FinOps X 2026 | News Break with J.R. Storment, FinOps Foundation & Nishant Gupta, Salesforce

Clips
More from FinOps X 2026

J.R. Storment

Executive Director

FinOps Foundation

Nishant Gupta

EVP, Engineering

Salesforce

play_circle_outline Managing AI token costs and rising tokenomics discipline

play_circle_outline FOCUS 1.4 spec for harmonizing billing and cost data

play_circle_outline From Subscriptions to Metering: Managing TCO with Token Telemetry, Observability, and Accurate Attribution

play_circle_outline Rapid accelerated adoption of tokens versus prior cloud FinOps timeline

play_circle_outline Practitioner guidance: experiment, share best practices, build observability

Info

News Break with J.R. Storment, FinOps Foundation & Nishant Gupta, Salesforce

J.R. Storment

Executive Director FinOps Foundation

Nishant Gupta

EVP, Engineering Salesforce

search

John Furrier

>> Welcome back to theCUBE's live stream here in San Diego for FinOps X 2026. I'm John Furrier, host of theCUBE, with Paul Nashwaty. The big story here is the rise of tokenomics, and every single company will start thinking about the effectiveness of tokens, AI factories, and providing that innovation. We had two great guests here. We have the executive director of the foundation as well as a leading practitioner, J.R. Storment, VP and GM of the Linux Foundation as well as the executive director of the FinOps Foundation. They announced a new foundation, Tokenomics Foundation. We'll get into that. And Nishant Gupta, chief availability officer of Salesforce. We want more tokens available, highly available, high availability of tokens. J.R., Nishant, thanks for coming on here. Great to see you again.

J.R. Storment

>> Thanks. Good to see you, John.

John Furrier

>> So keynote was great, great presentation. I love how you had practitioners come on and people in the community talking about their view of this waiver on, because FinOps has evolved with cloud. Now you have the AI wave coming in. Really strong messaging around this future is directionally going this direction, but it's still evolving. That was the key theme of the keynote. Please elaborate.

J.R. Storment

>> Well, I've been listening closely to the community's last six months, and one thing became really clear, is that the work of managing AI spend is really complex. There's a lot of overlaps with what FinOps does, FinOps for AI, over the years. But as we talk to more and more advanced vendors in this area, we hear that managing tokens and the whole tokenomics stack, there's a whole new set of skills and capabilities that you have to get into. And it's a bigger sphere of influence from the token side all the way out to the business models on the far end. So while we're seeing there's a ton of overlap, and we don't know where the lines are yet, we realized we needed to do a deep dive into defining this new discipline of tokenomics.

John Furrier

>> Nishant, you guys at Salesforce are very active on the agentic. Agentforce, obviously, top line. I think the conference is maybe renamed Agentforce. Dreamforce, now Agentforce. Key role of data with agents are going to be big, generative AI. Talk about your role in the community and your new title, the chief availability officer. I kind of teased out, I guess, a storage paradigm of highly available, high availability, but this is happening.

Nishant Gupta

>> Yes. The title is not too far away from that. My job is to make sure that our services are highly available for our customers, and we take that role very seriously. A lot of our customers run mission-critical applications on Salesforce, and we want to make sure the platform is highly available. So the focus of my role to a large degree is making sure that we are very reliable and available. The other aspects that's within my purview, and that's the practitioner piece, is managing all of the Salesforce infrastructure budget and how we can find the optimizations. And then in the last year or so, a lot of significant focus has been on managing the token spend and finding optimizations around that.

John Furrier

>> We've been covering a lot of Salesforce over the years, the harmonization layer, the data cloud, they call it a new name now, but there's a lot of stuff going on in the Salesforce layers.

Nishant Gupta

>> Yes.

John Furrier

>> You guys have been doing this for years. Now you have this model routing concept you talked about on stage, J.R. A lot of the same things that we've seen in other waves are hitting. What's different? What's changed?

Nishant Gupta

>> I think the key thing that's different from my perspective, compared to traditional FinOps practice, is just the completely different nature of how to think about tokens. It's an abstract quantity. It's very hard to relate tokens to a business outcome. It's very hard to relate input and output tokens in a predictable manner. It's hard to relate how many of the regular reads versus cash reads are going to happen. And that all impacts the economics of it very significantly. So we need to have fundamental different ways of measuring it, getting better telemetry and observability around it. J.R. mentioned that during the keynote. And then the squeeze is real between what the CEOs are asking and what the CFO is asking, and you're sort of in the middle, and you got to make sense of it.

Paul Nashawaty

>> Yeah. Nishant, that's a very important point here. What we've seen, and I saw it on the keynote as well, J.R., I really thought it was amazing, the energy in that room was wild. It was crazy.

J.R. Storment

>> Fun.

Paul Nashawaty

>> Yeah, it was very cool. But one of the things that's happening is, FinOps leaders are really kind of pivoting. They're looking at it more from a holistic view across the entire ecosystem and not so much just the infrastructure. Can you talk a little bit about that because you touched on that, Nishant, and maybe you can piggyback on that?

J.R. Storment

>> I can start with something, which is, I would tell you, Nishant is in an executive group that we have of leaders who come together and talk about the challenges for FinOps. About three months ago, we got them together to talk about the challenges of FinOps, and all the auction was taken out of the room. They needed to talk about this token cost thing. But I want to be clear on that. The token cost thing is one tiny piece of the AI cost and value puzzle. And there's a ton of work to be done still for all these big companies who are on two, three, four clouds, they have data centers. So they're all still doing FinOps, and that's still the critical bread and butter, but they're all now facing this new edge that is kind of like cloud was 10 years ago. It's the highest variable spend that's scaling. It's still not, I don't know if it is for you, the biggest line item for many, but it's going on a vertical path 10 times the speed of cloud. So while they are connected, there's also more need in that side right now.

Nishant Gupta

>> Yeah. I think what I call is a token-adjacent spin, and we touched upon a little bit today. If you think about the amount of data that needs to be processed, what observability systems you're going to have, more CI/CD pipelines, if you're writing more code, more test harnesses, and all of that stuff, that's going to increase your cloud spend. So that's one dimension. And then we are also seeing a renaissance of the data centers, people running to hold their own data centers, and then the focus on economics of data centers as well in addition to the cloud. So I think there is combination of few things happening here.

John Furrier

>> The thing that jumped out at me on that point about the data centers, J.R., you actually use the word AI factories, which we've been doing a lot of coverage on. That's the big factory. You're seeing the big CapEx buildouts on the infrastructure side. That's enabling a lot of the activity. But the data center is where the engagement is, that's where the users are adopting GenAI. We're seeing the token usage. You showed some graphs there, we saw many graphs up onto the right, costs are going down. That's spawning more user growth, but all the CapEx going on in the data center has to fund those tokens. So you have this new growth in the data center, and you also brought up the edge, and then you added procurement in there. I didn't see that on my radar. So I want you to expand the dynamics, guys, between how these AI factories... We've been saying you put intelligence in the edge, you put intelligence into the networks, things happen. So a lot of that token activity is because there's value being created. It's not like, "Hey, I'm running an app in the cloud." This is like real users using data on-prem. That's what the crown jewels are.

Nishant Gupta

>> Right. Yep.

John Furrier

>> So this is now flipped over.

Nishant Gupta

>> It's flipped over, and it's gradient across various factors. I was surprised with the Accenture speaker today, $2 billion of CapEx a day. That's a mind-boggling number to think about. And then if you look at announcements coming out of Dell, I believe last week or two weeks ago, in terms of laptops and computer devices with more GPUs in it and more power on it. So you will see this proliferation of usage on the edge, on your personal devices, in the cloud, in their own data centers across the board.

John Furrier

>> There's a talk. There's a lot of narrative coming out of that Dell where you can say, "Hey, build your own token factories." You can get sandboxes built. This is where some development's going to go on. And the conversation is, "Okay, get it all done." And that's where the developers are. That's where the connection is. It's not just, "I'm going to buy, buy, buy." And then the costs are going up significantly to the companies. These surprise bills are coming out.

J.R. Storment

>> But what is fundamentally different on this wave versus the cloud wave, you mentioned developers, it's not just the developers now. In my tiny organization, Linux Foundation, our legal team is spinning a bunch. They use Gemini, I use Claude, but on top of that, to your point of edge, the difference as well is, FinOps as a discipline was optimizing, allocating all these things, resources that exist in typically clouds or data centers, but the demand is so much greater now across not just the engineers, but across our org, that yes, token factories on the desktop, token factories in space, token factories in all those places. We're going to see all those things emerge.

John Furrier

>> Now, on that point, I want you to finish, Paul, one second. Jensen, at GTC, showed the Pareto curves with Vera Rubin. You're starting to see premium tokens. You're starting to see levels, it's almost a portfolio of a product map. It's like, "Okay, entry level, run this query over here." So you're starting to see now the evolution of premium... This brings up FinOps, Tokenomics Foundation. So the world's moving so fast, it's already accelerated to levels of service.

Nishant Gupta

>> Level of service, as well as the level of GPUs, right? People are not retiring the 100s from 40 years ago because you can use them for smaller inferences in certain areas, whereas newer chips like Vera Rubin are mostly used for training purposes. So there's a combination across the board. To J.R.'s point, yes, everybody's a builder now, right? So whether you are at the engineering team or you're in the legal team or the sales team, everybody's a builder.

John Furrier

>> By the way, finance too are building too.

Nishant Gupta

>> Finance shadow.

John Furrier

>> Shadow AI has hit every department.

Nishant Gupta

>> Yes.

Paul Nashawaty

>> Well, I mean, it's very clear in the State of FinOps report, J.R. We're looking at it's moved far beyond cloud, right? It's moving into AI, SaaS, licensing, data platforms, all the different pieces, and the data center that John was talking about. But let's move on to another topic that was kind of out of the keynote. When looking at FOCUS 1.4, that was a big kind of announcement, that was big release that's happening here. It's a common language challenge, the harmonization across all these different platforms whether you're on-prem or in the cloud. Let's talk a little bit about that.

J.R. Storment

>> So the FOCUS spec was originally designed to normalize cloud billing data. What we saw over the few years since it's launched is that people are looking to get consistent cost and uses data across all the places they're consuming and paying, and that challenge is going to accelerate incredibly with tokens. The FOCUS 1.4 that's being announced tomorrow is going to help solve some of the data integrity, some of the deeper areas of how we reconcile and invoices together. But down the road, as we start to look at token consumption, we're not just talking about hundreds of millions of rows of billing data. We're talking about trillions of tokens. Take that to allocations per user, per session, per trace, this becomes a telemetry problem, an observability problem that now these FinOps teams, engineering teams, finance teams, procurement teams need to look at this in a totally new way.

Paul Nashawaty

>> One of the interesting things, my read on the keynote and all the pre-announcements and such is, there is a maturity level when it comes to FinOps, right? There are some organizations that are very mature, and they understand it very much. And then there are some that are like almost analysis paralysis, and they don't know how to get started. I think that's something that when I was looking at what the FOCUS release was coming out, one of the things that was interesting to me was it was frictionless. It was allowing for moving away from the individual spreadsheets, the CSV files, to all these different things and compiling it yourself, to more of a unified view that was frictionless. What we're seeing in our own research is, 67% of organizations are hiring generalists over specialists. So if they're not hiring specialists to do this work, then it's really hard to actually pull all the... I can have data-

J.R. Storment

>> Common languages.

Paul Nashawaty

>> Right. So this common language made a lot of sense. That was where my read was.

J.R. Storment

>> I think what we're finding is, the billing data, the focus data, the cost and uses data, it's really just about getting that TCO and the denominator in the value equation. Now, one of the great things about agentic and agentic costs is, agents, in theory, are going to do business tasks that we can now get to better unit economics, and we can say this thing at this outcome at this cost. So I think we're going to see that cascade further.

John Furrier

>> I mean, TCO's hard when it's a moving train, which it is now. Your slide you put up on the keynote, I thought, was the bold slide you put up at the beginning, AI subsidy error is ending, and #ThanksTokenMaxxing was the hashtag. But this really talks to... And again, Salesforce had a news this morning around buying a company called m3ter, and the metering era is upon us. The subscription era moving to metering, is this a tell sign to what needs to happen on the infrastructure side? Because everyone's talking about token costs and budgets because they're seeing the bills. But the underlying infrastructure, when you have agents running, you're running operations, the total cost of ownership goes really sideways when you screw up. So you're like, if you don't engineer the system, the token costs are pale in comparison to the TCO of bad design, bad use of resource. That's kind of like under the iceberg. See an iceberg under the water, what don't you see? So there's icebergs ahead maybe, or what's your guys' thoughts on this?

Nishant Gupta

>> I think it is certainly there. The total cost of ownership conversation with AI, it's only an extension. We need to understand that. I was having a conversation with a colleague over here yesterday, and we were talking about, should we be tagging prompts much like we tag resources or should we be tagging the tokens? And it is all about the attribution. I think we talked about... In the keynote, a lot of people touched upon what's the outcome, and how do you demonstrate the value of AI? In order to demonstrate the value, you need to be able to tag it, you need to be able to attribute the usage and say, "This particular agent or this particular service is spending so much money on the tokens. Are we getting the right value from it or not?"

John Furrier

>> Yeah, Paul brought that up with FOCUS. Yesterday we had many pregame conversations around FOCUS and the value. What's the valuation? I mean, I can almost see the finance spreadsheet folks thinking, "Okay, how do I value a project? What's the payback look like? Do I focus on cost or value?" So guys, what's that new narrative? What's that vision or direction of scoping the value proposition?

J.R. Storment

>> I hate to be the wet blanket, because FinOps and AI and all these things should be about value and intelligence. But the problems we're hearing right now from the big token spenders are not even, how do I get to the value, but how do I just allocate the spend? How do I understand the spend? Everybody's blowing token budgets. The SVPs and CFOs are like, who's responsible, which project, where does that need to go? So I mean, that's I think the fundamental challenge right now.

John Furrier

>> What are you hearing? Stack rank some of the challenges from that token tsunami cost.

J.R. Storment

>> I would summarize it just in, do I have a giant bill from AI provider, who's responsible, and what are they using it for?

Paul Nashawaty

>> Chargeback model.

J.R. Storment

>> I think chargeback gets to invoicing, and I'm just showing it and understanding so I can go talk to that person and say, "Should you be doing this?" But then ultimately, yes, that's all a building block to get to how do we show the value on it.

John Furrier

>> Companies are struggling today. What you're saying is that where did this come from?

J.R. Storment

>> Yeah. Yeah.

John Furrier

>> Who's using those tokens?

J.R. Storment

>> Right, right.

Nishant Gupta

>> There was an article in Wall Street Journal, I believe, a couple of days ago which says 25% of the companies have good visibility.

John Furrier

>> That's a stretch maybe.

Nishant Gupta

>> And more than 25% of the companies only look at their stuff after the bill comes. 30 days is too late. There was another article where somebody spent $500 million in a month. If you're going to be spending that kind of money or anywhere near that, you want to know this on a daily basis as opposed to 30 days later.

John Furrier

>> I interviewed a CIO, I asked him, "How do you value the project?" He goes, "We look at very pragmatic as this early days, trying to understand what's going on." He's like, "If we have the cost of tokens, what's the human equivalent cost?" So now they're like, "I don't really need that if the cost to do tokens for AI but is greater than the human cost unless there's a major needle moving moment."

J.R. Storment

>> There are two really interesting conversations I'm hearing from engineering leaders, and frankly, up from the CEOs as well. The first one is, how do we allocate our token budgets to give the right people tokens? And token is beginning to be a bonus thing, engineers wanting to get more leverage. Do you give 10 actors more? The other side of this is, of course, the question of, well, which of the work should humans even do and how do we balance that? And that's where tokenomics is a discipline, and I think is an emerging discipline because it is yet to be fully defined. It's going to start to hit more into those labor implications and the pricing model implications as we see that.

Paul Nashawaty

>> Yeah. I think that that goes back to your FinOps' move into the boardroom conversation, because I mean that really comes down to what is going to be changing for skillsets in order to understand how to actually execute appropriately. I mean, to John's point, I was talking to one organization, they were measuring the success of their employees by the number of tokens they were spending, which was crazy, right?

J.R. Storment

>> Those are the leaderboards that everybody's moving away from.

Paul Nashawaty

>> Right. So they have to move away from that.

John Furrier

>> I was talking to someone yesterday in the hallway here, and the conversation was we're spending more tokens figuring out who's using what tokens. So again, tokens to figure out tokens. So there's a lot of growth. The other conversation we had here on theCUBE was, the cloud way was a 10-year cycle on FinOps, and this was a shorter time series now, going on now, highly compressed accelerated adoption, accelerated telemetry. So a lot of those kind of... How are you guys seeing that? How's that playing out into FinOps Foundation and now the new token foundation?

J.R. Storment

>> The shift has been... For a lot of years, FinOps was a conversation at that VP, SVP, EVP level of trying to understand the costs. Now, tokens, and whether that be FinOps tokens or tokenomics as we emerge these things, is suddenly a CEO level conversation. I was at a private forum with some global CIOs at Fortune 25s a couple months ago, and I heard them saying words like, "This is an existential crisis for us." And some of these are 100-plus-year-old companies, right? So it's pretty fascinating to see how that's happening. But I think what we're going to see change in all this is, we've got to build on the tenants that we've been building in FinOps land for years, visibility and accountability, and pushing data to the edge of the engineers to make decisions, but we do have to do it 10 times as fast and with a much smaller unit now. But in the end, it is going to force the value conversation much sooner than it came before.

John Furrier

>> Nishant, I want to ask you because we're seeing two things going on at the same time, the growth of the FinOps and token cost side, a lot of usage, but the AI infrastructure is also accelerating where you can get more. You mentioned the H100s inference, you're starting to see different cluster sizes. You don't need the big rack scale systems for certain use cases, especially in the data center. So you got this AI evolution on the infrastructure side booming. The convergence, where does it connect? What's your vision on that?

Nishant Gupta

>> I think that's where you have the G1's paradox, right? So you're seeing some of the older models and they're much cheaper now because they can run on older hardware and you have the newer models which are more expensive, but the growth is accelerating. I see this as an example of... And I do not know if it is really... There's a narrative around the inflationary aspect of the AI. I don't see it that way right now. I'll give you an example. The cost of your iPhone is the same over the last 10 years. You can buy TV for 1,000 bucks from Costco and you could buy a TV for 1,000 bucks 10 years ago as well. So your spend remains the same. You get better value and better product from it, but nobody's going to be using a GPT 3.0 model anymore for any real work at this point. So I think that remains to be seen in terms of how we make the mix of it. The other challenge is, we talked a little bit about model routing on the use cases. I think that's still a maturing practice to be able to better understand what problem sets for what models and what price point are the right things to do. I don't think we have fully cracked that thing yet.

John Furrier

>> Guys, great to have you on. Great keynote. Final question, J.R., for you is, on the foundation, what should people know about the change with FinOps and now the Tokenomics Foundation? What's the big change that they should know about here in this era?

J.R. Storment

>> The most important thing is that the FinOps Foundation is to keep doing everything it's been doing. We're keeping a single consolidated practitioner community. We're keeping doing our working groups and our community calls and our virtual and regional events. We're going to expand that sphere of influence to touch on some new areas, and we're going to bring a new part of this conversation into it, which is the tokenomics side. There's going to be a dedicated foundation for this that goes deep into working groups and defining the primitives of this, but we're not going to separate these things out in communities. The people are going to intersect and align. We can't stop doing the FinOps work, as Nishant said to me earlier. We still got to do the cloud FinOps, and we also need to go do the tokenomics work.

John Furrier

>> Decoupled, but highly cohesive, but connected.

J.R. Storment

>> Exactly. We'll share some operational resources, but dedicated SME teams.

John Furrier

>> Nishant, for practitioners that are trying to frame the problem statement, what's your advice? How should they lean into this? Because remember, when you lean into these emerging trends, there's downstream benefits, maybe a little pain and suffering, but if you get it right-

Nishant Gupta

>> Yeah. So with the FinOps side, we had best practices for the last 20 years that you can rely on. Nobody has best practices for the last six months. So you got to blaze the trail in some sense. Our focus on a practitioner side and a huge focus in the amount of money that we are spending on tokens is going to be around understanding the trends, understanding the usage, the patterns, and how best to optimize. We don't have the luxury to wait on somebody else to tell what the best practices are. I think we are somewhat racing the trail on that.

John Furrier

>> The phrase eating your own dog food, drinking your own champagne, client zero was kind of like regulated for a few firms, but that every company has to do that now. They got to roll their sleeves up.

Nishant Gupta

>> I think doing that in vacuum as an experiment is okay. And that's why I'm excited about tokenomics. But we need to bring those best practices and share it broadly within the industry, not only from a practitioner standpoint, but also from different tier models standpoint. Those who are provider of this data and visibility and insights, they need to be part of the conversation.

John Furrier

>> It's a great time to be an engineer. It's a great time to be an AI builder. J.R., Nishant, thank you for coming on theCUBE.

Nishant Gupta

>> Thank you very much.

J.R. Storment

>> Thank you, John. Thank you, Paul.

John Furrier

>> Kicking off the day here. I'm John Furrier, host of theCUBE, with Paul Nashawaty, my co-host with theCUBE Research. We'll be right back after this short break.

News Break with J.R. Storment, FinOps Foundation & Nishant Gupta, Salesforce

search

John Furrier

>> Welcome back to theCUBE's live stream here in San Diego for FinOps X 2026. I'm John Furrier, host of theCUBE, with Paul Nashwaty. The big story here is the rise of tokenomics, and every single company will start thinking about the effectiveness of tokens, AI factories, and providing that innovation. We had two great guests here. We have the executive director of the foundation as well as a leading practitioner, J.R. Storment, VP and GM of the Linux Foundation as well as the executive director of the FinOps Foundation. They announced a new foundation, Tokenomics Foundation. We'll get into that. And Nishant Gupta, chief availability officer of Salesforce. We want more tokens available, highly available, high availability of tokens. J.R., Nishant, thanks for coming on here. Great to see you again.

J.R. Storment

>> Thanks. Good to see you, John.

John Furrier

>> So keynote was great, great presentation. I love how you had practitioners come on and people in the community talking about their view of this waiver on, because FinOps has evolved with cloud. Now you have the AI wave coming in. Really strong messaging around this future is directionally going this direction, but it's still evolving. That was the key theme of the keynote. Please elaborate.

J.R. Storment

>> Well, I've been listening closely to the community's last six months, and one thing became really clear, is that the work of managing AI spend is really complex. There's a lot of overlaps with what FinOps does, FinOps for AI, over the years. But as we talk to more and more advanced vendors in this area, we hear that managing tokens and the whole tokenomics stack, there's a whole new set of skills and capabilities that you have to get into. And it's a bigger sphere of influence from the token side all the way out to the business models on the far end. So while we're seeing there's a ton of overlap, and we don't know where the lines are yet, we realized we needed to do a deep dive into defining this new discipline of tokenomics.

John Furrier

>> Nishant, you guys at Salesforce are very active on the agentic. Agentforce, obviously, top line. I think the conference is maybe renamed Agentforce. Dreamforce, now Agentforce. Key role of data with agents are going to be big, generative AI. Talk about your role in the community and your new title, the chief availability officer. I kind of teased out, I guess, a storage paradigm of highly available, high availability, but this is happening.

Nishant Gupta

>> Yes. The title is not too far away from that. My job is to make sure that our services are highly available for our customers, and we take that role very seriously. A lot of our customers run mission-critical applications on Salesforce, and we want to make sure the platform is highly available. So the focus of my role to a large degree is making sure that we are very reliable and available. The other aspects that's within my purview, and that's the practitioner piece, is managing all of the Salesforce infrastructure budget and how we can find the optimizations. And then in the last year or so, a lot of significant focus has been on managing the token spend and finding optimizations around that.

John Furrier

>> We've been covering a lot of Salesforce over the years, the harmonization layer, the data cloud, they call it a new name now, but there's a lot of stuff going on in the Salesforce layers.

Nishant Gupta

>> Yes.

John Furrier

>> You guys have been doing this for years. Now you have this model routing concept you talked about on stage, J.R. A lot of the same things that we've seen in other waves are hitting. What's different? What's changed?

Nishant Gupta

>> I think the key thing that's different from my perspective, compared to traditional FinOps practice, is just the completely different nature of how to think about tokens. It's an abstract quantity. It's very hard to relate tokens to a business outcome. It's very hard to relate input and output tokens in a predictable manner. It's hard to relate how many of the regular reads versus cash reads are going to happen. And that all impacts the economics of it very significantly. So we need to have fundamental different ways of measuring it, getting better telemetry and observability around it. J.R. mentioned that during the keynote. And then the squeeze is real between what the CEOs are asking and what the CFO is asking, and you're sort of in the middle, and you got to make sense of it.

Paul Nashawaty

>> Yeah. Nishant, that's a very important point here. What we've seen, and I saw it on the keynote as well, J.R., I really thought it was amazing, the energy in that room was wild. It was crazy.

J.R. Storment

>> Fun.

Paul Nashawaty

>> Yeah, it was very cool. But one of the things that's happening is, FinOps leaders are really kind of pivoting. They're looking at it more from a holistic view across the entire ecosystem and not so much just the infrastructure. Can you talk a little bit about that because you touched on that, Nishant, and maybe you can piggyback on that?

J.R. Storment

>> I can start with something, which is, I would tell you, Nishant is in an executive group that we have of leaders who come together and talk about the challenges for FinOps. About three months ago, we got them together to talk about the challenges of FinOps, and all the auction was taken out of the room. They needed to talk about this token cost thing. But I want to be clear on that. The token cost thing is one tiny piece of the AI cost and value puzzle. And there's a ton of work to be done still for all these big companies who are on two, three, four clouds, they have data centers. So they're all still doing FinOps, and that's still the critical bread and butter, but they're all now facing this new edge that is kind of like cloud was 10 years ago. It's the highest variable spend that's scaling. It's still not, I don't know if it is for you, the biggest line item for many, but it's going on a vertical path 10 times the speed of cloud. So while they are connected, there's also more need in that side right now.

Nishant Gupta

>> Yeah. I think what I call is a token-adjacent spin, and we touched upon a little bit today. If you think about the amount of data that needs to be processed, what observability systems you're going to have, more CI/CD pipelines, if you're writing more code, more test harnesses, and all of that stuff, that's going to increase your cloud spend. So that's one dimension. And then we are also seeing a renaissance of the data centers, people running to hold their own data centers, and then the focus on economics of data centers as well in addition to the cloud. So I think there is combination of few things happening here.

John Furrier

>> The thing that jumped out at me on that point about the data centers, J.R., you actually use the word AI factories, which we've been doing a lot of coverage on. That's the big factory. You're seeing the big CapEx buildouts on the infrastructure side. That's enabling a lot of the activity. But the data center is where the engagement is, that's where the users are adopting GenAI. We're seeing the token usage. You showed some graphs there, we saw many graphs up onto the right, costs are going down. That's spawning more user growth, but all the CapEx going on in the data center has to fund those tokens. So you have this new growth in the data center, and you also brought up the edge, and then you added procurement in there. I didn't see that on my radar. So I want you to expand the dynamics, guys, between how these AI factories... We've been saying you put intelligence in the edge, you put intelligence into the networks, things happen. So a lot of that token activity is because there's value being created. It's not like, "Hey, I'm running an app in the cloud." This is like real users using data on-prem. That's what the crown jewels are.

Nishant Gupta

>> Right. Yep.

John Furrier

>> So this is now flipped over.

Nishant Gupta

>> It's flipped over, and it's gradient across various factors. I was surprised with the Accenture speaker today, $2 billion of CapEx a day. That's a mind-boggling number to think about. And then if you look at announcements coming out of Dell, I believe last week or two weeks ago, in terms of laptops and computer devices with more GPUs in it and more power on it. So you will see this proliferation of usage on the edge, on your personal devices, in the cloud, in their own data centers across the board.

John Furrier

>> There's a talk. There's a lot of narrative coming out of that Dell where you can say, "Hey, build your own token factories." You can get sandboxes built. This is where some development's going to go on. And the conversation is, "Okay, get it all done." And that's where the developers are. That's where the connection is. It's not just, "I'm going to buy, buy, buy." And then the costs are going up significantly to the companies. These surprise bills are coming out.

J.R. Storment

>> But what is fundamentally different on this wave versus the cloud wave, you mentioned developers, it's not just the developers now. In my tiny organization, Linux Foundation, our legal team is spinning a bunch. They use Gemini, I use Claude, but on top of that, to your point of edge, the difference as well is, FinOps as a discipline was optimizing, allocating all these things, resources that exist in typically clouds or data centers, but the demand is so much greater now across not just the engineers, but across our org, that yes, token factories on the desktop, token factories in space, token factories in all those places. We're going to see all those things emerge.

John Furrier

>> Now, on that point, I want you to finish, Paul, one second. Jensen, at GTC, showed the Pareto curves with Vera Rubin. You're starting to see premium tokens. You're starting to see levels, it's almost a portfolio of a product map. It's like, "Okay, entry level, run this query over here." So you're starting to see now the evolution of premium... This brings up FinOps, Tokenomics Foundation. So the world's moving so fast, it's already accelerated to levels of service.

Nishant Gupta

>> Level of service, as well as the level of GPUs, right? People are not retiring the 100s from 40 years ago because you can use them for smaller inferences in certain areas, whereas newer chips like Vera Rubin are mostly used for training purposes. So there's a combination across the board. To J.R.'s point, yes, everybody's a builder now, right? So whether you are at the engineering team or you're in the legal team or the sales team, everybody's a builder.

John Furrier

>> By the way, finance too are building too.

Nishant Gupta

>> Finance shadow.

John Furrier

>> Shadow AI has hit every department.

Nishant Gupta

>> Yes.

Paul Nashawaty

>> Well, I mean, it's very clear in the State of FinOps report, J.R. We're looking at it's moved far beyond cloud, right? It's moving into AI, SaaS, licensing, data platforms, all the different pieces, and the data center that John was talking about. But let's move on to another topic that was kind of out of the keynote. When looking at FOCUS 1.4, that was a big kind of announcement, that was big release that's happening here. It's a common language challenge, the harmonization across all these different platforms whether you're on-prem or in the cloud. Let's talk a little bit about that.

J.R. Storment

>> So the FOCUS spec was originally designed to normalize cloud billing data. What we saw over the few years since it's launched is that people are looking to get consistent cost and uses data across all the places they're consuming and paying, and that challenge is going to accelerate incredibly with tokens. The FOCUS 1.4 that's being announced tomorrow is going to help solve some of the data integrity, some of the deeper areas of how we reconcile and invoices together. But down the road, as we start to look at token consumption, we're not just talking about hundreds of millions of rows of billing data. We're talking about trillions of tokens. Take that to allocations per user, per session, per trace, this becomes a telemetry problem, an observability problem that now these FinOps teams, engineering teams, finance teams, procurement teams need to look at this in a totally new way.

Paul Nashawaty

>> One of the interesting things, my read on the keynote and all the pre-announcements and such is, there is a maturity level when it comes to FinOps, right? There are some organizations that are very mature, and they understand it very much. And then there are some that are like almost analysis paralysis, and they don't know how to get started. I think that's something that when I was looking at what the FOCUS release was coming out, one of the things that was interesting to me was it was frictionless. It was allowing for moving away from the individual spreadsheets, the CSV files, to all these different things and compiling it yourself, to more of a unified view that was frictionless. What we're seeing in our own research is, 67% of organizations are hiring generalists over specialists. So if they're not hiring specialists to do this work, then it's really hard to actually pull all the... I can have data-

J.R. Storment

>> Common languages.

Paul Nashawaty

>> Right. So this common language made a lot of sense. That was where my read was.

J.R. Storment

>> I think what we're finding is, the billing data, the focus data, the cost and uses data, it's really just about getting that TCO and the denominator in the value equation. Now, one of the great things about agentic and agentic costs is, agents, in theory, are going to do business tasks that we can now get to better unit economics, and we can say this thing at this outcome at this cost. So I think we're going to see that cascade further.

John Furrier

>> I mean, TCO's hard when it's a moving train, which it is now. Your slide you put up on the keynote, I thought, was the bold slide you put up at the beginning, AI subsidy error is ending, and #ThanksTokenMaxxing was the hashtag. But this really talks to... And again, Salesforce had a news this morning around buying a company called m3ter, and the metering era is upon us. The subscription era moving to metering, is this a tell sign to what needs to happen on the infrastructure side? Because everyone's talking about token costs and budgets because they're seeing the bills. But the underlying infrastructure, when you have agents running, you're running operations, the total cost of ownership goes really sideways when you screw up. So you're like, if you don't engineer the system, the token costs are pale in comparison to the TCO of bad design, bad use of resource. That's kind of like under the iceberg. See an iceberg under the water, what don't you see? So there's icebergs ahead maybe, or what's your guys' thoughts on this?

Nishant Gupta

>> I think it is certainly there. The total cost of ownership conversation with AI, it's only an extension. We need to understand that. I was having a conversation with a colleague over here yesterday, and we were talking about, should we be tagging prompts much like we tag resources or should we be tagging the tokens? And it is all about the attribution. I think we talked about... In the keynote, a lot of people touched upon what's the outcome, and how do you demonstrate the value of AI? In order to demonstrate the value, you need to be able to tag it, you need to be able to attribute the usage and say, "This particular agent or this particular service is spending so much money on the tokens. Are we getting the right value from it or not?"

John Furrier

>> Yeah, Paul brought that up with FOCUS. Yesterday we had many pregame conversations around FOCUS and the value. What's the valuation? I mean, I can almost see the finance spreadsheet folks thinking, "Okay, how do I value a project? What's the payback look like? Do I focus on cost or value?" So guys, what's that new narrative? What's that vision or direction of scoping the value proposition?

J.R. Storment

>> I hate to be the wet blanket, because FinOps and AI and all these things should be about value and intelligence. But the problems we're hearing right now from the big token spenders are not even, how do I get to the value, but how do I just allocate the spend? How do I understand the spend? Everybody's blowing token budgets. The SVPs and CFOs are like, who's responsible, which project, where does that need to go? So I mean, that's I think the fundamental challenge right now.

John Furrier

>> What are you hearing? Stack rank some of the challenges from that token tsunami cost.

J.R. Storment

>> I would summarize it just in, do I have a giant bill from AI provider, who's responsible, and what are they using it for?

Paul Nashawaty

>> Chargeback model.

J.R. Storment

>> I think chargeback gets to invoicing, and I'm just showing it and understanding so I can go talk to that person and say, "Should you be doing this?" But then ultimately, yes, that's all a building block to get to how do we show the value on it.

John Furrier

>> Companies are struggling today. What you're saying is that where did this come from?

J.R. Storment

>> Yeah. Yeah.

John Furrier

>> Who's using those tokens?

J.R. Storment

>> Right, right.

Nishant Gupta

>> There was an article in Wall Street Journal, I believe, a couple of days ago which says 25% of the companies have good visibility.

John Furrier

>> That's a stretch maybe.

Nishant Gupta

>> And more than 25% of the companies only look at their stuff after the bill comes. 30 days is too late. There was another article where somebody spent $500 million in a month. If you're going to be spending that kind of money or anywhere near that, you want to know this on a daily basis as opposed to 30 days later.

John Furrier

>> I interviewed a CIO, I asked him, "How do you value the project?" He goes, "We look at very pragmatic as this early days, trying to understand what's going on." He's like, "If we have the cost of tokens, what's the human equivalent cost?" So now they're like, "I don't really need that if the cost to do tokens for AI but is greater than the human cost unless there's a major needle moving moment."

J.R. Storment

>> There are two really interesting conversations I'm hearing from engineering leaders, and frankly, up from the CEOs as well. The first one is, how do we allocate our token budgets to give the right people tokens? And token is beginning to be a bonus thing, engineers wanting to get more leverage. Do you give 10 actors more? The other side of this is, of course, the question of, well, which of the work should humans even do and how do we balance that? And that's where tokenomics is a discipline, and I think is an emerging discipline because it is yet to be fully defined. It's going to start to hit more into those labor implications and the pricing model implications as we see that.

Paul Nashawaty

>> Yeah. I think that that goes back to your FinOps' move into the boardroom conversation, because I mean that really comes down to what is going to be changing for skillsets in order to understand how to actually execute appropriately. I mean, to John's point, I was talking to one organization, they were measuring the success of their employees by the number of tokens they were spending, which was crazy, right?

J.R. Storment

>> Those are the leaderboards that everybody's moving away from.

Paul Nashawaty

>> Right. So they have to move away from that.

John Furrier

>> I was talking to someone yesterday in the hallway here, and the conversation was we're spending more tokens figuring out who's using what tokens. So again, tokens to figure out tokens. So there's a lot of growth. The other conversation we had here on theCUBE was, the cloud way was a 10-year cycle on FinOps, and this was a shorter time series now, going on now, highly compressed accelerated adoption, accelerated telemetry. So a lot of those kind of... How are you guys seeing that? How's that playing out into FinOps Foundation and now the new token foundation?

J.R. Storment

>> The shift has been... For a lot of years, FinOps was a conversation at that VP, SVP, EVP level of trying to understand the costs. Now, tokens, and whether that be FinOps tokens or tokenomics as we emerge these things, is suddenly a CEO level conversation. I was at a private forum with some global CIOs at Fortune 25s a couple months ago, and I heard them saying words like, "This is an existential crisis for us." And some of these are 100-plus-year-old companies, right? So it's pretty fascinating to see how that's happening. But I think what we're going to see change in all this is, we've got to build on the tenants that we've been building in FinOps land for years, visibility and accountability, and pushing data to the edge of the engineers to make decisions, but we do have to do it 10 times as fast and with a much smaller unit now. But in the end, it is going to force the value conversation much sooner than it came before.

John Furrier

>> Nishant, I want to ask you because we're seeing two things going on at the same time, the growth of the FinOps and token cost side, a lot of usage, but the AI infrastructure is also accelerating where you can get more. You mentioned the H100s inference, you're starting to see different cluster sizes. You don't need the big rack scale systems for certain use cases, especially in the data center. So you got this AI evolution on the infrastructure side booming. The convergence, where does it connect? What's your vision on that?

Nishant Gupta

>> I think that's where you have the G1's paradox, right? So you're seeing some of the older models and they're much cheaper now because they can run on older hardware and you have the newer models which are more expensive, but the growth is accelerating. I see this as an example of... And I do not know if it is really... There's a narrative around the inflationary aspect of the AI. I don't see it that way right now. I'll give you an example. The cost of your iPhone is the same over the last 10 years. You can buy TV for 1,000 bucks from Costco and you could buy a TV for 1,000 bucks 10 years ago as well. So your spend remains the same. You get better value and better product from it, but nobody's going to be using a GPT 3.0 model anymore for any real work at this point. So I think that remains to be seen in terms of how we make the mix of it. The other challenge is, we talked a little bit about model routing on the use cases. I think that's still a maturing practice to be able to better understand what problem sets for what models and what price point are the right things to do. I don't think we have fully cracked that thing yet.

John Furrier

>> Guys, great to have you on. Great keynote. Final question, J.R., for you is, on the foundation, what should people know about the change with FinOps and now the Tokenomics Foundation? What's the big change that they should know about here in this era?

J.R. Storment

>> The most important thing is that the FinOps Foundation is to keep doing everything it's been doing. We're keeping a single consolidated practitioner community. We're keeping doing our working groups and our community calls and our virtual and regional events. We're going to expand that sphere of influence to touch on some new areas, and we're going to bring a new part of this conversation into it, which is the tokenomics side. There's going to be a dedicated foundation for this that goes deep into working groups and defining the primitives of this, but we're not going to separate these things out in communities. The people are going to intersect and align. We can't stop doing the FinOps work, as Nishant said to me earlier. We still got to do the cloud FinOps, and we also need to go do the tokenomics work.

John Furrier

>> Decoupled, but highly cohesive, but connected.

J.R. Storment

>> Exactly. We'll share some operational resources, but dedicated SME teams.

John Furrier

>> Nishant, for practitioners that are trying to frame the problem statement, what's your advice? How should they lean into this? Because remember, when you lean into these emerging trends, there's downstream benefits, maybe a little pain and suffering, but if you get it right-

Nishant Gupta

>> Yeah. So with the FinOps side, we had best practices for the last 20 years that you can rely on. Nobody has best practices for the last six months. So you got to blaze the trail in some sense. Our focus on a practitioner side and a huge focus in the amount of money that we are spending on tokens is going to be around understanding the trends, understanding the usage, the patterns, and how best to optimize. We don't have the luxury to wait on somebody else to tell what the best practices are. I think we are somewhat racing the trail on that.

John Furrier

>> The phrase eating your own dog food, drinking your own champagne, client zero was kind of like regulated for a few firms, but that every company has to do that now. They got to roll their sleeves up.

Nishant Gupta

>> I think doing that in vacuum as an experiment is okay. And that's why I'm excited about tokenomics. But we need to bring those best practices and share it broadly within the industry, not only from a practitioner standpoint, but also from different tier models standpoint. Those who are provider of this data and visibility and insights, they need to be part of the conversation.

John Furrier

>> It's a great time to be an engineer. It's a great time to be an AI builder. J.R., Nishant, thank you for coming on theCUBE.

Nishant Gupta

>> Thank you very much.

J.R. Storment

>> Thank you, John. Thank you, Paul.

John Furrier

>> Kicking off the day here. I'm John Furrier, host of theCUBE, with Paul Nashawaty, my co-host with theCUBE Research. We'll be right back after this short break.