VAST Forward 2026 | Phil Manez, VAST & Ace Stryker, Solidigm

Clips
More from VAST Forward 2026

Phil Manez

Vice President, GTM Execution

VAST Data

Ace Stryker

Director of AI & Ecosystem Marketing

Solidigm

The AI boom is turning flash storage into a critical infrastructure battleground

An AI-driven flash shortage is emerging as one of the defining infrastructure challenges of the machine learning boom. More than a problem solved by waiting out the typical hardware cycle, the shortage is forcing a larger rethink on how enterprises can get more from existing capacity.Unlike past supply crunches, this shortage is hitting even the biggest buyers. The current pattern suggests the shortage is unlikely to resolve on a normal product-cycle timeline, according to Phil Manez (pictured, right), vice president of go-to-market execution at Vast Data Inc

play_circle_outline Flash shortage affecting even the largest customers

play_circle_outline AI's Memory Crunch: RAG, Vectorization, Long Contexts, and KV Caches Driving DRAM and NAND Demand

play_circle_outline Supply Crunch: Years-Long Fab Builds Drive Memory Price Surges and Supplier Stock Volatility

play_circle_outline VAST and Solidigm partnership optimizes existing deployed flash usage

Info
Transcript

Phil Manez, VAST & Ace Stryker, Solidigm

Phil Manez

Vice President, GTM Execution VAST Data

Ace Stryker

Director of AI & Ecosystem Marketing Solidigm

In this interview from VAST Forward in Salt Lake City, Phil Manez, vice president of go-to-market execution at VAST Data, and Ace Stryker, director of AI and ecosystem marketing at Solidigm, join theCUBE’s Dave Vellante and Rebecca Knight to unpack a supply shock most enterprises can’t ignore: a flash and memory crunch being accelerated by AI. Manez argues the demand curve isn’t behaving like a normal cycle – it’s being rewritten by inference at scale, where new applications and agentic workflows keep generating fresh data and chewing through capacity faster ... Read more

explore Keep Exploring

Why won't this shortage resolve within a typical product cycle? add

Why is the memory market (DRAM and NAND demand) experiencing a structurally different surge now, even though training-data availability has plateaued? add

What impact is AI demand having on DRAM and NAND supply and prices, why is increasing capacity difficult, and how can technologies like Amplify and VAST improve storage efficiency to mitigate those constraints? add

How did this program come together, and what role did customer needs, scale challenges, and the collaboration with VAST and Solidigm play in its development? add

bolt Powered by CUBE AI

Phil Manez, VAST & Ace Stryker, Solidigm

search

Rebecca Knight

>> Good afternoon everyone and welcome back to theCUBE's live coverage of VAST Forward here in Salt Lake City. I'm your host, Rebecca Knight, alongside Dave Vellante at VAST's first ever customer conference. It's pretty exciting to be here.

Dave Vellante

>> Yeah, it's a big deal. I love to watch the progression of these events over the years. They start small, and they get big and big and really big.

Rebecca Knight

>> That's the thing. We'll be here years from now remembering that quaint little conference in Salt Lake one day. I would like to welcome our next guest to the show. We have Phil Manez. He is the VP of go-to-market execution at VAST. Welcome, Phil.

Phil Manez

>> Thank you so much for having us.

Rebecca Knight

>> And Ace Stryker, director of AI and ecosystem marketing at Solidigm. Thank you so much for coming on the show, Ace.

Ace Stryker

>> Thank you for having me. It's a pleasure.

Rebecca Knight

>> Yeah. So I'm going to start with you, Phil, and talking about the shortage that few people see, and maybe even quite know is going on, because there ... Most people outside of infrastructure perhaps don't realize that there is a flash shortage underway. How serious is it, and why isn't this something that resolves in a traditional or typical product cycle?

Phil Manez

>> Yeah, I think it's very serious. I think when we've seen shortages before, you kind of have a class of company that's able to cut the line and not feel that shortage. We're not seeing that right now. We're seeing the biggest customers on the planet not be able to get the allocation that they need, both for kind of business as usual growth, and at the same time, trying to bring all these new applications and AI into their environment. So it's serious, and it's something we expect to really play out through this year, potentially really into maybe the middle of next year. I'm sorry, what was the second question?

Rebecca Knight

>> It was why isn't this something that resolves in the typical product cycle?

Phil Manez

>> Yeah. I think when you see something like this, obviously suppliers look at growth projections, and we're just seeing a snowball of AI effect here, right? Where the growth is just not something that you can forecast, because the market adoption of AI, the investment of AI, is not something that we have any kind of history to compare it to. So obviously when you're building capacity in, you're planning for some level of growth, and maybe some slack in there, and it just got chewed through so fast, based on the snowballing effect of AI infrastructure and the advancement. And it just adds, AI adds new types of data, on top of the data that it's using already. So it really is just this exponential growth curve that I don't know how anyone would have planned for.

Dave Vellante

>> And I think that the suppliers are obviously somewhat cautious, or maybe skeptical. As somebody who used to follow very closely the disk drive business when there were 80 disk drive manufacturers, it was a highly cyclical business, and it was always these boom and bust cycles. Actually, I remember doing analysis over a decade, and it was a negative cashflow business, but one of the executives in the business once told me that if you're not there when the cycle hits, it's actually worse than in a situation where it's over capacity. And so now, everybody's in the same situation. Whereas before, if one company was there and others weren't, they would run the table. Now everybody, right? Ace is in this, basically the same situation, trying to spin up the fabs, get those going as fast as possible, but it doesn't look like the demand is waning anytime soon. What are you seeing in the marketplace?

Ace Stryker

>> You're absolutely right. Yeah. Anybody who's been following the memory game for a while has seen these peaks and valleys and is kind of used to these cycles, right? By all accounts, what's happening here is something fundamentally different and structural to where we're headed in the future. If you rewind a couple of years, there was a lot of prognosticating about the amount of data going into training foundation models. And that was a big deal, and where are we going to put all that? And then we sort of maxed out on that. And you had these model developers saying, "Well, there's no more public data really available on the planet to throw at these models, so we got to work with what we have." But what we've seen since then is an explosion, not on the training side, but on the inference side, really driving massive demand for DRAM and NAND bits. That's a factor of, obviously, adoption, more and more applications, AI solving problems and people using it more often. In many cases, AI agents using AI models, and generating prompts, and through API accesses, placing demand on AI systems more frequently than humans. But then you also have some of the sort of elements of the data pipeline within inference that are really generating a ton of incremental data as well. And by that, I mean things like retrieval augmented generation. You have to vectorize all this data and make it quickly searchable and accessible by AI models. All that has a storage cost, it's got to live somewhere. The big thing in the news recently has been inference context and the KV cache, right? I think we all heard or learned about the ICMSP announcement from Nvidia at CES. These models with these context windows that are just growing and growing, and these longer loops, more iterations in a given interaction with a model, all of that has incredible storage implications, and it does not appear that this is a cyclical thing, right? That this is likely to wane anytime soon. And so that's where we find ourselves in 2026, in what's been called the storage and memory dry year, or the memory Armageddon. I've heard all sorts of clever names for this thing. But it's real and it feels like it's here to stay.

Dave Vellante

>> I call it the memory YOLO moment. You just got to go for it.

Ace Stryker

>> That's good.

Dave Vellante

>> Damn the torpedoes. So, okay. So Phil, could we talk about the architecture that you guys have? You guys announced, I think, a NAND, or it was flash reclamation program weeks or months ago. I can't remember now. What is it about the architecture, the DASE architecture that allows you to be more efficient? Can you take us through that and explain to the audience what that's all about?

Phil Manez

>> Yeah, absolutely. So we recently announced our VAST Amplify program, which is allowing customers to take flash they've already deployed in more legacy, less inefficient infrastructures, and bring it into VAST. And for us, the efficiency really comes from ... I look at it in three areas. One is that you have capacity that you always lose to protecting from failures. And when you take a legacy model designed for something like spinning disk where you're going to see a high rate of failures, and based on the limitations of those architectures and bring it over to flash, you lose a lot of your flash before you've even stored any data. So we have an incredibly efficient protection scheme. It has a high level of resilience, so you're going to be able to lose more SSD simultaneously than these old models. But at scale, we're only giving up less than 3% of the capacity to protection. And if you look at a legacy data lake, you could be giving up two-thirds of your capacity before you store anything, waiting for failures. And even a good system is more like 20 to 25. So before we're even kind of doing really cool stuff, we're just giving up less overhead by leveraging a more modern protection scheme that is fit for our architecture and fit for flash.

Dave Vellante

>> Without having to clean up the wasted space, and go into the attic. Nobody wants to clean up their attic or their closets, their garage, whatever. You always put that off. "I'll do it next week, wait for a rainy day." But this is sort of automated. It just comes-

Phil Manez

>> Absolutely....

Dave Vellante

>> with the package.

Phil Manez

>> Part of the architecture, you don't have to do anything, you don't have to create protection schemes or anything. So that's kind of step one. It's kind of the easiest. And depending on the type of technology you're using, that could give you almost three times more capacity per raw flash that you buy. The next thing that we do is we have some really compelling data reduction technologies. So we use compression, we use dedupe. Those have kind of been around for a long time, but we added something called similarity data reduction on top of it. And this is really important for unstructured data, because similarity takes two different chunks of data and says, "Are you mostly the same?" And if you look at using dedupe in unstructured data, what I find is a bunch of chunks of data that are almost exactly alike, but not exactly alike. And dedupe, it's either one or zero. Similarity says, "Hey, there's a lot of commonality here. Can we take advantage of that?" So we do that, and that gives us typically double or more data reduction than every other technology on the planet. So now we talk about, after I've done my overhead, we can store two times more capacity or more than you already could.

Dave Vellante

>> And that's because, wait, when you reconstitute, you can infer the delta, or ...

Phil Manez

>> So all of this is enabled by our architecture, right? So we have this distributed architecture, we have these stateless compute nodes that get to see every single device and they see a global shared metadata structure, right? So we don't have to worry about doing data reduction and creating this horrible performance experience. Everyone gets to see all the data that's already in the system already. And essentially what we do is we run a series of algorithms when the data lands. It's kind of like a vector similarity search. We were doing this years ago that said, "Hey, these two chunks of data are basically similar." So if I was looking for two cats, I would find that these three things have a lot in common. We do something similar for data reduction.

Dave Vellante

>> So what are the economics of this? Why is this so important? How have prices changed over the last, even several months? But then if we go back a year ago, we're talking about significant price increases and a lack of supply, right? Can you give us, however you can frame that, quantify it in subjective terms or actual numbers?

Ace Stryker

>> Yeah. Well, the demands that AI is placing on hardware infrastructure really show up in both DRAM and NAND. And obviously those are two components that play in very different price ranges, but both of them are kind of hit by this supply constraint, right? And we've seen prices multiply quarter over quarter quite substantially. You've seen the consequence of that in some cases in the public stock prices of some of these folks that make this stuff too, right? They're up like a thousand percent in the last six months, right? It's a very real situation. And the challenge is increasing capacity is not a short term proposition. Building a new NAND fab or DRAM fab is a three-year process, right? And-

Dave Vellante

>> And billions.

Ace Stryker

>> And billions and billions of dollars. And folks, the demand is there today, right? And so that's where things like Amplify get us so excited, right? Because it's really about how can we get more efficient with the resources that we do have today? How can we recover some of that lost capacity that maybe we're spending on sort of the traditional duplication and erasure coding approaches, where VAST can help us recover some of that space and do more with less. It's like one factor of general sort of name of the game in AI outcomes for the last few years, which has been efficiency across the board. It's been energy efficiency, it's been space efficiency, right? All of these things have been kind of the critical constraints for folks wanting to deploy AI to solve problems. And we can add to that now, the NAND and the DRAM supply efficiency, or capacity efficiency, in terms of making better use of what you can get from suppliers in a constrained environment, or, like Amplify is looking at, what's already deployed out there that maybe isn't being used as optimally as it could be.

Rebecca Knight

>> So the demands that AI is placing on capacity are unprecedented. You can't procure your way out of this, and it seems to require what you're describing, a real shift in how enterprises think about storage efficiency. I'm curious about the conversations that you're having here at VAST Forward. Are customers ... How much do they recognize this issue, and in terms of what you're telling them about the flexibility, the cost savings, are their eyes lighting up, or do they even quite realize that this issue exists to the extent that it does?

Phil Manez

>> Yeah. I think we've seen the awareness come in waves, right? So I think even back when maybe we were around super compute time, some of the AI labs were getting really spooked. And then I would say, as you kind of got closer to the end of the year, the bigger enterprises that have tied in supplies and they're making big purchases started really getting spooked. And now it feels like everyone's aware, but I do think it's a forcing function to think about efficiency. So I think some of the other things that we've looked at in terms of efficiency, it was kind of the third thing that we're able to do with VAST is a platform that can support a variety of workloads. So, and if you look at a customer's environment, they've got capacity all over the place, but they can only use it for one thing. So I've got legacy data lake capacity. I can't support AI and inference with that. I've got, this only does this one protocol. Maybe I don't have multi-tenant, so I've built islands for my tenants. So when you think about efficiency, it's not just erasure coding and data reduction, it's around capacity planning, and workload flexibility, and then looking at our product and saying, "If I took one system that's horribly inefficient, that's my data lake," that's kind of where we see a lot of the worst offenders of using triplication and poor data reduction and be able to take that and give you enough capacity to grow for a couple years. And at the same time, I can now run AI workloads on that infrastructure. That is something that I think that they haven't considered in a way before, because it's not just business as usual versus investment, it's like, delete data, or stop storing stuff, or push off AI initiatives that we might need to survive as an organization. So I think it's very much eye-opening, and from what I've seen in the conversations that we're having with customers, Amplify feels like a lifeline that they didn't even consider possible before.

Dave Vellante

>> Well, when we've seen situations like this in the past, they've generally been short-lived and everybody forgets about it. Prices drop, everything becomes abundant, and then that was yesterday's news. To Ace's point, it doesn't feel like this is, because of the expense of building new fabs, the ... I mean, you know what the capacity is, unless the demand tanks, which there seems to be no indication of that given what the hyperscalers just announced, it feels like it's going to go well into 2027. So this program is not just a short-lived thing. I mean, and there's ongoing benefits, which are wonderful. But in the next, I don't know, 12 to 18 months, seems like this is going to bring real value to people. So how did you guys come together with this? What's the partnership like? Is there any kind of other KV cache magic that you guys are working on that you can talk about?

Phil Manez

>> Yeah. I would say, just from how the program came together, we've been working to bring some of the most amazing new systems in, right? A lot of what we've done over AI for the last couple of years, VAST and Solidigm, is building some of the world's biggest supercomputers, right? When we look at the biggest GPU deployments working together, and as everything that VAST does that's pretty smart, customers brought us the idea. So actually before the supply chain, we had a couple major customers come to us and say, "We've got some things that are just not really working at scale." We typically solve problems at scale, and there's a lot of things that don't work at scale. It doesn't matter about efficiency. It's like, "Can you do this really big?" And they came to us and said, "We've got things that aren't working. We'd like to get more value out of it. Can you help us?" It wasn't about supply, it wasn't about cost, it was about scale. And then that enabled us to have a process where we could look and work together. And now with Solidigm understanding that they had a lot of customers deploying capacity, knowing that they're not in a position to maybe sell them as much as they'd like. I think good job, both of us. We sold the world out of flash working together, kudos to us. But they're working with their customers, even if it's not about selling them something today, it's about making sure they're getting the most out of what they've sold. So they know what customers are ringing in flash, and is this a way that they can get more out of it?

Rebecca Knight

>> Powerful note to end on. Phil, ace, thank you both so much for coming on theCUBE. A really great conversation.

Phil Manez

>> Thank you for having us.

Ace Stryker

>> A pleasure. Thank you.

Dave Vellante

>> Thank you.

Rebecca Knight

>> I'm Rebecca Knight for Dave Vellante. Stay tuned for more of theCUBE's live coverage of VAST Forward, presented by Solidigm. You're watching theCUBE, the leader in tech enterprise news and analysis.