FinOps X 2026 | Sabapathy Arumugam, Corestack

Clips
More from FinOps X 2026

Sabapathy Arumugam

Chief Technology Officer

Corestack

play_circle_outline Tokenomics for AI: Measuring Usage, Cost, and Business Value Beyond Traditional Cloud FinOps

play_circle_outline AI/agent governance: security, data leakage prevention, lifecycle, attribution

play_circle_outline Neo-Clouds and Multi-Agent AI: Multi-Cloud, Multi-Modal Deployments for Data Sovereignty

play_circle_outline Evolving FinOps: Attributing AI, SaaS, and Cloud Investments to Revenue with Visibility and KPIs

Info
Transcript

Sabapathy Arumugam, Corestack

Sabapathy Arumugam

Chief Technology Officer Corestack

In this interview from FinOps X 2026, Sabapathy Arumugam, chief technology officer of CoreStack, joins theCUBE's John Furrier and theCUBE Research's Paul Nashawaty to discuss how FinOps is evolving from cloud cost control into a broader discipline for managing AI value and technology investment. Arumugam draws a sharp distinction between traditional cloud FinOps — where spend on compute, network and storage maps to known applications — and AI FinOps, where enterprises are paying for intelligence measured in tokens. The defining question, he argues, is whether... Read more

explore Keep Exploring

How is AI spending different from cloud spending, and what challenges does that create for measuring and attributing AI “intelligence” (token usage, model choice, prompts, multi‑agent behavior) to business value? add

Why is AI/agent governance more important than cloud governance, and what factors (e.g., security, data leaks, agent lifecycle, attribution, and handling of confidential or personal information) should such governance address? add

How should enterprises approach hybrid/multi‑cloud AI architectures — including on‑prem vs cloud tradeoffs, data sovereignty and security, cost allocation across environments, use of multiple models/providers (multi‑agent/multi‑modal), and the choice between SaaS and headless deployments? add

How can one gain the credibility to say which technology investments (cloud, SaaS, AI, etc.) a company should make? add

bolt Powered by CUBE AI

Sabapathy Arumugam, Corestack

search

John Furrier

>> Welcome back over to theCUBE's livestream here in San Diego for FinOps X 2026. I'm John Furrier, your host of theCUBE with my co-host Paul Nashawaty for theCUBE Research. We're breaking down the AI value, AI for FinOps and all that's in between is driving a lot of the value and also measuring the cost. Tokens are here. Tokenomics is the new foundation. It's all kind of expanding superfast. Saba Arumugam is here, Chief Technology Officer at CoreStack. Saba, thank you for coming on.

Sabapathy Arumugam

>> Thank you for having me here.

John Furrier

>> Great to have a CTO on, because there's a lot of engineering conversation. My first question is on the AI value, the theme this week is it's just an engineering culture, rising superfast, building the tools and platforms to enable the ability to manage the chaos that is costs from cloud growth, but now you got the AI growth.

Sabapathy Arumugam

>> Yeah, absolutely. I think the industry is seeing the enormous growth in cloud in the last 10 years, but in the last four years we see the AI is actually taking the world by storm. In FinOps for cloud, it's all about optimizing the spend on compute, network, storage. We know what applications that we are running and we know what is the value of the application. So it is somewhat relatable to see what is the investment that we are making for an application and what is the investment that we are making for cloud. But when it comes to AI, we are paying for intelligence. We are not paying for compute here. We are paying for intelligence and the intelligence is measured in terms of token. Agents, inferencing, all of that. So what is the AI intelligence that you're getting and what is the business value of the intelligence is the biggest problem the industry is facing today. If we have to get the value of the intelligence, it needs to provide the business value.

Paul Nashawaty

>> Yeah, yeah.

Sabapathy Arumugam

>> Right? So it's all about getting the attribution of the intelligence to the business value.

John Furrier

>> How is it different? How is the AI spend different than cloud in this new era? Managing tokens, for instance.

Sabapathy Arumugam

>> Yeah. I mean the foundation models are emerging and there are multiple models actually keep coming and what model to use for what kind of intelligence that you need. You can easily spin up your prompt and then it may cost $100. The value of the output, is it worth $100? That is the biggest challenge that is going on. And when you are running a multi-agentic model, you can have as many number of prompts, but what prompts, what model that is being used and what is the memory that it uses and what is the output that we are getting. I think that's the biggest challenge that is going on with the current trend.

John Furrier

>> Yeah, good.

Paul Nashawaty

>> Yeah. So when we're looking at the differences ... I like where you were going with that because it kind of changes up like the bridging the gap between the old and the new, right? It's kind of like that conversation. But when we think about that, managing these AI spend, the tokens, et cetera, everything comes along that you were just talking about. It is different than managing traditional cloud or traditional data center FinOps opportunities.

Sabapathy Arumugam

>> Absolutely.

Paul Nashawaty

>> Previously, and even with this event, this event has evolved over the years. We got an understanding of cloud control, cloud cost control. We understand that. We know and if you're not doing it, you can kind of figure out how to do it pretty easily. This is different though.

Sabapathy Arumugam

>> Correct.

Paul Nashawaty

>> What's different about it?

Sabapathy Arumugam

>> The difference is basically in a SaaS world, you're paying for subscription.

Paul Nashawaty

>> Yes.

Sabapathy Arumugam

>> In an AI world, there is a subscription and there is a token. When you are consuming your token, what kind of ... There are multiple applications trying to consume this API endpoints and then the token consumption happen. Or if a developer using AI to develop and they're basically getting the context of the entire code, doing a code generation and all of that. So how much of token consumption happening for each of the use cases and how do we track the token efficiency against the output that we are getting? It is all about the key performance indicator. What is the outcome that we get? It's all about business outcome rather than the intelligence.

Paul Nashawaty

>> No, I agreed with that. But what do you think about ... Now, this is something that's been kind of a theme that's been going on for the last couple of days at least, is the fact that there's a different cost that should be associated for different token usage. For example, innovation versus production, right?

Sabapathy Arumugam

>> Correct.

Paul Nashawaty

>> And you want to be able to innovate so you can add to production, but that cost of innovation, it should be a development or a sandbox environment. It shouldn't cost the same. Would you agree with that?

Sabapathy Arumugam

>> Absolutely. When it comes to production, we need to attribute it to the application and what is the business value of the application. But when it comes to innovation, we should not be stopping the acceleration of innovation, but at the same time there are too many experimentation happens, right? Yeah. What experiments provide what value, I think that is the biggest gamble that happens, right? Everybody wants to use AI and everybody wants to create business value on AI. And for that, there are multiple experiments happen. Which experiments provides the value. I think that's the biggest question that is going on. So I think at an enterprise level, there needs to be attribution for production workload separately. For innovation and research, there needs to be a separate attribution and how much of budget that is required for innovation and how much are we spending on the AI and what is the outcome of the experiment that needs to be attributed, right?

John Furrier

>> Saba, I want to ask you about the cloud AI transition in context to governance. We saw with data lakes, governance was a real enabler, a lot of agentic activity happening. Agents will be big in FinOps and tokenomics. What's the role of governance specifically on how that ties in with FinOps? Because we hear tagging, attribution you just mentioned, governance and identity, all these things matter. How do you look at that as a CTO for this world? Because a lot of build out's coming fast. What's the role of governance?

Sabapathy Arumugam

>> I think we feel that AI governance or agentic governance is way more important than a cloud governance. They're more important than just actually looking at a cost. The reason is actually when we are looking at agent, there is a non-human identity being created. I think every human can create, go and create multiple agents and we basically bring in lot of data as a context to the agent, right? So in terms of governance for AI, we need to look at security, we need to look at the data leaks, we need to look at the life cycle of the agent, we need to look at the attribution of the agent. So the entire things need to be looked at to governance. Basically, the cost for AI is just one element of it, but when we look at governance, it includes security, data leaks, attributions, everything actually. That's what governance means. You need to be able to reason why you are running the agent and what data that you're giving as a context to the agent and are we giving any confidential information? Are we giving any personal information? All the time we see that we basically put some credit card information, right? How do we avoid such data leaks being provided as a context? Some of the passwords, we should not be sending a passwords to the agent. So these are all the elements that needs to come under the governance umbrella.

Paul Nashawaty

>> Yeah. So it comes back to the conversation of what's suitable when it comes to FinOps, what's suitable for AI and where you need human judgment to kind of be put in place, right? Because some things are just not ready for AI. So I think that's where you were going with that. From a governance perspective, you still need to have a human in the loop to make sure things are working. Is that accurate?

Sabapathy Arumugam

>> Absolutely. I think the human in the loop is very, very important for AI. AI need to be looked at for intelligence, analysis, reasoning, but the decision making needs to be with the human. And I think where AI helps is providing the recommendation along with the reasoning for human to take decision faster. It's all about decision making and AI helps in accelerating the decision making for the human, but the human needs to take a final decision based on the business contact, because the human knows what is the business is all about, a lot more context that human have. Yeah, LLMs are trained with a lot of data, but the business context the human has so that AI provides, based on all the context, it provides the reasoning and the human takes the final decision based on the business context.

John Furrier

>> That brings up a lot of the conversations we've been having. It's been cloud and AI have been two key things, the data. But with the explosion of tokens, the on-premises is now impacted. But it's hybrid computing, so it's distributed. Control planes are becoming important, but to get the cost across the environments, talk about the multi-environment cross connection, because I can have a lot of cloud. I can have now a lot of on-prem because a lot of data there too. You mentioned unlocking that proprietary or unique data, that's the crown jewels, that's the value and the users are adopting. So it's like a perfect storm for value creation, but now you got, "Okay, where's the costs?" Do I bring it on prem? Do I have an AI factory? So there's a lot of cross environment and cross use case, SaaS versus headless. What's your thoughts on those two things?

Sabapathy Arumugam

>> I think the AI is accelerating this multi-cloud, multi-modal hybrid environments to life and it needs to be, right? So currently, every AI application that I built now is a multi-agentic model. The reason why multi-agentic models are built is actually each model provides a different efficiency, right? For some use cases, cloud gives a better efficiency, OpenAI gives better efficiency, or a Gemini gives better efficiency. So when the applications are built, they need to be able to leverage the different model for the different use cases that they are building. There is model on the one side and the data security is on the other side, right? When organization looks at the confidentiality of the data, they want to maintain certain data within their network. That's where the hybrid computing comes into play. I think now the neoclouds are coming, the neo clouds are primarily coming for the data sovereignty of the enterprises. The enterprises want to maintain their own data within their infrastructure and use the compute from a hyperscalers, but my data resides within my GPU infrastructure. And there are SaaS applications that are being used. So basically, the combination of SaaS, the neoclouds, the private environment, all of them plays a major role in getting the value.

John Furrier

>> Okay. I have to ask you, because I love where this is going, because I want you to explain the difference between data management in the old way, because we've seen governance in data management, but I won't say niche, but in that segment, governance is now mainstream because of agents and the value it's creating. What's the contrast and difference between data management of yesterday and today?

Sabapathy Arumugam

>> I think that's a good question. Actually, earlier the data management is all about the data that I have in my database, the data that I have in my file system, right? Now, the data governance is about what is the data that I am feeding to my agents, right? Earlier, a lot of this context actually remind within the enterprise infrastructure. Now, every user in an organization have access to the agent. So the context that we are giving to the agent is not only coming from the files, not only coming from the databases, the user can upload all of this information, right? How do we protect the data that is actually flowing to the agent from the end user? There is application wise data governance, there is a end user-wise data governance. The end users can put any of the enterprise data into the cloud. How do we look at this data and how do we avoid the leak is one of the major problem on the data governance side.

John Furrier

>> That's where intelligence comes in handy here, because agents are basically workers. So imagine data management having more customers.

Sabapathy Arumugam

>> Correct. Exactly.

John Furrier

>> That's what's happening.

Sabapathy Arumugam

>> Absolutely.

John Furrier

>> So you got to have a different mechanism, different intelligence, they're human triggered, machine triggered, a lot more data management and activities.

Sabapathy Arumugam

>> Absolutely. Absolutely. I mean, the value of the agent depends on what the context. The context resides in the data. And I think what data that we feed into the agent plays a major role in getting the recurred intelligence and at the same time not compromising the security and feeding all the data to the agent. Yeah.

Paul Nashawaty

>> Yeah. But this is also related to when you have ... This is a board level conversation because there's financials involved, there's lots of ... And there's also discussions about like, should execs are asking, should we invest more in AI? What we see in our own research is 25% of IT's budget in 2025 was dedicated to AI projects. Okay, that's fair. That was about experimentation and innovation, right? But 26 a year of implementation, right? Now, organizations are going, "Should I invest more?" And from a FinOps perspective, where can the response be from related to FinOps, related to the FinOps practices? How do you get the credibility to say, "Yes, this is what you should be investing in"?

Sabapathy Arumugam

>> The FinOps foundation is moving in the right direction of not just looking at the business value for cloud, it's a business value for the technology investment that includes cloud, SaaS and AI. I mean, that's the direction FinOps is going and that's what the industry needs. From an exec standpoint, what is very important is the visibility of where we are spending the money and the attribution to the various functions that includes what is my spend on production, what is my spend on research, what is my spend on developments? The attribution is very important. Then the third thing is what is the business outcome for this technology investment, right? If we are spending this 20% on AI, what is the business outcome that we are getting from this 20%? If the company's revenue reminds the same, what is the value of this 20% investment, right? So end of the day, the budget that you're allocating for AI need to provide the recurred, the revenue growth for the enterprises, right? How do we measure the business outcome for the AI experimentation, the AI spend that we are doing is the most important factor for the FinOps practitioner. And the practitioner need to roll up this data and then provide to the exec for taking the right decision on where to invest to get the recurred value. The AI spend can be on various things. You can be on the research, it can be on the development, of core development, or providing certain output from a business standpoint. Where do we spend this budget to get the recurred business outcome? That data needs to be provided to the execs. Yeah.

John Furrier

>> Saba, great to have you here. Thanks for sharing that great insight. Love the CTO perspective. Governance, the center of the value proposition, data management's exploding on mainstream real time. More users, more agents.

Sabapathy Arumugam

>> Yes, absolutely.

John Furrier

>> Different behavior patterns. Thanks for sharing.

Sabapathy Arumugam

>> Thank you. Thank you.

John Furrier

>> Really appreciate it.

Sabapathy Arumugam

>> Absolutely.

John Furrier

>> I'm John Furrier, with Paul Nashawaty with theCUBE. We're getting all the data here. We're like agents of data here on theCUBE at FinOps. Hopefully it's on the money. Thanks for watching.