 Welcome back to theCUBE's coverage here. On location, Adib is re-invent Amazon Web Services annual user conference. This is theCUBE. I'm John Furrier, your host with Dave Vellante, my co-host. Our 13 years covering Adib is our 11th year covering Amazon. We've been here every year, 11 years at re-invent. Dave has been quite the journey. In fact, I'll never forget 2013 was our first year. We were just talking about it. And Jerry Chen was walking through and he said, back then we didn't have a schedule. He was like, hey, you can free. Come on up. Oh, this is James Hamilton. Come on up. He runs the other way. Andy Jackson, I don't want to have another cube. But we got everybody else. Jerry Chen just left VMware. The newly-invented partner at Greylock. Back again. He's been on every single year on theCUBE. Thanks for having me, Nan. Great to see you. Nice to see you, man. Legend. Now you're rich. You're a big BC. You're a fat cat. What's going on? How do you see it? I don't think of that fat yet. Trying to be the cool cat. It's amazing, right? 2013, 10 years running. 11, 12 years guys been doing this. It changed from like basically one ballroom. We were all in 11, 12 years ago to this thing up and down, the strip. A lot of goodies that year. And you were just saying that it's both, it's tough to do to have a lot of respect for what they've done is both doing incremental and improvements as well like big releases, right? And so how do you like give a little catnip for all the audiences? Yeah, yeah. So, I mean, 80 billion, you know? It's a real business. I had 10%. What was your line? Win or take all? Or win or take most? Win or take most, yeah, yeah. If I remember correctly? I said, I don't know. It's win or take all. I think it's win or take most. And it's interesting because over the past 10 years, Google and Azure clearly have made their inroads to different various levels of success. AI, as we talked about, has turned the tables completely. And I think you're seeing two or three different cloud paradigms. You have kind of an edge cloud. Guys like Cloudflare, you have classic clouds. Like Amazon did in early days. And these AI clouds, right? And so AI clouds now are all about GPUs, InfiniBand, low power, whereas like edge clouds are close to the user and the classic clouds is Amazon. Now clearly, Amazon, Azure, and Google also want to own the AI clouds. And so I think it's a race now which company can add AI cloud portfolio and win in that market. And this is potentially the opening for Google. It hasn't shown up in the numbers yet. We really haven't seen it. Well, the arms dealer in all this is NVIDIA and it's interesting to see Jensen on stage because DGX Cloud was not, Amazon was not included. I think they passed on that. Okay, from what I've been reporting and it's not been officially reported but it's kind of what I'm hearing. And then they've enabled CoreWeave and other companies to create these sub clouds or other castles, if you will, in the clouds. And that's just opened up the HPC market who know how to put stuff together with bare metal and data centers. So you're starting to see purpose built large scale GPU clusters. And I pressed Adam on that. I think to your point around data centers you might see not necessarily a pendulum swing back but more inertia to private clouds and private data centers for security reasons and privacy reasons, right? So there was this inevitable shift towards public cloud but now large enterprises want to train their data. They want private data. And so increasingly we're seeing more of our customers in our portfolio want to run open source models in their data center or in the private cloud. So cloud's still the inevitable wave but I see you're going to see this new trend of new stacks of GPUs and supercomputers inside your data centers. So is there an opportunity for startups to go after that or is that going to be like a Dell or an HPE that does that? What's the opportunity? I think, you know, I was invidious to stock we all should have bought for John's comment. I think there'll be opportunity for startups not interested to build systems because systems are hard to make margin but software, right? So think about all the software you need to train, fine tune, run the models and build these applications. How do you replicate the stack that Betterock has or Azure has or Google has in your VPC in your data center? So like Lama index, one of our seed investments it's all about RAG, Retrieval Augmented Generation how to build these data pipelines of your enterprise data with these new models inside a public cloud or private cloud. So I think you're going to see new software stacks be created. And by the way, congratulations. Great investment, Jerry is on the other Jerry from Lama index is on theCUBE in our studio with Savannah and team and Howie. So this is again, one of those areas where is it a white space that's going to get rolled over or is it actually beach head for an opportunity to get in the game and iterate quickly? And is it an iterate quickly like the web was and get out front and get fast? Yeah, when you see open AI do like LLM or RAG out of the box, what does that say? Well, I think it's the right question, Dave. Is the LLMs, the foundation models, is that a winner take all market or winner take most market? Going back to the question we had 2013. If it's just one giant model all open AI or all barter, whatever from Google then the startup ecosystem probably is decimated. But you think there's a long, fat tail of open source models, other foundation models not just open source but like we're investors in inflection, we're investors in adept, there's ethnographic, there's coherent. If there's a multiple models out there then companies like Lama index have a place to exist, right? So again, if it's a winner take all, maybe not, if it's a winner take some or a winner take most then there's a vibrant ecosystem for Lama index to see. And that's your bet. Winner take, winner take. Winner take a lot. I think one, two, so let's keep proving our model and your thesis around especially in models the long tail power law. Two, but Lama index proves to me as well some of the RAG stuff is that there's a developer appetite and that keeping that in the local host which could be on premise in the data center. Maybe the data center is the new local host. I mean, but if you think about the data as the IP because you're testing and iterating, what's the observability? How do you manage the data? What's the memory recall on the retrieval? I mean, this is new data, net new. Net new. Net new, so it's not yet been tackled by off the shelf observability. Is net new stacks of net new how I use data? And David and I were talking about these applications how you do software development with AI is non-deterministic. You get three prompts or the same prompt to three different models, you get three different answers, right? So all of a sudden, you know- There's no memory. When we learned a code, it was deterministic. Software worked or not. Now with these LLMs, you get the same prompt three models, you're not exactly sure of the response every single time. So software development is going to change. You know, I was talking to this professor, CS of Stanford and they're like, yeah, intro CS is going to change the next five or 10 years. And so, I think about building software. It might not even exist. Right. Yeah, what do you call software development? So I think, I'm excited about Lamanix. I'm excited by a bunch of other stuff announced today from storage to cloud to whatever. But I do think, obviously AI is super exciting. It changes the cloud players. It changes the infrastructure players like database guys or Lamanegos and Ragspace. It changes the application. So I'm super bullish about the startup market. In 2013, we talked about, we asked you, do you think Amazon will go up the stack and start developing applications? You said, no, their strategy is really going to be to enable developers to build those apps to compete against their likes of Microsoft. So we were talking the other day and I think in the Q pod, I said, was that a failed strategy? And you said, was Netflix a failed strategy? So that was a good point. But looking back, I mean, Microsoft is in such a strong position right now. They've got the full stack from Silicon all the way up to the apps. So how do you see Amazon's position now in that context? Yeah, it's interesting, right? I mean, in argue that if you don't have a running game, you need a passing game to use sports analogy, right? And so to the extent that Azure's not going to recreate all the basic building blocks horizontally as dominant Amazon, you know, play to your strengths. So if you're a Microsoft, you have the full stack from Silicon onto the apps. So why not leverage your apps, right? That totally makes sense. Is there only move? Is there only move? You don't have a running game to pass the game. Google's doing a version of that, you know, whatever the RPO offense, whatever it is, right? Run pass option. And so they're not going to recreate the building blocks like Amazon, play to their strengths and both models will work, right? The market's big enough where it can have building blocks, building blocks, building blocks, Amazon and kind of full stack with Azure. Azure is doing very well now because they have the full stack, but they also have a head start on knowing how to build these applications for working with OpenEye for the past, you know, several years. And they can deliver value today and get paid for it again. And again, just like the way it was embryonic, pages load slow, they got faster. So the question is, as a startup or a company, get in the game, don't stare at the navel, don't raise your deck chairs, get in the game. The other point is the margin model is superior with Microsoft. I mean, even Bomber couldn't kill Microsoft. Oh. Well, I think we were talking about like distribution matters more than technology at the end of the day, right? And so you have the greatest IP, you know, the Damian vs. Goliaths, can you build IP before they build distribution or vice versa? So they got distribution, all these apps, Microsoft shoves AI through the distribution, which are their apps? On the flip side, if you're a startup, you might have great technology, you know, can you build distribution to compete? And that's the race, and that's why I have a job. What about the edge action? You're seeing some guys do Silicon and develop software for the edge, you know, not big numbers yet, but it looks like a giant TAM, I just can't get my hands around it. You know, it depends how you define edge, right? So your phone could be the edge, right? So there will be inference of your application, inference of some models at the edge, in your home, in your car, in your pocket, on your face, right? Will there also be inference in these local pops nearby? Maybe, right, I think Cloudflare announced they're doing some inference or AI inference edge. I think it starts like fly.io, we're talking about that. So I think it's reasonable to assume there will be some inference at the edge. The question is it pushed all the way to your device or to some like pop at the edge, like a Cloudflare or fly or something else, or just centralized? Is your fan of the Cloudflare model? I do, especially around security. I'm admiring, what Cloudflare's done great is they built out this infrastructure on pops, and once they have that, they just sell more stuff, security, R3, you know, their database, there's object storage. I'm going to invest a couple of Kato networks, similar ideas, security company, but they build their own pops of network security, and now they sell more and more security services through the same network of pops, right? Again, both companies have the kind of- Control the route. It's like, don't be dependent on Amazon, build your own infrastructure. Kato says, hey, we're going to do cloud security, cloud networking on our own infrastructure. Jerry, great to have you on. We've got like 30 seconds left. What are you working on now? Give the audience a taste of what's your thesis now? What are you looking for for companies? What are you investing in? It's still the same stuff. Seed Series A, super early enterprise software. I would want to be the right side of history. So there are things like faster data, more data, cheaper storage costs like S3 going down. So when I see trends like that, right? Trends of, you know, object storage going down, getting faster, more data and volumes going up, I'm riding those waves. And so early States founders, data, cloud and AI, but- Check sizes? Anywhere from a few hundred K to like $10 million. So Seed Series A, but it's really looking for great founders, all things from AI on down. And Silicon performance like we've never seen before. It's incredible, right? We're all going to benefit. So I'm super excited about the 2024. All right, CUBE coverage here in location. We're in the MongoDB Emerald Club. No, Emerald Lounge. Emerald Lounge. Emerald Club is a car rental thing. It's awesome here. Thanks to Mongo for having us here on the set. Stay with theCUBE. We'll be right back after this break.