AWS Summit NYC 2024 | Keynote Analysis Pt. 1 with Zeus Kerravala | ZK Research

theCUBE.net

info Help
Sign in
Register

Clips
News
More from AWS Summit NYC 2024

Zeus Kerravala

Founder & Principal Analyst

ZK Research

From code to cloud, how is AWS shaping the gen AI landscape? TheCUBE keynote analysis from AWS Summit NY

The generative artificial intelligence wave has entered the infrastructure stage. Hyperscalers and solutions companies are laying out the gen AI blueprint for simplified, secure and scalable deployment.One such name is Amazon Web Services Inc., whose messaging signals an accelerating focus on enterprise AI, marked by rapid innovation, an expanding partner ecosystem and support for broadened app layers.On set with theCUBE at AWS Summit New York 2024.“I think [AWS’ Matt Wood] used the platform to really establish the fact that AWS is not just a participant in the gen AI wars, they are the leader in it,” said Zeus Kerravala (pictured, left), founder of ZK Research and CUBE Collective analyst

play_circle_outline Matt Wood Highlights AWS Leadership in genAI at NYC Summit with 236 Updates

play_circle_outline Discussion on enterprise vs. developer focus, ecosystem partnerships, and App Studio

play_circle_outline Importance of guardrails, responsible AI, and data sources in genAI development

play_circle_outline Analysis of AWS ecosystem viability, Partner program success, and future directions

Info

Keynote Analysis Pt. 1 with Zeus Kerravala | ZK Research

Zeus Kerravala

Founder & Principal Analyst ZK Research

search

>> Hello, welcome back to theCUBE's coverage here at New York City for AWS Summit. I'm John Furrier, your host of theCUBE, and Zeus Kerravala is here, part of our Cube Collective. Also, principal and owner of ZK Research. Zeus, it's becoming a tradition that you come on with the keynotes. We're going to have Sarbjeet come on later with Stackpane to give his analysis, but we do this keynote analysis every time. It's almost like a tradition.

Zeus Kerravala

>> Yes. tradition.

>> Great to get your perspective. Look, I mean, we've just been traveling around. You go to a lot of the events, you see the keynotes. This is kind of a quasi monster event because it's like a summit for Amazon. It's not re:Invent, but it's New York City. It's a ground zero for all the action in finance, Wall Street, but also there's a ton of developer community building out here. You're seeing the VCs from California come here. It's a heavy AI culture here in New York. We've all been seeing it for years, but now it seems to be popping really hard on all cylinders here in New York. So it's not just FinTech or these AdTech areas where there's real-time information and low latency like , generative AI is bringing a whole nother wave. And so in the keynotes that you heard, Matt Wood, what's your analysis?

Zeus Kerravala

>> Yeah, well, they call this AWS Summit in New York. They should call it genAI Summit because all the content was about that. And I felt like this was the year Matt really laid down the hammer, "We are the genAI company." I think if you look at a lot of the media coverage of genAI, people have written that AWS has fallen behind, Microsoft's stolen the show or whatever. And I think he used the platform to really establish the fact that AWS is not just a participant in the genAI wars, they are the leader in it. In fact, he put up some good data points that over the last year they had 236 updates, I think the number was, which he said was more than all the other cloud companies combined. And then he rolled out customer after customer, Nasdaq, Bayer, there was one other customer, but what they were trying to do with it. I recently just talked to Rocket Mortgage as well. They've built their whole platform on it, they're doing a session here. So I think between the customers he had on stage and a lot of the data points they gave and a lot of the innovation they rolled out, Matt Wood's point is, "We're not just playing in the genAI space, we're leading it." And if cloud is indeed the platform for genAI, then rightfully so that they've been the leader in the cloud really since its inception and that would make sense.

>> Yeah, a couple of observations on my end. One, Swami was not here. Last year he had the keynote. The size of the event was smaller. Remember, last year it was a packed house. Remember the long lines? I'm not sure about this, I'm going to check on the reporting on this to get more detail, but I think they might've curtailed the registration so that a lot of the homeless didn't come in and you have protesting.

Zeus Kerravala

>> We had a lot of protestors last year.

>> Last year was a lot of protests. So again, instant free pretty much here. And Matt's been around for a while pre being the poster child at genAI, he's been spearheading a lot of those stuff with Swami back in the day. I remember those interviews we did on theCUBE. Did you think that he hit the mark on the consumer side as well, or developer side? Because clearly with Q, I mean, they seem to be double down on Q. I get it, but not everyone needs Q. They need the models. And I think he hit a home run with the models merging, models interacting. I thought that was something we've been reporting on theCUBE for over a year now, that models would interact. But was it too enterprisey? And was there any hope for developers, like the classic startup? I saw a startup in the hallway here, it's like, "I'm doing a startup, I don't need VCF or VMware to automate Kubernetes. I'm doing my own thing. I don't really need Q." So what was your takeaway in terms of the percentage of enterprise content versus, say, pure developer?

Zeus Kerravala

>> Yeah, it was interesting because developer was there, but it was at the tail end of it, and then he gave some data points around CodeWhisperer and how that's been used with genAI to help write code faster. But I did think it actually took a backseat to a lot of the enterprise things. He talked a lot about models and I think their approach to genAI has been there isn't one model to rule them all, they're going to be open, they're going to take more of a platform approach. But I think to answer your question on developer, it was represented but not as heavily, you're right, as it could have been, and it was at the tail end of that keynote. So they weren't leading with developer.

>> I mean, I like the code assistant thing, don't get me wrong, I love that, but that's not the developer I'm talking about. I'm talking about the developer that had success with Gen 1 Amazon Web Services. "I want to build an app and I don't yet know what it's going to be."

Zeus Kerravala

>> A real person.

>> "I want spin up a server. I don't want to buy a server. Oh, they got storage and queuing, that's all I need because I'm not going to go buy a Supermicro box and host it at my home and some hosting provider. It's going to take too much time. I'm just going to play with the-"

Zeus Kerravala

>> I don't know if that's what the industry is right now with genAI though, John. I think a lot of it is people wanting to buy components that they can just deploy out of the box and be able to use that. I think the developer and the builder part will come. I just think right now there is such an arms race within businesses to get genAI into their contact centers, into their workforce, into their employees, that if you actually went through the process of trying to build it ... And understand, too, even the developer stuff they were talking about was using genAI to help write code faster versus building your own genAI. Which obviously NASDAQ's doing that and things like that, but Amazon is trying to make it easier for customers to deploy.

>> Well, I guess what I'm saying is is that there's two types of developers. There's the developers that need help and develop stuff. They work for a company or whatever. And then there's a startup, entrepreneur. I'm talking about the entrepreneur. What is the genAI use case for an entrepreneur? They're going to see an opportunity and build an app. And so I just can't see in the Bedrock and the Anthropics why ... It seems like a lot of extra stuff to get around just to do something. And I yet haven't seen that entrepreneurial match yet. It's so easy.

Zeus Kerravala

>> Yeah, you're talking about the incubator developer.

>> Or just, "Hey, I have an idea. I want to build a new Instagram or a new news app that automates newsfeeds," or-

Zeus Kerravala

>> No, I don't think that was really represented here much. Obviously those companies still use Amazon a lot, but Amazon's obviously been trying to bring more of their components together for enterprise. And, look, there's a tipping point here with cloud. I think the large enterprise was really adverse to cloud for a long time, but most of them have all gone in with both feet now. And if they were adverse to it, genAI is the thing that's actually forced them into it.

>> It's interesting, it's interesting. We're living in a world now where you have entrepreneurship post hyperscalers at full production. Remember, Amazon was just this little cloud that just took advantage of arbitrage from the startups. Now you got startups and all the enterprises using the cloud. So I guess what I liked about his keynote, which I thought was kind of entrepreneurial, was if I was an entrepreneur, what would I do? I love the multiple models interacting. I thought that was good. And then access to data sources I thought was a pretty good part of the keynote because at the end of the day, you're ultimately rolling data around if you're a developer. Like if I'm going to build something cool, it's probably going to be very enterprise-like. I mean, there's no Airbnb from scratch because that's mobile phone and SaaS, now it has to be in the cloud at some level. To your point, it's not a good comparison to say entrepreneurship in 2005 is a lot different than entrepreneurship to 2025.

Zeus Kerravala

>> I'm not sure you can do it not in the cloud. I just think you wouldn't have access to the data, the scale. To your point, building your own VCF stack isn't simple and I'm not sure it's the payback anymore for it. That's really become a large enterprise thing. I did like the focus on guardrails too, and I think that is something that even entrepreneurs can focus on, is making sure that AI's being done responsibly. And I think that's something AWS is trying to do since it launched. And it's funny because it's using the AI to check AI. But I think that's an important part of it, is understanding the subtle nuances of what's a good response, what's a bad response, what's a good outcome, what's a bad outcome. And they're trying to build a lot of that checking into their models to make sure that it's safe for customers to deploy. And that's an important part. Whenever I talk to customers, they are scared to death right now, John, of ... Almost every CIO talk to says their bosses are hammering them, "We have to have an AI strategy." But everyone remembers, in the early days of chatbots and things that didn't go so well, and no one wants to be on CNN as the company that botched their AI. The responsible AI piece of it has to be an integrated part of it in order for companies to have the confidence to be able to-

>> Guardrails are going to give really good parameters. I think the other thing that I liked was the App Studio, to seeing where that goes. What's your take on App Studio?

Zeus Kerravala

>> Well, again, it's been built to simplify the rollout of genAI within apps. And so they've got a bunch of connectors now with a bunch of other companies. And so this is more of an ecosystem approach for AWS, they've always been good at that. And so I think that we're not building these things in isolation. We have to understand how it interconnects with other applications, and I think the number of partners they had there was a good proof run for it.

>> I mean, one of the things I think is that, and I want to get your thoughts on this, Rob Strechay and I were talking about this earlier in the day before the keynote, their Partners and developers, and mainly Partners, is a real competitive advantage for Amazon. Because they own AIS and they have Silicon, they got the performance going on. Graviton, Ford, Dash Shipping, the platforms of service. What do you see in the Partner side? What are you hearing from Partners? The Competency Program is starting to seeing 60-plus people now globally coming in that are saying, "We are now AI competent," and they have a high bar. There's a front lines, I want to say general contractor like mentality. Anyone who's built a house or has been involved in construction knows that you hire a general contractor and they bring piece parts together and they get suppliers. And you see a similar trend going on here with genAI, where there's a ton of integration and it's not easy in the enterprise. So you have this new general contractor kind of role that-

Zeus Kerravala

>> You're talking about the reseller partners? Yeah, yeah.

>> You got the reseller partners, you got the distribute ... The distributor's more on the resale. But we also got the GSIs, the Global System Integrators. That's right Deloitte out here, front and center. You got Accenture, you got IBM is huge on this.

Zeus Kerravala

>> Well, that's where Ruba Borno has made a huge difference. She came over from Cisco, who's to me the marquee partner company in tech, and she's revamped their partner organization to be more ... Just create partner value in a lot of different ways, it's not just Marketplace. To your point, there's GSIs, there's VARs, there's things like that that all can participate in the AWS ecosystem. And I think she's had a huge impact, which I thought she would. I talked to her when she first joined and she was really excited about the partner opportunity here.

>> It's a clean sheet of paper. But she had to do some cleanup. They have now collapsed and merge APN with Marketplace, so one team go to the market with synergies.

Zeus Kerravala

>> Yeah, it created a lot of inconsistencies then. And obviously they wanted everybody in Marketplace, but I think-

>> Not everyone's ready.

Zeus Kerravala

>> Yeah. Well, and it's not the only delivery model. Marketplace is fantastic for certain things and a lot of customers are going to buy from that, but a lot of customers prefer the traditional partners. And a lot of them, to your point, especially big enterprises, go the GSI route. And so you got to be able to cater all of those different things do that.

>> What's your analysis on their ecosystem? Obviously it's robust, it's vibrant. Always has been. If you go to re:Invent, re:Invent's become the industry show. We were speculating, we're fearful that Amazon, if they become too internal focused, it becomes an Amazon show versus an industry event. What does Amazon Web Services need to do to keep reinvent to be a bellwether industry show, what specifically around the ecosystem? Is it money making? Is it delivery? What's your analysis of ecosystem viability?

Zeus Kerravala

>> To your point, it is ecosystem. And if they do want to reinvent to be the industry show for cloud, not just the Amazon show, this is where they need to make sure they have a steady stream of ecosystem partners. I'm hoping Jensen shows up there. We'll see if he gets as much time as he did at HPE.

>> We'll get the stopwatch out, clock . He probably won't get much time.

Zeus Kerravala

>> So what did Matt say on stage? He said the AWS cloud is the best place to run Nvidia. If that's true, let's start seeing some proof points on that. And so if the world's going AI and Nvidia is the centerpiece of that, then that Nvidia ecosystem needs to marry with the AWS ecosystem. And so we should see a pretty big dose of Nvidia at re:Invent if they're actually true to that.

>> You, myself, Bob Laliberte with theCUBE research team have been talking a lot about how networking is evolving in this new systems architecture. You mentioned Nvidia. We talk about clustered systems, they call it the AI Factory. So you start to see, and Amazon's had this for a while, where it's a collection of stuff that makes the system work for the power that is needed in generative AI. So that's kind of a general statement, but let's assume that for a second. The characteristics of the network and the characteristics of storage and compute are going to change depending upon what gen I is hitting it. So we're seeing that now. And I want to bring up networking, because networking is that last area that's going to impact storage. Doug Gourlay was recently promoted to the CEO job at Qumulo.

Zeus Kerravala

>> I saw that.

>> That's a storage company. You see the success of VAST Data. Doug's a networking guy, so connect the dots, networking is now a key part of this storage system because you're moving packets into storage and they're going to be doing inference or training, it might have different read/write capabilities. That was once only an AdTech problem or a finance problem. So now you're starting to see the infrastructure at networking. What's your take on this whole systems view? I mean, what does your data show you and what's your opinion? Because you've been covering networking for a while. It's a big part of it.

Zeus Kerravala

>> It's funny when Charlie Giancarlo, who's also a network person, when he joined Pure Storage, I remember shooting him a text message saying, "Charlie, you went to a storage company. Ha, ha, ha, what's up with that? Are you ready for retirement?" And he actually walked me through the thesis, which I've lived with since then, that when you think about AI ... And understand, this was seven years ago.

>> Exactly. I know the AI was coming next, I'm laughing already.

Zeus Kerravala

>> But he said in order for AI to really hum, you need a fast processor, hence, we're moving to GPUs. You need fast networking. And Arista, Cisco, Juniper, all those companies are playing that, and you need fast storage. And the three of them have to work in sync in order to make AI work. If one of those components isn't up to par with the other ones, it's going to be the bottleneck, and the money you're going to spend on the other two, you're not going to get your payback. And so how do you get those three things to work together? You have to take a systems approach to it. If you were to actually try and build these things in isolation and hope that they work together, I mean good luck. Yeah, I mean, even look how just with general computing. You've got slow processors, slow network or whatever, it has a huge effect on performance. And then when you raise the stakes to the AI level where the costs are so much higher that you can't-

>> And the stakes too are higher because the business is being bet on it.

Zeus Kerravala

>> Yes, you're right. And so I think this is where it's going to be interesting to see, with all the money companies spent on GPU based systems, if they've indeed made the investments on the storage side or the networking side. If not, I think we could see some pullback on spending on AI systems while companies try and get the rest of their infrastructure.

>> I was speculating and riffing with Rob Stretchy, and I'll riff with you for a second, I want to get your thoughts on this. You almost think, remember the old days of the old internet, over provisioning, it was a big problem. And you overprovision your network all out, we overbuilt. Do you feel like there might be a scenario where people bought too many GPUs and realize, "Hey, I got to run this on power now. Oh, I'm going to need other stuff." Because what's happening, it feels like there's a frenzy for GPUs. Obviously Nvidia stock's through the roof, the software with it, but there's other factors. Amazon's got a very efficient sustainability ratio. They just hit there seven years earlier than on their sustainability goals today that was announced. So there's some benefit to running maybe GPUs service on the cloud, if you're using other stuff. And then energy and power become huge. What's your take on this? Am I fantasizing in the wrong direction here?

Zeus Kerravala

>> No, I've been a little worried about that, that we overspent on GPU systems. We're not really getting the bang for the buck for it because we're not sure how to run them properly. There's nothing wrong with Nvidia tech. It works.

>> Yes, of course. There's a line around the corner to get them.

Zeus Kerravala

>> But I was talking with CISO or a CIO about this, that he doesn't want the company to use them incorrectly. There's a lot of things you can just use search for and that it's like one 10th the compute power of genAI. And so as they try and refine, where do you use genAI, where do you use basic search, where do you use basic analytics, the amount of AI they use might be less than their thinking.

>> It's idle time, basically.

Zeus Kerravala

>> Yeah, yeah. So I do think companies have overspent on it and I am expecting over the back half of the year, I'm not trying to make a stock prediction on Nvidia or anything.

>> I think Dave Vellante and the team have some data on idle time, or another entrepreneur. I talked to an entrepreneur about this. And there's kind of a vibe going on in Silicon Valley right now that looks a lot like the server virtualization days, pre virtualization, where, "Hey, we have ll these servers and they're not being utilized. Let's create a virtualization, virtual server." And not to say that's going to happen here, but managing workloads to take advantage of capabilities is interesting.

Zeus Kerravala

>> Yeah. Well, ideally, like you think what typical server utilization runs 30, 35%. If you're spending the money you are on DGX systems and you're running them at 30%, I mean, you're wasting a ton of money. And so this is where the industry and the systems approach can actually help companies maximize that utilization. And I do think it's something to watch for the end of the year. We'll know more at re:Invent by the time that rolls around, but I do think the industry is at risk of seeing a pullback on spend.

>> Zeus, great to see you. Thanks for the keynote analysis and extra commentary. Zeus Kerravala here on theCUBE, covering AWS Summit New York City. It's our popup CUBE. I'm John Furrier, the host, got more coming. Got a lot of great guests coming up, Matt Wood, we got some customer examples, partner examples from Amazon Web Services team here showing some great product. We got some hot startups as well, so stay with us.

Keynote Analysis Pt. 1 with Zeus Kerravala | ZK Research

search

>> Hello, welcome back to theCUBE's coverage here at New York City for AWS Summit. I'm John Furrier, your host of theCUBE, and Zeus Kerravala is here, part of our Cube Collective. Also, principal and owner of ZK Research. Zeus, it's becoming a tradition that you come on with the keynotes. We're going to have Sarbjeet come on later with Stackpane to give his analysis, but we do this keynote analysis every time. It's almost like a tradition.

Zeus Kerravala

>> Yes. tradition.

>> Great to get your perspective. Look, I mean, we've just been traveling around. You go to a lot of the events, you see the keynotes. This is kind of a quasi monster event because it's like a summit for Amazon. It's not re:Invent, but it's New York City. It's a ground zero for all the action in finance, Wall Street, but also there's a ton of developer community building out here. You're seeing the VCs from California come here. It's a heavy AI culture here in New York. We've all been seeing it for years, but now it seems to be popping really hard on all cylinders here in New York. So it's not just FinTech or these AdTech areas where there's real-time information and low latency like , generative AI is bringing a whole nother wave. And so in the keynotes that you heard, Matt Wood, what's your analysis?

Zeus Kerravala

>> Yeah, well, they call this AWS Summit in New York. They should call it genAI Summit because all the content was about that. And I felt like this was the year Matt really laid down the hammer, "We are the genAI company." I think if you look at a lot of the media coverage of genAI, people have written that AWS has fallen behind, Microsoft's stolen the show or whatever. And I think he used the platform to really establish the fact that AWS is not just a participant in the genAI wars, they are the leader in it. In fact, he put up some good data points that over the last year they had 236 updates, I think the number was, which he said was more than all the other cloud companies combined. And then he rolled out customer after customer, Nasdaq, Bayer, there was one other customer, but what they were trying to do with it. I recently just talked to Rocket Mortgage as well. They've built their whole platform on it, they're doing a session here. So I think between the customers he had on stage and a lot of the data points they gave and a lot of the innovation they rolled out, Matt Wood's point is, "We're not just playing in the genAI space, we're leading it." And if cloud is indeed the platform for genAI, then rightfully so that they've been the leader in the cloud really since its inception and that would make sense.

>> Yeah, a couple of observations on my end. One, Swami was not here. Last year he had the keynote. The size of the event was smaller. Remember, last year it was a packed house. Remember the long lines? I'm not sure about this, I'm going to check on the reporting on this to get more detail, but I think they might've curtailed the registration so that a lot of the homeless didn't come in and you have protesting.

Zeus Kerravala

>> We had a lot of protestors last year.

>> Last year was a lot of protests. So again, instant free pretty much here. And Matt's been around for a while pre being the poster child at genAI, he's been spearheading a lot of those stuff with Swami back in the day. I remember those interviews we did on theCUBE. Did you think that he hit the mark on the consumer side as well, or developer side? Because clearly with Q, I mean, they seem to be double down on Q. I get it, but not everyone needs Q. They need the models. And I think he hit a home run with the models merging, models interacting. I thought that was something we've been reporting on theCUBE for over a year now, that models would interact. But was it too enterprisey? And was there any hope for developers, like the classic startup? I saw a startup in the hallway here, it's like, "I'm doing a startup, I don't need VCF or VMware to automate Kubernetes. I'm doing my own thing. I don't really need Q." So what was your takeaway in terms of the percentage of enterprise content versus, say, pure developer?

Zeus Kerravala

>> Yeah, it was interesting because developer was there, but it was at the tail end of it, and then he gave some data points around CodeWhisperer and how that's been used with genAI to help write code faster. But I did think it actually took a backseat to a lot of the enterprise things. He talked a lot about models and I think their approach to genAI has been there isn't one model to rule them all, they're going to be open, they're going to take more of a platform approach. But I think to answer your question on developer, it was represented but not as heavily, you're right, as it could have been, and it was at the tail end of that keynote. So they weren't leading with developer.

>> I mean, I like the code assistant thing, don't get me wrong, I love that, but that's not the developer I'm talking about. I'm talking about the developer that had success with Gen 1 Amazon Web Services. "I want to build an app and I don't yet know what it's going to be."

Zeus Kerravala

>> A real person.

>> "I want spin up a server. I don't want to buy a server. Oh, they got storage and queuing, that's all I need because I'm not going to go buy a Supermicro box and host it at my home and some hosting provider. It's going to take too much time. I'm just going to play with the-"

Zeus Kerravala

>> I don't know if that's what the industry is right now with genAI though, John. I think a lot of it is people wanting to buy components that they can just deploy out of the box and be able to use that. I think the developer and the builder part will come. I just think right now there is such an arms race within businesses to get genAI into their contact centers, into their workforce, into their employees, that if you actually went through the process of trying to build it ... And understand, too, even the developer stuff they were talking about was using genAI to help write code faster versus building your own genAI. Which obviously NASDAQ's doing that and things like that, but Amazon is trying to make it easier for customers to deploy.

>> Well, I guess what I'm saying is is that there's two types of developers. There's the developers that need help and develop stuff. They work for a company or whatever. And then there's a startup, entrepreneur. I'm talking about the entrepreneur. What is the genAI use case for an entrepreneur? They're going to see an opportunity and build an app. And so I just can't see in the Bedrock and the Anthropics why ... It seems like a lot of extra stuff to get around just to do something. And I yet haven't seen that entrepreneurial match yet. It's so easy.

Zeus Kerravala

>> Yeah, you're talking about the incubator developer.

>> Or just, "Hey, I have an idea. I want to build a new Instagram or a new news app that automates newsfeeds," or-

Zeus Kerravala

>> No, I don't think that was really represented here much. Obviously those companies still use Amazon a lot, but Amazon's obviously been trying to bring more of their components together for enterprise. And, look, there's a tipping point here with cloud. I think the large enterprise was really adverse to cloud for a long time, but most of them have all gone in with both feet now. And if they were adverse to it, genAI is the thing that's actually forced them into it.

>> It's interesting, it's interesting. We're living in a world now where you have entrepreneurship post hyperscalers at full production. Remember, Amazon was just this little cloud that just took advantage of arbitrage from the startups. Now you got startups and all the enterprises using the cloud. So I guess what I liked about his keynote, which I thought was kind of entrepreneurial, was if I was an entrepreneur, what would I do? I love the multiple models interacting. I thought that was good. And then access to data sources I thought was a pretty good part of the keynote because at the end of the day, you're ultimately rolling data around if you're a developer. Like if I'm going to build something cool, it's probably going to be very enterprise-like. I mean, there's no Airbnb from scratch because that's mobile phone and SaaS, now it has to be in the cloud at some level. To your point, it's not a good comparison to say entrepreneurship in 2005 is a lot different than entrepreneurship to 2025.

Zeus Kerravala

>> I'm not sure you can do it not in the cloud. I just think you wouldn't have access to the data, the scale. To your point, building your own VCF stack isn't simple and I'm not sure it's the payback anymore for it. That's really become a large enterprise thing. I did like the focus on guardrails too, and I think that is something that even entrepreneurs can focus on, is making sure that AI's being done responsibly. And I think that's something AWS is trying to do since it launched. And it's funny because it's using the AI to check AI. But I think that's an important part of it, is understanding the subtle nuances of what's a good response, what's a bad response, what's a good outcome, what's a bad outcome. And they're trying to build a lot of that checking into their models to make sure that it's safe for customers to deploy. And that's an important part. Whenever I talk to customers, they are scared to death right now, John, of ... Almost every CIO talk to says their bosses are hammering them, "We have to have an AI strategy." But everyone remembers, in the early days of chatbots and things that didn't go so well, and no one wants to be on CNN as the company that botched their AI. The responsible AI piece of it has to be an integrated part of it in order for companies to have the confidence to be able to-

>> Guardrails are going to give really good parameters. I think the other thing that I liked was the App Studio, to seeing where that goes. What's your take on App Studio?

Zeus Kerravala

>> Well, again, it's been built to simplify the rollout of genAI within apps. And so they've got a bunch of connectors now with a bunch of other companies. And so this is more of an ecosystem approach for AWS, they've always been good at that. And so I think that we're not building these things in isolation. We have to understand how it interconnects with other applications, and I think the number of partners they had there was a good proof run for it.

>> I mean, one of the things I think is that, and I want to get your thoughts on this, Rob Strechay and I were talking about this earlier in the day before the keynote, their Partners and developers, and mainly Partners, is a real competitive advantage for Amazon. Because they own AIS and they have Silicon, they got the performance going on. Graviton, Ford, Dash Shipping, the platforms of service. What do you see in the Partner side? What are you hearing from Partners? The Competency Program is starting to seeing 60-plus people now globally coming in that are saying, "We are now AI competent," and they have a high bar. There's a front lines, I want to say general contractor like mentality. Anyone who's built a house or has been involved in construction knows that you hire a general contractor and they bring piece parts together and they get suppliers. And you see a similar trend going on here with genAI, where there's a ton of integration and it's not easy in the enterprise. So you have this new general contractor kind of role that-

Zeus Kerravala

>> You're talking about the reseller partners? Yeah, yeah.

>> You got the reseller partners, you got the distribute ... The distributor's more on the resale. But we also got the GSIs, the Global System Integrators. That's right Deloitte out here, front and center. You got Accenture, you got IBM is huge on this.

Zeus Kerravala

>> Well, that's where Ruba Borno has made a huge difference. She came over from Cisco, who's to me the marquee partner company in tech, and she's revamped their partner organization to be more ... Just create partner value in a lot of different ways, it's not just Marketplace. To your point, there's GSIs, there's VARs, there's things like that that all can participate in the AWS ecosystem. And I think she's had a huge impact, which I thought she would. I talked to her when she first joined and she was really excited about the partner opportunity here.

>> It's a clean sheet of paper. But she had to do some cleanup. They have now collapsed and merge APN with Marketplace, so one team go to the market with synergies.

Zeus Kerravala

>> Yeah, it created a lot of inconsistencies then. And obviously they wanted everybody in Marketplace, but I think-

>> Not everyone's ready.

Zeus Kerravala

>> Yeah. Well, and it's not the only delivery model. Marketplace is fantastic for certain things and a lot of customers are going to buy from that, but a lot of customers prefer the traditional partners. And a lot of them, to your point, especially big enterprises, go the GSI route. And so you got to be able to cater all of those different things do that.

>> What's your analysis on their ecosystem? Obviously it's robust, it's vibrant. Always has been. If you go to re:Invent, re:Invent's become the industry show. We were speculating, we're fearful that Amazon, if they become too internal focused, it becomes an Amazon show versus an industry event. What does Amazon Web Services need to do to keep reinvent to be a bellwether industry show, what specifically around the ecosystem? Is it money making? Is it delivery? What's your analysis of ecosystem viability?

Zeus Kerravala

>> To your point, it is ecosystem. And if they do want to reinvent to be the industry show for cloud, not just the Amazon show, this is where they need to make sure they have a steady stream of ecosystem partners. I'm hoping Jensen shows up there. We'll see if he gets as much time as he did at HPE.

>> We'll get the stopwatch out, clock . He probably won't get much time.

Zeus Kerravala

>> So what did Matt say on stage? He said the AWS cloud is the best place to run Nvidia. If that's true, let's start seeing some proof points on that. And so if the world's going AI and Nvidia is the centerpiece of that, then that Nvidia ecosystem needs to marry with the AWS ecosystem. And so we should see a pretty big dose of Nvidia at re:Invent if they're actually true to that.

>> You, myself, Bob Laliberte with theCUBE research team have been talking a lot about how networking is evolving in this new systems architecture. You mentioned Nvidia. We talk about clustered systems, they call it the AI Factory. So you start to see, and Amazon's had this for a while, where it's a collection of stuff that makes the system work for the power that is needed in generative AI. So that's kind of a general statement, but let's assume that for a second. The characteristics of the network and the characteristics of storage and compute are going to change depending upon what gen I is hitting it. So we're seeing that now. And I want to bring up networking, because networking is that last area that's going to impact storage. Doug Gourlay was recently promoted to the CEO job at Qumulo.

Zeus Kerravala

>> I saw that.

>> That's a storage company. You see the success of VAST Data. Doug's a networking guy, so connect the dots, networking is now a key part of this storage system because you're moving packets into storage and they're going to be doing inference or training, it might have different read/write capabilities. That was once only an AdTech problem or a finance problem. So now you're starting to see the infrastructure at networking. What's your take on this whole systems view? I mean, what does your data show you and what's your opinion? Because you've been covering networking for a while. It's a big part of it.

Zeus Kerravala

>> It's funny when Charlie Giancarlo, who's also a network person, when he joined Pure Storage, I remember shooting him a text message saying, "Charlie, you went to a storage company. Ha, ha, ha, what's up with that? Are you ready for retirement?" And he actually walked me through the thesis, which I've lived with since then, that when you think about AI ... And understand, this was seven years ago.

>> Exactly. I know the AI was coming next, I'm laughing already.

Zeus Kerravala

>> But he said in order for AI to really hum, you need a fast processor, hence, we're moving to GPUs. You need fast networking. And Arista, Cisco, Juniper, all those companies are playing that, and you need fast storage. And the three of them have to work in sync in order to make AI work. If one of those components isn't up to par with the other ones, it's going to be the bottleneck, and the money you're going to spend on the other two, you're not going to get your payback. And so how do you get those three things to work together? You have to take a systems approach to it. If you were to actually try and build these things in isolation and hope that they work together, I mean good luck. Yeah, I mean, even look how just with general computing. You've got slow processors, slow network or whatever, it has a huge effect on performance. And then when you raise the stakes to the AI level where the costs are so much higher that you can't-

>> And the stakes too are higher because the business is being bet on it.

Zeus Kerravala

>> Yes, you're right. And so I think this is where it's going to be interesting to see, with all the money companies spent on GPU based systems, if they've indeed made the investments on the storage side or the networking side. If not, I think we could see some pullback on spending on AI systems while companies try and get the rest of their infrastructure.

>> I was speculating and riffing with Rob Stretchy, and I'll riff with you for a second, I want to get your thoughts on this. You almost think, remember the old days of the old internet, over provisioning, it was a big problem. And you overprovision your network all out, we overbuilt. Do you feel like there might be a scenario where people bought too many GPUs and realize, "Hey, I got to run this on power now. Oh, I'm going to need other stuff." Because what's happening, it feels like there's a frenzy for GPUs. Obviously Nvidia stock's through the roof, the software with it, but there's other factors. Amazon's got a very efficient sustainability ratio. They just hit there seven years earlier than on their sustainability goals today that was announced. So there's some benefit to running maybe GPUs service on the cloud, if you're using other stuff. And then energy and power become huge. What's your take on this? Am I fantasizing in the wrong direction here?

Zeus Kerravala

>> No, I've been a little worried about that, that we overspent on GPU systems. We're not really getting the bang for the buck for it because we're not sure how to run them properly. There's nothing wrong with Nvidia tech. It works.

>> Yes, of course. There's a line around the corner to get them.

Zeus Kerravala

>> But I was talking with CISO or a CIO about this, that he doesn't want the company to use them incorrectly. There's a lot of things you can just use search for and that it's like one 10th the compute power of genAI. And so as they try and refine, where do you use genAI, where do you use basic search, where do you use basic analytics, the amount of AI they use might be less than their thinking.

>> It's idle time, basically.

Zeus Kerravala

>> Yeah, yeah. So I do think companies have overspent on it and I am expecting over the back half of the year, I'm not trying to make a stock prediction on Nvidia or anything.

>> I think Dave Vellante and the team have some data on idle time, or another entrepreneur. I talked to an entrepreneur about this. And there's kind of a vibe going on in Silicon Valley right now that looks a lot like the server virtualization days, pre virtualization, where, "Hey, we have ll these servers and they're not being utilized. Let's create a virtualization, virtual server." And not to say that's going to happen here, but managing workloads to take advantage of capabilities is interesting.

Zeus Kerravala

>> Yeah. Well, ideally, like you think what typical server utilization runs 30, 35%. If you're spending the money you are on DGX systems and you're running them at 30%, I mean, you're wasting a ton of money. And so this is where the industry and the systems approach can actually help companies maximize that utilization. And I do think it's something to watch for the end of the year. We'll know more at re:Invent by the time that rolls around, but I do think the industry is at risk of seeing a pullback on spend.

>> Zeus, great to see you. Thanks for the keynote analysis and extra commentary. Zeus Kerravala here on theCUBE, covering AWS Summit New York City. It's our popup CUBE. I'm John Furrier, the host, got more coming. Got a lot of great guests coming up, Matt Wood, we got some customer examples, partner examples from Amazon Web Services team here showing some great product. We got some hot startups as well, so stay with us.