RAISE Summit '26 | Greg Matson, Solidigm

Clips
More from RAISE Summit '26

Greg Matson

Senior VP, Head of Marketing & Products

Solidigm

play_circle_outline SSDs as an Intelligent Memory Layer: Extending HBM/DRAM for High-Capacity GPU-Adjacent Storage

play_circle_outline Beyond HBM/DRAM: Storage as the Intelligence Layer to Scale Agentic Inference

play_circle_outline From HDDs to SSDs: Solidigm's High-Capacity Drives Enable New AI Supercomputing Architecture for GPU Training and Inference

play_circle_outline Shift from PC-era architecture to supercomputer-era AI factories and infrastructure

play_circle_outline Importance of keeping GPUs fed for continuous token generation and efficiency

play_circle_outline From PCIe Gen4 to Gen6: Direct‑Attach SSDs Power Hyperscalers' AI Storage Rip‑and‑Replace

play_circle_outline Token economics: storage performance reduces token costs and improves throughput

play_circle_outline Token economics: storage performance reduces token cost and improves throughput

Info

Greg Matson, Solidigm

Greg Matson

Senior VP, Head of Marketing & Products Solidigm

search

John Furrier

>> Palo Alto Studio Connection, Silicon Valley and Wall Street.

Greg Matson

>> I'm John Furrier, co-host here with Dave Vellante, my co-host.

John Furrier

>> Welcome to the RAISE Summit coverage brought to you by Solidigm with support from Tensordyne, d-Matrix, and Argentum. Welcome to theCUBE, everyone, live streaming here in Paris, France for the RAISE Summit. This is day 1 kickoff. We had day 0. Yesterday, MACHINA Summit was the physical AI and robotics, kind of a sidecar event. But as that growth continues, the main story is the growth of AI here at the RAISE Summit. Action-packed house here with all the top leaders in the global tech scene, certainly with AI driving. We're kicking off the show with Greg Matson, Senior Vice President, Head for Products and Marketing at Solidigm. Greg, great to have you on, and thank you guys for sponsoring theCUBE. Really appreciate the support. And again, the success for Solidigm continues. I know you guys, The demand curve's crushing you. So welcome back to theCUBE, great to see you.

Greg Matson

>> Well, first it's great to be here at RAISE, right? It's our first time being at RAISE, and I can't believe the collection of leaders from across the AI landscape, so it's very exciting to be here.

John Furrier

>> it's really exciting for me, even following you guys, been following the growth for many, many years now, and the past 2 years in particular, what you guys have done on the product side has been incredible. I've always been kind of falling over myself looking at how small and how performant the drives are on the SSD. But last night at the Versailles event, everyone was clamoring, give me some product. You guys are in high demand. And I think if you look at what NVIDIA and all this large-scale AI factories is, the AI growth is a completely new infrastructure build-out. And the AI growth here at the RAISE Summit, Mark Cuban was just on stage, they're talking about Lovable and all the apps are coming in and the power for the infrastructure is really enabling that. But the density, which used to be, oh, there's too much dense equipment on the board, is actually a feature because they want to get the components together, and you guys have had a great product. So talk about that dynamic, one, the demand, and two, how the infrastructure has changed to get this high-performance supercomputing capability.

Greg Matson

>> Yeah, absolutely. First is, you're right, the demand is exploding. it started a couple of years ago with training, where the need for high-capacity, high-performance storage very adjacent to the GPUs was all of a sudden, center stage, right? We needed as high capacity as we could get. But now at the inference, as we go from last year into this year inference phase into agentic inference, it's exploding even more. And where you're talking about, the growth in inference, basically blowing out for the HBM and the DRAM footprint. Needing to be extended into storage. And now storage is actually a whole new storage tier that's being created to extend the memory for the system. And that memory is really where the data is being housed, that's being worked on, and that's taking that data and turning it actually into the intelligence. So I almost call it the intelligence layer is now being housed on storage.

John Furrier

>> And then I bring up the density thing because the market's flipped, it's inverted now. As you're seeing a whole different paradigm for engineering. We've been talking many times on theCUBE Pod about, Windows and Intel, and during that era for the processor advancements that powered the PC revolution, it was always that operating system that brought the applications in. So Windows had applications, and what we're seeing now with AI is that those large-scale AI factories, and now the enterprise clusters, have to have that power. So every year we're seeing more and more functionality from NVIDIA, But what's interesting is that it's not the PC era, it's the supercomputer era. And Solidigm's at the center of it because there's no more hard disk drives anymore. That was the old era.

Greg Matson

>> That was old days.

John Furrier

>> The SSDs have come in and they play a critical role because now you're seeing desktop computers that are AI-enabled. Again, servers first, systems first, now desktops. You guys are the key element. Explain that nuance because SSDs and high-bandwidth memory work together. They're not just like, hey, there's memory and a hard disk. Talk about the dynamic.

Greg Matson

>> The— definitely the era of memory and a hard disk is over in AI, right? Because that HBM needs an extension, right? It needs to be fed continuously with a lot of data. And high-capacity SSDs are really what's coming into play, over the past couple of years. It was 30 terabytes, it was 60 terabytes, it's 122 terabytes per drive. Now you have a system full of drives, a rack full of systems, and you're talking about petabytes worth of storage per GPU cluster. And that's really to make sure that you keep your GPUs fed. And while the GPU is the most expensive part of your infrastructure, you want that thing to be humming 100% of the time generating tokens. And if it's down, right, waiting for data, then you're wasting your money on GPUs and you're getting— it's just much less efficient without that high-capacity storage right next to the

John Furrier

>> GPU.It's funny, I was talking to someone yesterday, they're like, yeah, you're seeing the models, OpenAI, Anthropic, you're going to start to see the performance. If you're waiting, it feels like dial-up internet. It's like, so speed is critical. And if you look at storage, I was talking with VAST Data as well, the storage equation, they don't even call themselves storage, they call themselves a data platform. So the data layer for inference context windows, the things that are powering Replit, all these apps have to have that high performance. Talk about that dynamic because you guys sit in the middle, you guys enable it. What are you guys doing today in terms of product? What's coming down the pike? I know the demand curves, there's a line around the corner for products, but the storage equation has changed. The people that you sell to and you're embedding into, these are large-scale systems that are powering the performance for AI.

Greg Matson

>> The demand scale is for sure changing. And we can look out as far as we can see and we have demand. And I think that's the same for anyone in this AI infrastructure space. But the infrastructure now is really the asset that's going to power the AI revolution.

John Furrier

>> How about the enterprise? The enterprise is a new thing this year, not new, but the crossover from hard disk drives to SSD, but also the configurations on-premises. And certainly here at RAISE Summit, sovereignty is a huge issue. You start to see these new deployments. Can you share any commentary on how those are going, how you see those happening?

Greg Matson

>> Well, first is, the problems or the challenges from an IT infrastructure at scale, at the hyperscale level, and out at the edge at the enterprises are the same. They need high capacity, they need high performance in a space-efficient, power-efficient footprint. And high-capacity SSDs are really the only way to get that at the storage layer, or as VAST talks about it, the data layer. And I even call it the intelligence layer, because you take that data and process it and turn it into actionable information.

John Furrier

>> When you have conversations here at RAISE and around the industry, what's the current state-of-the-art conversation right now? What are people asking you for? What is the top-of-mind concern, opportunity?

Greg Matson

>> Well, of course, AI, AI, AI, right?

John Furrier

>> We saw that, you got mobbed last night.

Greg Matson

>> Yeah, it was pretty good. And it's nice to be there, right? Because memory and storage has a lot of times been the afterthought. Now it's really the forethought because people are realizing that AI is data, right? That is the key. And data's housed in storage and it's worked on in memory. And that's really why people are really starting to think about those two things first, or at least in parallel with their GPU and their system decisions.

John Furrier

>> Yeah, there's a big conversation. Liquid-cooled has been super hot, and most people think it's for large-scale systems. You guys have pioneered that one piece. Outside of that, talk about the product mix. What are you seeing there in terms of deployments?

Greg Matson

>> Yeah, first is liquid-cooled. We did introduce the world's first liquid-cooled SSD and continue that development as we see NVIDIA especially pushing out, completely fanless systems. And so every component within the rack will be liquid-cooled, whether it needs to be or not, because they want to reduce the power and the space associated with the fans. And that's actually great for the data center environments because it also creates a new level of efficiency and environmental friendliness in terms of water consumption and things like that, and power consumption. It also allows us to run at way higher performance levels, and at a higher power per unit, but higher performance overall at the system level.

John Furrier

>> All right. This year at the RAISE Summit, I'd like to get your take on what's your current vibe. Obviously sovereignty is a hot conversation. What that means is people are building out essentially large-scale infrastructure in regions and countries and enterprises have sovereignty. What does that mean for folks? And what do you think the top story is so far at RAISE? What's coming out of the current show floor right now for you?

Greg Matson

>> Well, first is day one in the morning. Yeah, but I think sovereignty is important, right? We talked to a lot of people already last night and today from different parts of the world, all interested in the very similar things, right? Building their own infrastructure in the most efficient way, a state-of-the-art way possible, I think is still really the thing that we've heard so far.

John Furrier

>> And performance-wise, what are you hearing from people asking for performance? Where's that puck going? How do you see that playing out?

Greg Matson

>> Well, on the SSD side, we're going from PCIe Gen4 to Gen5 in the high-capacity storage space, and then from Gen5 to Gen6 as the Blackwell systems move to Rubin systems, and that's really for the direct attach SSD space. So really, performance is key, capacity is key, but most importantly, supply and availability is key.

John Furrier

>> Yeah, so you're very popular. What are some of the cloud dynamics? Got the hyperscalers, the neo-clouds, and the enterprises all booming right now in terms of buildout. What are some for the— trends you're seeing there?

Greg Matson

>> Some of the things— so the hyperscalers had massive amounts of storage infrastructure that already existed for traditional compute. What they found is, about 12 to 18 months ago, they're like, uh-oh, hey, this doesn't keep up. And so they're really doing something they hadn't done in a long time. There's rip and replace going on in terms of the big storage infrastructure. And that's something that, really they had been letting assets run for as long as they possibly could. And we have some drives that have been going on in production for more than 10 years. So now they're deploying modern storage, high-capacity storage. And we're really seeing this high-capacity adoption across the board with the hyperscalers, also with the NeoClouds, also at the edge. So it's really a very similar thing across the span for customers.

John Furrier

>> price performance has always been a benchmark in technology. Hardware, software. Now you got token costs. These are now new economics.

Greg Matson

>> Yeah.

John Furrier

>> How are you guys looking at performance and price on the SSD side, and how does that translate in your mind to some of these token economies that are developing? Because tokens are the new currency in AI. That's now got a performance element and cost to it. Talk about the price performance dynamic that you guys are constantly iterating on.

Greg Matson

>> Well, we have, of course, prices are going up. It's great for us. Historically, it's been a challenge sometimes to invest in the capital infrastructure that you need to build out SSDs. Now it becomes affordable. It's becoming essential infrastructure for AI. And it's also essential for making those tokens more and more affordable. As you're building an AI factory to generate tokens. So if you're waiting on data, then you're not generating tokens.. And even a simple 15-word prompt can generate as much as 40,000 tokens through the context for that entire prompt. That's maybe 5 to 10 gigabytes for data. Think of that as one person's simple prompt. And now you go, okay, I have X number of employees using Y number of agents, and you can go quickly into the petabytes worth of just context storage. And so needing that high capacity and high-performance storage actually improves your token economics and lowers your cost for tokens and improves the overall token output in your factory.

John Furrier

>> It's funny to see, I brought up the PC revolution because I think there's a very comparable where you had that hard disk drive, it was separate from the processor. Today the processor is GPU, HBM memory, SSD, it's kind of a new architecture that's actually one thing. You mentioned context window, you look at KV cache for instance with NVIDIA, the networking has to move the data to the GPU as fast as possible. So in a way, the SSDs are becoming the processor by default because they have to be right there. They can't be slow. They got to be tight.

Greg Matson

>> Right there and on and ready to go. So high performance, available, and high capacity to make sure that you're having all that data very adjacent. you think about, if you see a book, There's information, there's data over here. You read a book, that starts to become information to you. You think about the book, you talk to somebody about the book, now you're creating intelligence around that book. That's exactly what's happening at the inference layer. And you need to have that data, basically the GPU and the data become now a seamless system through—

John Furrier

>> and you're seeing out here the examples, the enablement for making that work. Powers the AI revolution. And I think that's a story people are starting to understand. How do you explain all this when people who aren't in tech in your family or friend group, like, hey Greg, what's going on with all this AI stuff? Well, how do you explain it?

Greg Matson

>> Well, partly a little bit what I— that's why I brought up the book analogy, because it's something that anyone can think about or anyone can relate to. If you— I try to liken it back to the cloud era. So if you go back into the beginning of the internet era, maybe late '90s, early 2000s, and this is, that was the time when infrastructure started to be built for the internet. But the internet didn't really budge until, for decades, it's been building out. Now it's, everyone's life is online. And that's where we are kind of at the AI infrastructure timeframe is we're here, it sounds like big business, Stocks are going crazy, revenues for everyone in the hardware business are going crazy, but it's just the beginning. And so it's just the worldwide accelerated compute infrastructure buildout is just in its infancy.

John Furrier

>> It's a whole new architecture. I was talking last night about just the 5G buildout. Just think about this, 5G. We know our phones, 1G, 2G, 3G, 4G, just 5G telecom, close to $1 trillion just on that. If you look at the totality for just 3G to 5G, it's trillions and trillions of dollars. And people kind of scratch their heads like, whoa, so much money's being spent on AI. These are new roads being paved. This is concrete being poured metaphorically. So a whole nother generation of architecture. This isn't a data center refresh. This is a whole nother thing.

Greg Matson

>> Brand new from the ground up, built for acceleration.

John Furrier

>> And you guys are, and how are you guys designing into that? Continuing to get closer to the chips, working on density, smaller the form factor's the edge, share that a little bit with us.

Greg Matson

>> So we start at the silicon level. So each memory chip has to get denser, smaller in terms of bit density, and so we come out and have a very, fast cadence for new memory chips, but then we build those into SSDs. We've been engineering at the package level. So we stack a lot of those chips in one package. We try to make those as small as possible and stack as many as we can in each package. And then at the form factor level, there's different form factors for different chassis designs as well as different interface speeds for the drive. So we have to actually fan out a number of form factors. And then of course there's air-cooled and liquid-cooled form factors.

John Furrier

>> you can't even call them drives anymore because the crossover from hard disk drive to SSD makes it memory-based. It's memory.

Greg Matson

>> It's memory.

John Furrier

>> And so you guys are the inside power engine for the inference era for AI. Solidigm inside is what's powering this whole generation.

Greg Matson

>> Absolutely. The infrastructure is the data layer. The data layer is the storage layer. that's the Solidigm layer.

John Furrier

>> Infrastructure is the new killer app. Okay, just let's end by talking about some of the partnerships. Just share the relationships you guys have. you're working with everybody. Just talk about some of the big names.

Greg Matson

>> Yeah, first is, it's very public, today you'll see me on stage with VAST Data, one of our great partners and great winners in terms of AI infrastructure storage. We partner with all the different ISVs though, Weka, MinIO, we have, big customer relationships with the big hyperscalers that you can imagine, both in the US but also globally. So we want to make sure that we're partnering with the neo clouds, as well. We've been on stage a lot with CoreWeave, so they're the one a lot of people know, but many, many others as

John Furrier

>> well.And the liquid cooling gets you into the big systems too, because everything on the high, high end is liquid

Greg Matson

>> cooled.Oh, everything that's coming out over the next couple of years, nothing will be air-cooled.

John Furrier

>> Well, great design. You guys have, I love, love that form factor. I first saw it, I remember saying, that's not real. I can't believe that's going to happen. Congratulations.

Greg Matson

>> Well,

John Furrier

>> thank you.Thanks for coming on

Greg Matson

>> theCUBE.Thanks, John. Appreciate

John Furrier

>> it.All right. Solidigm powering the next generation, Solidigm inside the AI factories and all the AI growth that's happened. It's just the beginning. the roads are being paved, concrete's being poured for the AI infrastructure. These data centers are expanding to the edge as well. We're going to do our part here on theCUBE to cover it. I'm John Furrier, your host. Thanks for watching.

Greg Matson, Solidigm

search

John Furrier

>> Palo Alto Studio Connection, Silicon Valley and Wall Street.

Greg Matson

>> I'm John Furrier, co-host here with Dave Vellante, my co-host.

John Furrier

>> Welcome to the RAISE Summit coverage brought to you by Solidigm with support from Tensordyne, d-Matrix, and Argentum. Welcome to theCUBE, everyone, live streaming here in Paris, France for the RAISE Summit. This is day 1 kickoff. We had day 0. Yesterday, MACHINA Summit was the physical AI and robotics, kind of a sidecar event. But as that growth continues, the main story is the growth of AI here at the RAISE Summit. Action-packed house here with all the top leaders in the global tech scene, certainly with AI driving. We're kicking off the show with Greg Matson, Senior Vice President, Head for Products and Marketing at Solidigm. Greg, great to have you on, and thank you guys for sponsoring theCUBE. Really appreciate the support. And again, the success for Solidigm continues. I know you guys, The demand curve's crushing you. So welcome back to theCUBE, great to see you.

Greg Matson

>> Well, first it's great to be here at RAISE, right? It's our first time being at RAISE, and I can't believe the collection of leaders from across the AI landscape, so it's very exciting to be here.

John Furrier

>> it's really exciting for me, even following you guys, been following the growth for many, many years now, and the past 2 years in particular, what you guys have done on the product side has been incredible. I've always been kind of falling over myself looking at how small and how performant the drives are on the SSD. But last night at the Versailles event, everyone was clamoring, give me some product. You guys are in high demand. And I think if you look at what NVIDIA and all this large-scale AI factories is, the AI growth is a completely new infrastructure build-out. And the AI growth here at the RAISE Summit, Mark Cuban was just on stage, they're talking about Lovable and all the apps are coming in and the power for the infrastructure is really enabling that. But the density, which used to be, oh, there's too much dense equipment on the board, is actually a feature because they want to get the components together, and you guys have had a great product. So talk about that dynamic, one, the demand, and two, how the infrastructure has changed to get this high-performance supercomputing capability.

Greg Matson

>> Yeah, absolutely. First is, you're right, the demand is exploding. it started a couple of years ago with training, where the need for high-capacity, high-performance storage very adjacent to the GPUs was all of a sudden, center stage, right? We needed as high capacity as we could get. But now at the inference, as we go from last year into this year inference phase into agentic inference, it's exploding even more. And where you're talking about, the growth in inference, basically blowing out for the HBM and the DRAM footprint. Needing to be extended into storage. And now storage is actually a whole new storage tier that's being created to extend the memory for the system. And that memory is really where the data is being housed, that's being worked on, and that's taking that data and turning it actually into the intelligence. So I almost call it the intelligence layer is now being housed on storage.

John Furrier

>> And then I bring up the density thing because the market's flipped, it's inverted now. As you're seeing a whole different paradigm for engineering. We've been talking many times on theCUBE Pod about, Windows and Intel, and during that era for the processor advancements that powered the PC revolution, it was always that operating system that brought the applications in. So Windows had applications, and what we're seeing now with AI is that those large-scale AI factories, and now the enterprise clusters, have to have that power. So every year we're seeing more and more functionality from NVIDIA, But what's interesting is that it's not the PC era, it's the supercomputer era. And Solidigm's at the center of it because there's no more hard disk drives anymore. That was the old era.

Greg Matson

>> That was old days.

John Furrier

>> The SSDs have come in and they play a critical role because now you're seeing desktop computers that are AI-enabled. Again, servers first, systems first, now desktops. You guys are the key element. Explain that nuance because SSDs and high-bandwidth memory work together. They're not just like, hey, there's memory and a hard disk. Talk about the dynamic.

Greg Matson

>> The— definitely the era of memory and a hard disk is over in AI, right? Because that HBM needs an extension, right? It needs to be fed continuously with a lot of data. And high-capacity SSDs are really what's coming into play, over the past couple of years. It was 30 terabytes, it was 60 terabytes, it's 122 terabytes per drive. Now you have a system full of drives, a rack full of systems, and you're talking about petabytes worth of storage per GPU cluster. And that's really to make sure that you keep your GPUs fed. And while the GPU is the most expensive part of your infrastructure, you want that thing to be humming 100% of the time generating tokens. And if it's down, right, waiting for data, then you're wasting your money on GPUs and you're getting— it's just much less efficient without that high-capacity storage right next to the

John Furrier

>> GPU.It's funny, I was talking to someone yesterday, they're like, yeah, you're seeing the models, OpenAI, Anthropic, you're going to start to see the performance. If you're waiting, it feels like dial-up internet. It's like, so speed is critical. And if you look at storage, I was talking with VAST Data as well, the storage equation, they don't even call themselves storage, they call themselves a data platform. So the data layer for inference context windows, the things that are powering Replit, all these apps have to have that high performance. Talk about that dynamic because you guys sit in the middle, you guys enable it. What are you guys doing today in terms of product? What's coming down the pike? I know the demand curves, there's a line around the corner for products, but the storage equation has changed. The people that you sell to and you're embedding into, these are large-scale systems that are powering the performance for AI.

Greg Matson

>> The demand scale is for sure changing. And we can look out as far as we can see and we have demand. And I think that's the same for anyone in this AI infrastructure space. But the infrastructure now is really the asset that's going to power the AI revolution.

John Furrier

>> How about the enterprise? The enterprise is a new thing this year, not new, but the crossover from hard disk drives to SSD, but also the configurations on-premises. And certainly here at RAISE Summit, sovereignty is a huge issue. You start to see these new deployments. Can you share any commentary on how those are going, how you see those happening?

Greg Matson

>> Well, first is, the problems or the challenges from an IT infrastructure at scale, at the hyperscale level, and out at the edge at the enterprises are the same. They need high capacity, they need high performance in a space-efficient, power-efficient footprint. And high-capacity SSDs are really the only way to get that at the storage layer, or as VAST talks about it, the data layer. And I even call it the intelligence layer, because you take that data and process it and turn it into actionable information.

John Furrier

>> When you have conversations here at RAISE and around the industry, what's the current state-of-the-art conversation right now? What are people asking you for? What is the top-of-mind concern, opportunity?

Greg Matson

>> Well, of course, AI, AI, AI, right?

John Furrier

>> We saw that, you got mobbed last night.

Greg Matson

>> Yeah, it was pretty good. And it's nice to be there, right? Because memory and storage has a lot of times been the afterthought. Now it's really the forethought because people are realizing that AI is data, right? That is the key. And data's housed in storage and it's worked on in memory. And that's really why people are really starting to think about those two things first, or at least in parallel with their GPU and their system decisions.

John Furrier

>> Yeah, there's a big conversation. Liquid-cooled has been super hot, and most people think it's for large-scale systems. You guys have pioneered that one piece. Outside of that, talk about the product mix. What are you seeing there in terms of deployments?

Greg Matson

>> Yeah, first is liquid-cooled. We did introduce the world's first liquid-cooled SSD and continue that development as we see NVIDIA especially pushing out, completely fanless systems. And so every component within the rack will be liquid-cooled, whether it needs to be or not, because they want to reduce the power and the space associated with the fans. And that's actually great for the data center environments because it also creates a new level of efficiency and environmental friendliness in terms of water consumption and things like that, and power consumption. It also allows us to run at way higher performance levels, and at a higher power per unit, but higher performance overall at the system level.

John Furrier

>> All right. This year at the RAISE Summit, I'd like to get your take on what's your current vibe. Obviously sovereignty is a hot conversation. What that means is people are building out essentially large-scale infrastructure in regions and countries and enterprises have sovereignty. What does that mean for folks? And what do you think the top story is so far at RAISE? What's coming out of the current show floor right now for you?

Greg Matson

>> Well, first is day one in the morning. Yeah, but I think sovereignty is important, right? We talked to a lot of people already last night and today from different parts of the world, all interested in the very similar things, right? Building their own infrastructure in the most efficient way, a state-of-the-art way possible, I think is still really the thing that we've heard so far.

John Furrier

>> And performance-wise, what are you hearing from people asking for performance? Where's that puck going? How do you see that playing out?

Greg Matson

>> Well, on the SSD side, we're going from PCIe Gen4 to Gen5 in the high-capacity storage space, and then from Gen5 to Gen6 as the Blackwell systems move to Rubin systems, and that's really for the direct attach SSD space. So really, performance is key, capacity is key, but most importantly, supply and availability is key.

John Furrier

>> Yeah, so you're very popular. What are some of the cloud dynamics? Got the hyperscalers, the neo-clouds, and the enterprises all booming right now in terms of buildout. What are some for the— trends you're seeing there?

Greg Matson

>> Some of the things— so the hyperscalers had massive amounts of storage infrastructure that already existed for traditional compute. What they found is, about 12 to 18 months ago, they're like, uh-oh, hey, this doesn't keep up. And so they're really doing something they hadn't done in a long time. There's rip and replace going on in terms of the big storage infrastructure. And that's something that, really they had been letting assets run for as long as they possibly could. And we have some drives that have been going on in production for more than 10 years. So now they're deploying modern storage, high-capacity storage. And we're really seeing this high-capacity adoption across the board with the hyperscalers, also with the NeoClouds, also at the edge. So it's really a very similar thing across the span for customers.

John Furrier

>> price performance has always been a benchmark in technology. Hardware, software. Now you got token costs. These are now new economics.

Greg Matson

>> Yeah.

John Furrier

>> How are you guys looking at performance and price on the SSD side, and how does that translate in your mind to some of these token economies that are developing? Because tokens are the new currency in AI. That's now got a performance element and cost to it. Talk about the price performance dynamic that you guys are constantly iterating on.

Greg Matson

>> Well, we have, of course, prices are going up. It's great for us. Historically, it's been a challenge sometimes to invest in the capital infrastructure that you need to build out SSDs. Now it becomes affordable. It's becoming essential infrastructure for AI. And it's also essential for making those tokens more and more affordable. As you're building an AI factory to generate tokens. So if you're waiting on data, then you're not generating tokens.. And even a simple 15-word prompt can generate as much as 40,000 tokens through the context for that entire prompt. That's maybe 5 to 10 gigabytes for data. Think of that as one person's simple prompt. And now you go, okay, I have X number of employees using Y number of agents, and you can go quickly into the petabytes worth of just context storage. And so needing that high capacity and high-performance storage actually improves your token economics and lowers your cost for tokens and improves the overall token output in your factory.

John Furrier

>> It's funny to see, I brought up the PC revolution because I think there's a very comparable where you had that hard disk drive, it was separate from the processor. Today the processor is GPU, HBM memory, SSD, it's kind of a new architecture that's actually one thing. You mentioned context window, you look at KV cache for instance with NVIDIA, the networking has to move the data to the GPU as fast as possible. So in a way, the SSDs are becoming the processor by default because they have to be right there. They can't be slow. They got to be tight.

Greg Matson

>> Right there and on and ready to go. So high performance, available, and high capacity to make sure that you're having all that data very adjacent. you think about, if you see a book, There's information, there's data over here. You read a book, that starts to become information to you. You think about the book, you talk to somebody about the book, now you're creating intelligence around that book. That's exactly what's happening at the inference layer. And you need to have that data, basically the GPU and the data become now a seamless system through—

John Furrier

>> and you're seeing out here the examples, the enablement for making that work. Powers the AI revolution. And I think that's a story people are starting to understand. How do you explain all this when people who aren't in tech in your family or friend group, like, hey Greg, what's going on with all this AI stuff? Well, how do you explain it?

Greg Matson

>> Well, partly a little bit what I— that's why I brought up the book analogy, because it's something that anyone can think about or anyone can relate to. If you— I try to liken it back to the cloud era. So if you go back into the beginning of the internet era, maybe late '90s, early 2000s, and this is, that was the time when infrastructure started to be built for the internet. But the internet didn't really budge until, for decades, it's been building out. Now it's, everyone's life is online. And that's where we are kind of at the AI infrastructure timeframe is we're here, it sounds like big business, Stocks are going crazy, revenues for everyone in the hardware business are going crazy, but it's just the beginning. And so it's just the worldwide accelerated compute infrastructure buildout is just in its infancy.

John Furrier

>> It's a whole new architecture. I was talking last night about just the 5G buildout. Just think about this, 5G. We know our phones, 1G, 2G, 3G, 4G, just 5G telecom, close to $1 trillion just on that. If you look at the totality for just 3G to 5G, it's trillions and trillions of dollars. And people kind of scratch their heads like, whoa, so much money's being spent on AI. These are new roads being paved. This is concrete being poured metaphorically. So a whole nother generation of architecture. This isn't a data center refresh. This is a whole nother thing.

Greg Matson

>> Brand new from the ground up, built for acceleration.

John Furrier

>> And you guys are, and how are you guys designing into that? Continuing to get closer to the chips, working on density, smaller the form factor's the edge, share that a little bit with us.

Greg Matson

>> So we start at the silicon level. So each memory chip has to get denser, smaller in terms of bit density, and so we come out and have a very, fast cadence for new memory chips, but then we build those into SSDs. We've been engineering at the package level. So we stack a lot of those chips in one package. We try to make those as small as possible and stack as many as we can in each package. And then at the form factor level, there's different form factors for different chassis designs as well as different interface speeds for the drive. So we have to actually fan out a number of form factors. And then of course there's air-cooled and liquid-cooled form factors.

John Furrier

>> you can't even call them drives anymore because the crossover from hard disk drive to SSD makes it memory-based. It's memory.

Greg Matson

>> It's memory.

John Furrier

>> And so you guys are the inside power engine for the inference era for AI. Solidigm inside is what's powering this whole generation.

Greg Matson

>> Absolutely. The infrastructure is the data layer. The data layer is the storage layer. that's the Solidigm layer.

John Furrier

>> Infrastructure is the new killer app. Okay, just let's end by talking about some of the partnerships. Just share the relationships you guys have. you're working with everybody. Just talk about some of the big names.

Greg Matson

>> Yeah, first is, it's very public, today you'll see me on stage with VAST Data, one of our great partners and great winners in terms of AI infrastructure storage. We partner with all the different ISVs though, Weka, MinIO, we have, big customer relationships with the big hyperscalers that you can imagine, both in the US but also globally. So we want to make sure that we're partnering with the neo clouds, as well. We've been on stage a lot with CoreWeave, so they're the one a lot of people know, but many, many others as

John Furrier

>> well.And the liquid cooling gets you into the big systems too, because everything on the high, high end is liquid

Greg Matson

>> cooled.Oh, everything that's coming out over the next couple of years, nothing will be air-cooled.

John Furrier

>> Well, great design. You guys have, I love, love that form factor. I first saw it, I remember saying, that's not real. I can't believe that's going to happen. Congratulations.

Greg Matson

>> Well,

John Furrier

>> thank you.Thanks for coming on

Greg Matson

>> theCUBE.Thanks, John. Appreciate

John Furrier

>> it.All right. Solidigm powering the next generation, Solidigm inside the AI factories and all the AI growth that's happened. It's just the beginning. the roads are being paved, concrete's being poured for the AI infrastructure. These data centers are expanding to the edge as well. We're going to do our part here on theCUBE to cover it. I'm John Furrier, your host. Thanks for watching.