Hi, everyone, and welcome to The Six Five Summit: AI Unleashed 2026. For this AI infrastructure spotlight, we will be exploring how enterprise AI infrastructure is evolving as organizations look to securely bring AI closer to their business data, wherever that data resides. I am here with my colleague, Matt Kimball, and joining us is Karan Batta, Senior Vice President of Oracle Cloud Infrastructure. Karan, welcome to Six Five. It is nice to be here. Thank you so much. This is going to be a fun conversation, Karan. We are going to run you through the wringer here. I hope you are ready. Let us go. I work at Oracle, so it is not too dissimilar. It's called a normal half hour at Oracle. Yep. Let's start out setting the stage. AI, we've been talking about it for a while, right? Early days was models, models. We're really starting to see moving from those discussing foundational models and training and experimentation into production. We're really starting to see that, and as we watch enterprise organizations move from proof of concept and pilots and really starting to try to scale, one of the things we're seeing is that infrastructure's really come back into focus. We've moved away from talking about this model versus that model. From your perspective, what's changed in how enterprises should be really thinking about AI infrastructure over the last year or so? Yeah. Look, I think, the overall thought process, I would say, is it's gone from how do I build an AI model to essentially how do I operationalize AI across my entire organization, right? That's the key message that I would talk about. A year ago, a lot of the customers were like, "Well, which model should I use? What GPU should I buy? How fast can I train?" They're still important questions, but not for the enterprise customer, right? Today, I think I would say CIOs are asking very different set of questions. How do I connect my AI to my enterprise data? How do I run governance on it? How do I keep my sensitive information to myself within the bounds of my data center or my boundaries? How do I scale AI across thousands of employees and business processes, right? Actually, Oracle is going through that significant change ourselves. How do I control costs over time, right? That's actually the recent conversation a lot of the times is, great, we're using AI, but costs are ballooning. How do I control that? Is it through inferencing? Is it through building my own model? How do I think about the cost aspect of it, right? So this is why AI infrastructure has become really strategic, right? It's no longer a standard work alone, workload, so to speak. It's really embedded into, I would say, all of our applications. Especially working at Oracle, we're just not just an infrastructure company, we're an applications company as well, right? So it's embedded into ERP, HCM, supply chain, healthcare, financial services, manufacturing. So basically, every enterprise workflow that you can think of, right? Infrastructure is just much more than compute. It's more than GPU compute anyway. It's everything from security, identity, databases, massive datasets. We're seeing the enterprises go through the motions and be a bit more pragmatic about they don't want 10 different AI platforms. They want the data to come, rather than data come into AI, they want AI to come to the data. There's lots of change going on as they really figure out, how do I embed AI into all of my business processes, people, and function? That's essentially the big change. There's no answer yet, but I think they're going through that sea change now. Something you alluded to just a minute ago, one of the big challenges being governance. There was a time when mission-critical data for businesses existed in Oracle Databases, snug and secure in our own data centers. Part of the cloud revolution was Oracle giving people the ability to have mission-critical data in a lot of different places. Now we're in the AI era, and people are trying to juggle innovation speed with safety, security, and governance. How does that change what people expect from Oracle or from any cloud or on-prem infrastructure provider? I'll start with the last statement I just made, which was, most customers increasingly want the AI to come to their data, not the other way around. What that means is that, most enterprise data is not just sitting in one cloud. It's on premises, it's in an Oracle Database, maybe it's in some of the other cloud providers, maybe it's in some neo clouds. Some must actually remain inside a sovereign environment. With AI becoming critical infrastructure for a lot of governments, if you go outside the U.S., the geopolitical space, it's governed. It's regulated. AI changes all those assumptions. What customers are really telling us now is they want to keep the data where it lives. They want to apply AI consistently across all their environments, not Snowflake environments. They want to maintain a single security model and governance framework as well at the same time. For us, from an Oracle standpoint, this is essentially what distributed cloud has become. It's really become a significant differentiator for us. We've got now probably over 50 - 70 dedicated regions and Oracle Alloys that are deployed globally, and it's actually worked out really well for us because it's the same code, it's the same set of services, but the deployment model's slightly different depending on the needs that you have whether it's public cloud, whether it's a dedicated region at customer's facility, whether it's Cloud@Customer, if you've got your data sitting on an Oracle Exadata box, as an example, right next to your AI, whether it's sovereign cloud, partner clouds with Oracle Alloy. They expect the same level of services across all their different data pieces, and they want to apply AI across all of that, right? Because you're getting a single cloud platform with OCI, you're getting single governance layer and identity and security across all those pieces. Not to mention, I talked about Oracle Databases a little bit earlier. We have multi-cloud obviously, with our databases spanning across all of the different cloud providers, including OCI. This also expands to the data that's sitting in those other cloud providers as well. They don't want different strategies for AI across every environment. They actually want one single consistent operating model for all of it. When you talk about that, Karan, it's funny because I step back and I think multi-cloud has always been important for me as an enterprise CIO, right? There are a lot of reasons for that. There's operational consistency, or not consistency, but uptime, resilience. It's also I go to OCI for this, I go to AWS for that, and because there are specializations that sometimes, or services sometimes one cloud has that's just fantastic. It seems to me that has become even more strategic and more important as AI starts to play in the enterprise equation, right? Because of that kind of not just agility, but operational consistency, right? When you think about how multi-cloud continues to evolve over time, and especially as we start to see enterprise AI, what would you say, if an enterprise is looking two years out, an enterprise CIO, how do you think they would define success? What does that look like? Is it that just a single kind of operating plane for them or control plane for them, and is it that universal view of all their data? Is there more to it than that? Can you add a little bit more color on that? Yeah, look, I would say multi-cloud used to be more of a vendor lock-in strategy in the past, right? It's no longer that at all. It's actually about combining the best set of technologies depending on wherever they live, right? I would say five years ago when we started our relationship with Azure, as an example, a lot of the multi-cloud conversations were largely defensive. Today they're offensive, right? Customers are intentionally, I would say, choosing different clouds because each brings us unique trends, right? With AI, I think it only accelerates this trend, right? Because different models live in different clouds. Everybody's kind of taking their own bet on what cloud and what model's going to be sort of the key. Different enterprise applications are in different clouds. Databases are in different clouds as well, and that's one of the reasons why we have a multi-cloud platform for database. Specialized AI services continue to emerge in different verticals as well, right? We're obviously investing a lot in our applications from an AI perspective. So, success isn't going to be running identical infrastructure everywhere. It's going to be making those environments work really seamlessly across all the different environments, whether it's on-prem or whatnot, right? It's one of the big reasons why we invested five, six years ago in multi-cloud. We started with Azure. We started with the network connectivity, right? We wanted to make sure that network connectivity is there, because if you can't move data for cheap, nothing's really going to work, right? With AI, the data's sort of the key part of it. We started with Azure, we then expanded to Google Cloud, we then expanded to AWS. Now what we've become is the central core data hub almost I would call it, where you have network connectivity across all clouds back and forth, right? You're able to move that data free of charge across different clouds using multi-cloud. If your data is sitting in different clouds, it still sits on a singular platform, which is Oracle Database, and you can move that data around, you can use whatever models you want, wherever you want. Yeah. Maybe they want to use some verticalized service in Google, but maybe they want to run some core infrastructure in OCI, but then maybe they're running their data on-prem. Which is also why, kind of going back to the last question about distributed cloud, it's the same API at the end of the day. So you're able to build this control plane right on top, which sort of is fungible across many different environments at the end of the day. Look, I think customers aren't asking for multi-cloud because they love the complexity that it comes with. They're asking for it because they want the freedom to choose the best of the best, irrelevant of where it lives. Yeah. Some people like to collect cloud merit badges, though, and wear them on their jumpsuits like an F1 driver. Whenever we are having conversations about infrastructure, especially now, but really always, real business outcomes are always tied to these kinds of decisions. Today, I think you can almost cynically boil this down to the question of how much money can you save me so I can spend that money on tokens? If you are making the pitch for distributed cloud, can you nail that down to a single greatest benefit of distributed cloud? You have talked a lot about it already but we are in the elevator, we are riding up, we only have two more floors to go. What is the greatest advantage to distributed cloud? Look, I think distributed cloud is really ultimately about business outcomes, right? Not deployment models like I talked about, right? It really depends on the kind of customer, right, that is using it, right? Because each different customer will measure success slightly differently, I would say, when it comes to distributed cloud, right? Different industries have sort of different constraints. So like as an example, with healthcare it is going to be patient privacy, right? They have to have the patient records and things. It is going to be data residency, clinical AI, being close to the hospitals because of latency. For financial services, it is something completely different, which is regulatory compliance, low latency because if they are doing high frequency trading or, sensitive financial data as an example. With manufacturing, it is going to be things like factory automation, it is going to be AI near the production environments or predictive maintenance. Obviously with governments it is things like sovereignty, national security and so on. So it is every different industry has slightly different sort of benefits and success criteria. I think the thing that is key is that the data has got to live within the bounds that you have described. It has got to be right next to your core infrastructure at the end of the day that is already preexisting, and you have to have control over it. Those are sort of the three or four metrics that determine success. Hey, one last question for you, Karan, and I promise I will make it quick, but we saved the hardest one for last. We are in an interesting time, right? We are at a point where innovation is kind of, as in this industry, the tech industry has kind of hit breakneck speed, right? That is partly thanks to AI, but new silicon is coming out on an annual cadence. New models are updating frequently, new frameworks. It is just everything is moving so fast. As I talk with enterprise CIOs and IT leaders, it is like, it is really difficult to say, "Okay, here is the bet I am going to lay down today, and I want to make sure I am making the right bet because this thing is going to stick with me for years to come," right? So you are sitting back, you spend a lot of time thinking about these things for OCI. So as these enterprise leaders are kind of struggling with these decisions, do you have like principles or kind of like here are a few guiding posts for you to kind of think about or set out as you're making these strategic decisions that kind of brings you to what's next? At sort of the highest levels, there's probably a few things kind of like, I would say tailwinds. I think the key, like as an example, the first factor I would say is design for flexibility. Foundational models will continue to evolve and rapidly change, and so it's really hard to pick a winner or, hardware generations are going to continue to evolve. Everybody's initially it was GPUs from NVIDIA, then it was AMD, and then everybody else is coming out with custom ASICs. Infrastructure today should outlast sort of the today's model choices as well. So the fungibility and the flexibility is, I think, really, really key for these customers. The second thing I would say is building around openness. Customers, I would say they don't want like proprietary AI stacks, so they want open models, open frameworks, APIs. They want multi-cloud architectures. Our philosophy has been consistently customer choice, which is one of the reasons why we put the ability for customers to have data in other clouds on our stack. Then optimize for mere long-term economics. Inference is going to be the biggest operational expense for a lot of these customers. So thinking about how to optimize token cost, infrastructure efficiency, GPU utilization, network efficiency, operational simplicity, what kind of people you need. So I think those are kind of the key things. Then I think at the highest level, keeping infrastructure, I would say invisible is going to be key, because the best infrastructure always sort of disappears under the covers. Sure. Developers or scientists or data analysts shouldn't be thinking about infrastructure. Business users that are the token users shouldn't be thinking about infrastructure. They should just simply trust that the AI is going to work securely, performantly, and it's going to scale economically. So I think that's kind of the key set of principles I would think about. Well, Karan, thanks for joining us for this AI Infrastructure Spotlight. To our viewers, don't forget to hit subscribe, follow us on social media, and check out all of our Six Five Summit content at sixfivemedia.com/summit. We'll see you next-