat GPD, those are fake apps and so forth. Exactly, with no differentiation. No, but he's right, this is a really big point, because what is the gestation period of these apps? And you know, the conversation that's having, we should be having is, what's the acceleration compression timetable? Because one of the premises is that AI accelerates value, so the time to value or product market fit can be accelerating. So if that's now available, I think you're going to see ventures pop up fast and hit escape velocity. What's your take on that? How fast does AI change the game on product market fit and time to value? Yeah, I kind of said it. If you look at a cloud, right, you know, AWS came out, EC2, 2007-ish, right, it probably took at least five years before most people even understood it, right. I think at this time it will take a much shorter time. But it's not going to be five months, right. That's why six months later, you don't see co-pollets left and right. There are a lot of announcements, but that's a different story. The real products, it takes time too. I think the digestion of this new platform, it's, you are talking about year or two, right. Then there will be playbook, there will be best practice. And in three, five years it should be every year. I think it's the tsunami of apps. We heard yesterday about the, someone was talking about an example of software for house managing housing. Outside of managing one or two units to 300 is a gap. The cost to build the software is going to get lower. Therefore, these markets that were unreachable before are going to be open. That's why I do think, I do agree with the disruption. SAS gets disrupted in my view. Cloud, okay, so cloud and SAS. So it's going to be really interesting to see 10 years down who makes it through. AWS, potential disruption, Salesforce, service now. These guys going to be stronger or weaker 10 years from now? I think stronger. I mean, Howie makes a great point. And I keep coming back to the power law. If you look at, Matt Farman just talked about it, about proprietary models and you got third party models, then open source has got a slew of long tail, unique things coming out that might look small on paper, but have evergreen and unique domain expertise. And collectively they're large. Like audiences and the power law will stretch out and the head and the neck and the torso will be much fatter and then the long tail. So if that happens, then apps will take the same trajectory in my opinion. And the economics are irrelevant based upon the cost to build them and the available markets that are out there. So I think- I would even stretch this a little bit further. We have been talking about big company incumbents, right? I think that there is also another thing I'm seeing that is to start a company, to do a project, to turn out a product that people, lots of people would use, you probably don't need an army of people anymore, right? Think about a mid-journey, 11 people, right? Got entire world sort of around it. And I think talking about the programmer's job evolution I was referring to just now, I think what's going to happen is in the past, if I'm a product manager, probably I need to hire 10 people in order for me to do anything. Today, I can just leverage AI technology, right? And then get going without the 10 people. So you were asking me right before the show, we were chatting, hey, in the Starbland landscape, what's happening? And I felt like entrepreneurship will evolve a little bit towards the solopreneur, right? We'll see a lot more solopreneur than the traditional version of the entrepreneur. Because they got a smaller team. Because they have co-pilots, they have AI, their tech stack is not the same as before. And they can get a lot done. It's almost abstracted away. The tech stack could be abstracted away. Essentially, your CTO could be a co-pilot. And you can talk to it. And we're talking about super cloud, right? The definition of cloud is evolving. Definition of cloud in the past is compute storage. Now it's the large language model abstraction, right? And scale and data play a huge role in that. Data will be a huge thing. Okay, so final question before we wrap up here. The role of data has come out, if you believe the long tail of open source models, foundation models coming out, the role of data becomes the new IP. So intellectual knowledge, human knowledge, we check that it's IP. Data is valuable. Your thoughts on data as intellectual property? Yeah, I talk about it from a few different dimensions. One, we were talking about foundation model. People say, well, all the internet data has been already ingested, there's no room to grow, right? I actually disagree. One is there is, especially in the enterprise software, there's so much proprietary data that has yet to be understood, ingested to those foundation models yet, right? So that's one area. The other thing, I actually agree with Elon Musk last Friday, he was talking about, hey, the future, a lot of the data, the model will generate, they're just like AlphaGo, AlphaZero, AlphaGo sort of the be the world champion four by one, but AlphaZero is like, hands down why? Because it generates a lot of data automatically. So that's that. So let's just focus. The last thing I wanted to just focus on the enterprise side, right? Enterprise software. All the big companies, right? Have advantage by definition. That's a good story. The not so good story is, you still need to do a lot of work to understand what sort of data you have. What's the gap, right? Privacy issue, right? Do you really capture, did you capture 100% data you need to do the value or you only capture 60%? Do you have a plan, do you have an understanding? I think that will differentiate, especially for the incumbents, if you are kind of a very keen on those kind of things, understand the gap or understand the type of work, then you can potentially thrive. Otherwise, you have a lip service, yeah. We are the incumbents, we have the data, we have all the data, we have more data. But when you double click, triple click, that's not a case. I'll give you just one quick example. Sarah Gore, the ex general partner of Greylock, my good friend, I invited her to Palo Alto Network leadership. I did a five-sided chat, right? I asked her about the data side, and then she said she's backing up one of the co-pilot, the Github co-pilot architect, right? That person left. I'm not going to talk about what exactly he's doing, but his notion is, hey, I'm going to do a vertical area, and I know those company have all the data, so he went to those companies and then he found out. In order for me to go to co-pilot for the vertical areas, a lot of the data he needs is not there because they didn't care to collect those data because they didn't curate those data. So he now has to think, I wouldn't think from scratch, but think much harder than he originally thought, at least. So the point I'm trying to make is, it's not just the volume, you have it, you need to double click, triple click, and I would say that's quality data, quality, quality terms, the quantity. And then there's a paper, you were talking about the open source model, there was a paper just two months ago or so, less is more, you can use a thousand, high quality, fine-tuned data, Q&A, to get a very pretty high quality, sort of the llama-based, fine-tuned model, it's only a thousand, you don't need a gazillion. And that's very, very telling, quality, it's the quality, not the quantity. Yeah, and the startup community is booming, great insight. Howie, thanks so much for being part of our CUBE community and collective, your contributions are awesome. Love hearing the masterclass and what you're thinking. Love your blog posts, you post a lot on LinkedIn, check out Howie Shoe on LinkedIn, he's got a lot of great content. Of course, it's also the SVP of engineering for AI and ML at Palo Alto Networks. Howie, great to see you and thanks for coming on SuperCloud. Thank you. All right, I'm Jeffery David Lother at Howie Shoe here for the CUBE conversation around AI and SuperCloud 3, laying it all out. And don't forget, SuperCloud 4 is an October market calendar, so that's going to be all about AI. So we're going to go deep. We'll be right back with Silver Cup 3 after this short break.