The Data Center Race Behind AI: Solidigm's SVP on Why Storage, GPUs, and Scale Matter

If you're ready to scale AI and data intensive workloads without letting storage become the bottleneck, solidigm is built for, what's next? With enterprise SSD solutions designed for high capacity, performance and efficiency across the data center and edge, solidigm helps organizations store more, move faster and make every watt count. Learn more@solidigm.com that's S O L I D I G M.com solidigm.com hey everyone. Welcome to another episode of Future Ready Leadership. My guest today is Greg Mattson. He's the senior Vice president and head of marketing and products at solidigm. Greg, thank you for joining me. Hey Jacob, thank you so much for having me. I'm excited to be here. Yeah, likewise. Before we jump into the fun conversation stuff, can you give people some background information about solidigm? What do you guys do and what do you specifically do there? Okay, so first is, yeah, I'm the head of products and marketing at solidigm. So in the product role I am out working with our customers, ecosystem partners and kind of in the industry standards world to understand their storage requirements and you know, to help build the best products to help solve their infrastructure needs. And in the marketing type of role, I take those products, enable our sales team to deliver those products to market and get them designed in our customers. So kind of a full end to end ownership of our product line. Solidigm is the number one provider of high capacity storage solutions for AI and cloud computing as well as enterprise computing. And so We've been making SSDs for upwards of 15 years and are really solidly in that market kind of globally. Very cool. And obviously this is very relevant to the AI space. And even before we jumped on the call, I think it was, my son was asking me a question about this. He's turning 6 in just a couple of weeks. And you know, I have a daughter who's nine and they, they're familiar with AI and sort of like, you know, if we have a question, let's ask AI. But then they were asking me these like a high level, like what happens when you put your information into the computer and you ask Cloud or ChatGPT, like where, what's going on? Like where, why do you need a data center? Why do you need storage? Why like what is going on? What you. Cause obviously everyone's hearing the news about all these data center build outs and what the infrastructure means, but a lot of people don't understand conceptually like what actually happens? How does AI actually work? Why do you need these things. I obviously put this into Claude and so I'll give you kind of like the five steps that it gave me and you let me know sort of where the storage piece fits in here and why it's so critical. So step one, when you put something into, whether it's your phone or your computer, you obviously have the information that leaves your device. It gets securely encrypted and it goes to some sort of server endpoint. Then it has something called load balancing. So your request hits this load balancer and it figures out, okay, from all the available servers that are out there, where do we have space to actually put your request to give you back information. So the load balancer kind of sends things out there, then it gets obviously loaded onto these GPUs or TPUs or whatever that's out there. Your request then gets processed and so you get the tokens, which are basically three quarters of a word that gets broken into numbers, that gets run through a neural network, decoded back into text and sent back to you within a matter of milliseconds. And it's really mind boggling when you think about what happens when you use these AI tools. So how does the storage element fit in? Because a lot of the AI conversations we obviously talk about, you know, the nvidias, you know, the cooling, the data center, the infrastructure, this and that. Why do we need the storage piece first? If you're a six year old and nine year olds are asking about AI, that's pretty impressive, but I'm not sure if I would give them the answer you just described because I didn't tell them this yet. I'm testing this on you first. Well, we fit in. So that was kind of a consumer AI example, and enterprise AI works similarly, although I think you have a little bit more dedicated and known compute space allocated to you. But where we fit in is really in the data center side of things. And so storage is really critical in both training and inference. And when you're talking to ChatGPT, that's a trained model, so that model has already gone through massive amounts of computing to create the model. Yeah, maybe you can unpack what is trained, what is inference, because a lot of people are probably not familiar with any of these concepts, how these things work. So anything that you can unpack as much as possible, I think that'd be great. Yeah. So I mean, again, a normal person would not be involved in training at all, but they're interacting with a model that has been trained. So ChatGPT, Claude, etc. And those are really done with massive amounts of GPUs. But those GPUs need something to fuel them and to train a model. And that's data. And data is stored in storage. That's the fundamental purpose of storage. And so they need to have massive amounts of data available to them while running through their training processes to create the end model. Right now the modern models are now trillions of parameters and vectors that they sort through and process. And all of that data needs to be accessible to GPUs in a very real time kind of fashion. And so you really have two main areas of storage in the training process. One is kind of very close to the GPU and the GPU server. And what is a GPU for people? GPU is a graphical processor unit. It's really the product that Nvidia builds. It's really the ultra high performance accelerated compute engine that has the capabilities to process this amount, massive amount. So it's the brain, it's the everything behind. Yeah, call it the brain, but it's also the engine. Right, but engines need fuel and brains think and all that needs information to be able to process. And that's really where storage comes in. And so there's kind of GPU storage that's really close and quickly accessed by the GPU during training. And there's also network attached storage, which is massive amounts of data available, scraped from anywhere on the Internet that you can possibly think of. And the more data, the better in terms of model creation. Now that's the model, that's the part that nobody ever. That's training, creating the model. People interface with the model ChatGPT, but they don't know that the training even existed. When people start using the models, whether it's a person, consumer or a worker, they're actually starting to do kind of lightweight inference. And that's really where the model is thinking, it's drawing conclusions, it's processing different amounts of information to create the answer that you talked about. Got it. Okay. Because I know there's like two types, right? When they do like the benchmarks, there is with tools and without tools. And so with tools for people not familiar basically means these AI models have access to the Internet where they're not just kind of retrieving information that's stored, but they're actually searching for information that's available. So when you talk about storage, what is actually stored on these devices in these data centers. So if you have for example, a bunch of solidigm units in some sort of a massive data center, is it just storing all human data, all information, everything from like, why is the sky blue? To how do I get a stain out of a coffee table? Are we talking like all of that information is stored in these devices? In the training use case, yes, in the inference use case. Now this model has been trained already. And then you add in your own data. Say if you're working within the enterprise, you could have your own company data. That data would be stored in the storage and then added into that model. And so when you're querying, you're asking that model to not only use its data, but you can load in specific, whether it's analyst reports, whether it's a code repository, if you're writing software, whether it's your financial reporting system data or your human resources data. And you can point your inference engine to be able to draw from these different sources. And that data is stored in the storage in the data center. Interesting. Can you give us just kind of conceptually, how much data are we talking about? Like, like when we say storage, right? I mean, most people think of, you know, external hard drives. You know, everyone has external hard drives. We have hard drives in our computers. How much data are we looking at here? And I don't even know how to frame that question, whether it's how much data is in a typical data center, how much data storage do we need collectively for AI, but how maybe you can help frame how much data are we talking about when we talk about storage? It's actually quite incredible if I just take an Nvidia system as a reference because those are a. There's public available information to reference. And also it's kind of the name out there that you can, it's a product you can go buy too. You can't really buy Google tpus if you're an enterprise, right. You can only rent them. And Nvidia is very prescriptive with the, actually the amount of storage that they need for the GPU accessed storage. And so they need about 15 terabytes for the GPU local storage and another 15 terabytes for their context memory extension, which is kind of the thinking layer, the active memory portion while you're doing inference. And then there's another portion that's roughly about 30 terabytes. It can range though, depending on the user of network attached storage. So this is additional information for the GPU server to go out and retrieve and come back and process. And so our model shows that there's roughly 60 terabytes of data per GPU, which is insane Phenomenal amount and much, much higher than there ever has been coming kind of in your traditional cloud computer. Yeah. So 60 terabytes per GPU. And how many GPUs is the company or is a data center potentially using or tapping into? The biggest single new data center I believe was announced earlier this year and it was the first grounds up gigawatt scale data center. It has one gigawatt and that used, this is public information. They, they had about 550,000 GPUs. Oh my God. And so if you extrapolate that out, that turns out to be about 25 exabytes worth of storage for that one data center. Yeah. How much like you know, people hear that. I don't even know how large is an exabyte. Like if we were to think of all human information that's ever been put together, is that even an exabyte? Oh yeah, yeah, yeah, it is. Okay, but you can't, it's so hard to fathom. I couldn't give you a. Yeah, like it's just ridiculously massive. And this is just for one data center. Right, right, right. One gigawatt scale data center. Now they're building out many data centers that are even bigger than that, multiple gigawatts. And there's hundreds of gigabytes, I think Maybe in between 150 and 180 gigawatts of announced AI focused data centers capacity going to be brought online by like 2030. So, so can you give, give people some context around the data center? Like why do we need a dentist data center? Why are so many of these things being built? Because for most people who are out there that have been using these AI tools, you know, as far as how most of us think about it, we think about it kind of interacting through our laptop or through our phone. And so a lot of people in the news, you hear about all this news, you know, data centers being built, people protesting data centers. But what exactly what is happening in these data centers? Why do we need them? And why should most people actually care about them? Well first the Internet, since the dawn of the Internet was built in data centers. And so this is not a new phenomenon, Right. All your cloud computing, every device that you have, phone, gaming console, PC, all connect to the Internet all through data centers. Today these have just been traditionally built in a, you know, with kind of a CPU focused non accelerated compute environment. And they're just less resource intensive, but equally big in terms of size than accelerated compute. So AI really is the era of accelerated compute. It's where new generation of ultra high performance compute has come online. And those things take a lot more power than the traditional Internet, but they're doing a lot more right. They're processing much more information, much more, much more capable than your traditional cloud computing infrastructure. And the same thing even happens, not only just in the cloud and the huge data centers, but even if you're a hospital and you want to do medical image processing, you have a small data center in your hospital. If you're an office worker, you have a small rack of data center within kind of a medium to large business, the same challenges apply. Right. You need high compute performance, you need to be power and space efficient. And the more you can pack into a smaller amount is equally as important in these small data centers. And I use data centers kind of in quotes, but they're like a closet. It could be a small room compared to these GPUs in there. Yeah, yeah, but think about the world is accessing, billion people are accessing ChatGPT now and that takes a lot of compute. Yeah. If a billion people are doing something on it. And not to mention we haven't even. We're starting in the kind of agentic area where for companies now one person can have 1,000 agents working on their behalf 247 and. And they need access to this compute all the time. Yeah. Okay. So essentially for people kind of from a, I guess a basic conceptual standpoint, the data center is the energy required. It's kind of like the mitochondria in your cell where if you don't have it, the cell dies. So you need these data centers to basically power all these devices, all these AI tools, all the agents. Claude, ChatGPT, you need all of that stuff being powered by these data centers. So if you don't have these data centers, I guess that's where you start to see some of these throttling of usage, throttling of tokens, downtimes, inability to use some of these tools. Is that kind of the downside? If you didn't have data centers, you wouldn't have AI, period. Yeah, you just couldn't do the level of compute necessary to train these models and then to really do deep levels of inference on the models where you could do some kind of accelerated interesting things at a very small scale. But the power or, sorry, the benefit from these advanced technologies is really when you link many, many of them and thousands in the case I just talked about hundreds of thousands of these together and then you link them very much together with high speed networking with tons of storage attached and that kind of has to be co located in a space and that space is the data center. And so you can't bring high power infrastructure to every single spot in the world. So you need to concentrate it, you need to concentrate the cooling. And then you get the biggest benefit by having massive amounts of compute together. That's what's really enabled the kind of revolution of AI today. Do you have a sense of like, how are we doing in terms of energy consumption versus production? Like, are we, are we behind? How far behind are we? You know, how many more data centers do we need? How many more, like, are we where we need to be with all these things? From what I've been able to tell through lots of research and talking with customers is we're just getting started. We're at the very early phases of the AI infrastructure build out. I mean, this is going to be a multi decade build out, just like the Internet was. Wow. Multi decade. The Internet took 20 years plus to get to where it is today, where everything we do is kind of online. And these things are, these are big investments. They take a lot of infrastructure and just a lot of manufacturing and construction to put them together. So yes, it's going to take a long time. Yeah. Such an interesting space to be following. I mean, I don't remember, I mean you've been in this space for a while. I don't remember such a massive infrastructure build out for anything that we've ever seen. As far as just the number of electricians, construction jobs. It's just been fascinating to see how much is being put in here. And I think this year, what is it, around 700 or $800 billion is being put in to AI infrastructure build out just from the nvidias, the chatgpts, the anthropics. I mean, I wouldn't be surprised if it hit over a trillion dollars. Have you seen anything like this happen before in your entire life? This is not near this. I mean this is. Yeah, nothing so far has been near the scale of today. Even though there was a massive build out of fiber infrastructure, communications infrastructure and data centers in the, in the 90s and 2000s for cloud. This is, you know, order of magnitude or even more bigger than that. Yeah. And the reason is, is because it's proven that who has the infrastructure will end up winning. Right. And AI has the power to really revolutionize. It's really kind of the next industrial revolution. Right. Where it's going to really revolutionize how we work and the productivity that everyone gets out of, out of their employees and their companies and then really enable the next wave of breakthroughs that have never, we can't even dream of yet. Yeah, I saw a video that you did, this may have been a few months ago, so I wouldn't be surprised if you even exceeded this. But you were holding something that looked like a deck of cards and I think you said it was 120 terabytes. 122 terabytes. Yeah, we lost. Is that possible like on a deck of cards? That's how much that little kind of little thing was storing? Yes. Yeah. And it's, it was an incredible feat of engineering. Where we have actually kind of at the first is a solid state drive is built up of hundreds of memory chips inside of the drive as well as some other components that help control the memory chip. And so we started with the highest density memory chips built on our floating gate NAND technology. And we did really what we call extreme co engineering in terms of engineering every piece of that drive to be the most space efficient but high capacity product that we could possibly engineer. And we came out with that drive well over a year in advance of the next leading competitor. Crazy. I remember having the like bigger, the bigger drives, they were not quite the size of a shoebox but a little bit less and you would be lucky if you could get like 20, 30 gigs on there. And they were pretty meaty pieces. So the fact that you're getting 120 terabytes in something that looks like a deck of cards and it's pretty crazy to think about, like if you had something the size of this table for example that I'm sitting in, how much storage and data you could probably put on there would be astronomical. If you're scaling AI and data heavy workloads don't let storage slow you down. Solidigm's enterprise SSDs deliver high capacity, speed and efficiency from the data center to the edge. So you can store more, move faster and maximize every watt. Learn more@solidigm.com that's S O L I D I G M.com solidigm.com yeah, and that's it. But that's why the data centers that we're talking about right. Are consuming these drives like just as fast as we can possibly build them. In fact they want them even faster than we can build them. And the reason is because to get the most amount of data in the smallest amount of space and so they really revolutionize based on previous technologies or compared to previous technologies, you can shrink the amount of storage space and the power needed for that storage by as much as nine times by using these super dense drives. Yeah, it's pretty wild stuff. Yeah. Okay, so I think we established kind of from like the data center perspective, the role that you guys play in there. So you're basically handling all the storage that these AI models are using to kind of give to be trained, where companies can house their information so employees can kind of pull that information. So it's very clear, I think, why the storage piece is really important. I'm curious just from an internal perspective at Solidaim, are you guys using AI yourselves? And if so, are you able to share at all about what you're doing internally with it? Yeah, absolutely. First is what we've learned both through trial and error and research and then success is that first is kind of to adopt AI in the company successfully. We really have to start at the top and really create a culture of this is really a big change, change management exercise actually that you didn't think about at the beginning, to be frank, where you have to get people's mindsets able to think about changing. You have to enable them with the tools to change and understand how to basically redesign processes in an AI first environment. That really takes kind of a different leadership mindset. And then encouraging employees to be curious, willing to try, give them the time to be able to do their job, but also start figuring out how to do their job. The AI first way we've had last year we started it, we just actually completed our second Innovation Day challenge where we challenge our employees to come up with ways to use AI to improve their the way they do work. And really the excitement around that has been remarkable and some of the results have been remarkable. From AI powered robotic things to do some very repetitive tasks that are involved in our engineering processes, to automated software testing, to automated code writing. My team does uses AI. We're more of the business focused team. We use AI to do really quick statistical analysis on market forecasts or a collection of product requirements into one central location instead of multiple Excel sheets. And we can iterate and do scenario planning on which products should we build first and second and which customers first and second. And so it's early days, but we're seeing a lot of early successes. Once we got over that, learned about that, really that different mindset that you have to have kind of coming into it. Yeah. Have you seen, and I'm sure you've seen some of the reports out there, the whole like ROI debate is a Big area of question. Whereas if companies are seeing it on an enterprise side or if they're just seeing it on task based side, I mean it seems like based on what you're saying, you're clearly using it at least for your team on forecasts, your developers are using it for kind of coding. So it seems like you have the task based roi. But are you thinking about how do you measure the impact across the enterprise or is that not maybe like top of mind for you guys at the moment? Yeah, we're trying to really focus on is productivity and, and speed to results. Right. A shrinking timeline, say a development timeline for our product so that we can bring, if we can reduce the product delivery time mine by a quarter, two quarters and bring a new product to market that our customers want, then it's very measurable for us in terms of both revenue, employee efficiency and customer satisfaction. And so I don't think we're there. I wouldn't call us the fully functioning, mature AI organization yet. I don't know if there's any company that would fit that bill. Everyone is still, I think, experimenting and playing around with it and trying to figure out what do we use it for, how do we use it, should we use agents, should we not use agents? Every day you hear a story. There was one that came out last week where this company gave this agentic tool access to its code database and it just wiped everything out in nine seconds. Yes. Did you see that one? And there was no governance rules and things you want to put in place for sure. Yeah. And they said, well, we had governance rules in place but it overrode everything. And then it apologized for doing it. And then there was a case before that, a big law firm that was going to court for a criminal case. And it turns out that all a lot of the documents that this law firm was submitting, you know, they used AI and AI hallucinated and like made up cases and made up references. And they had to issue this really big apology to the judge and say we're sorry, we've been using AI. So I don't know, you know, even for me, I find that if I'm doing it, if I'm using some sort of, if I need a complex task done, sometimes I find that it does better in a complex task than it does in a simple task. Like if I need to brainstorm an idea for a new keynote that I'm delivering to an organization, it'll do a fantastic job. But if I want to find the top future of work, news related Stories of the day. It has a hard time doing that. Like, sometimes it'll make up your URLs, sometimes it'll send me stuff that it thinks that I want. I have to keep repeating that I only want stories from today, not from like five days ago. It's like the basic stuff I have more of a challenge with, but if I want something complex done, it does a great job. It's kind of a weird paradox. Yeah, that's interesting. I haven't seen the high level of hallucinations happening inside the environment. We have our tools inside of our. And they're vetted first. Right. We don't use. We have corporate versions of all the different models. We can't put company data, for example, in a personal account. It's been fairly productive for us, to be honest, so far, once we've figured out the right process and where to house the data and kind of rules around it. So I was actually. I do hear about those stories all the time. Yes, you mentioned something interesting around kind of housing data, because I know a lot of organizations are moving out of kind of the pilot phase into more of the production phase. But at least I'm hearing. I don't know if you're hearing this as well, but a lot of companies are having a hard time getting into that production phase just because of, you know, years and decades of bad infrastructure, of bad, I don't know, not standardizing data, maybe not having the storage requirements in place. Because a lot of people just think if you give everyone access, give everyone a ChatGPT license, like, you're done, right? Just pay the 20 bucks a month per user, everyone has ChatGPT, and you're off to the races. But few companies are really understanding that you need a very solid infrastructure internally for employees to take advantage of these tools, both from like a data storage standpoint, from a data standardization standpoint, just like having that infrastructure in place. Are you seeing that as well? And our company's kind of struggling. Like, you know, say you have the Ferrari engine, but you don't have the tires on the car. There's no seat where the driver's supposed to be sitting. So you really can't take advantage of the car. Well, that's a good analogy because we talked about kind of at that data center scale level. The same thing's true in enterprise, where if you have the models, you have GPU compute that's super accelerated. If that can't be fed with data near real time, then your GPUs sit idle or your workers for example, if they're doing something, they have a project going and they run out of memory, but they have no place to expand that memory. As their work goes on, boom, it goes away and then the GPU has to stop, recompute. You can get back to where you were, but it's work and it's not productive. And so really thinking about storage and modern storage technologies and taking the time to make sure that your future, your planning for the future, not utilizing past technologies is really important to get the most out of your AI infrastructure. So for us, our recommendation is high capacity, high performance storage. Your AI will use the capacity. And so what we're seeing both, you know, again at small scale and big scale is the challenge is very similar. And you know, we partner with some of the leading storage system providers for both smaller enterprises, medium enterprises, I mean down to research universities all the way through Fortune 500, but also even the hyperscaler and they kind of use very similar technology. It's just the amount, so similar technology, but it's just the amount that they're using. Where do you think this is going? So we've seen I think progress and evolution just in the, I don't know, just even in this year it's been wild. You know, if you just think about this past year, we've seen what, three, four, five new models come out, new plugins come out. I don't know how many data center announcements have been announced. Almost a trillion dollars in investment and we're not even halfway through the year. So if you had to kind of peer into your crystal ball and just anticipate where you think we're going to go through the end of this year, even maybe next year, you know, scary to think of that we'll be on what, ChatGPT 6.2 next year, Cloud Opus 5.3. Right. I mean these models, with each model that comes out, you really do notice an evolution in how these things work and how they operate. So how are you and your team thinking about where this is going to go? I mean even six months to a year from now? Yeah, first is it's not going to slow down in that time frame at all. I don't think it's going to slow down for many years. We talked about the multi decade build out. Of course it'll be, it'll probably be more, you know, it'll get asymptotic at some point. But you know, with the 150 to 180 gigawatts of data centers have been announced, very is just Scratching the surface in terms of how many have been built so far. And so we're planning our products, you know, increasing our production capacity. We're really trying to, you know, get ahead of the ball in terms of which products are going to be the key for the next generation of AI compute. But I don't think it's slowing down anytime soon. In fact, you can see it in the, the model providers, they filed for their IPOs and they still are both forecasting hundreds of trillions or billions, sorry, hundreds of billions of dollars just for training. And that's just a portion of their capex. I mean inference is really where it's at and so there's going to be hundreds of more billions and trillions of AI infrastructure built out for inference. It's wild. I saw, I don't know if you saw the article. The CTO of Uber came out and he basically said, well, we've, you know, in four months we've already exceeded our entire AI budget for the year. Just, I think in, I think it wasn't just in token costs, which is pretty insane to think about. Yeah, I've heard new job offers, you know, for employees are like, well, what's my token budget? That's the second question. What's my salary? Right. That's so bizarre. I don't know, is that going to be kind of like a new standard? I mean, it's a good question. I mean, I can't claim to predict the future on it. I can, you know, I focus on my storage and try to understand what the software is doing to storage and help co optimize with our storage, you know, with the software providers to make sure our products work well out of the box. Right. Where's storage going to go next? So if you have 122 terabytes and kind of a deck of cards, where is that going to go in a year, in two years? Like are you going to have like, I don't know, something as big as this SD card one day that can have, I don't know, hundred terabytes on something that small? Well, I mean first is that the deck of cards example will double and probably double again. Yeah. In the next couple of years. So we're working on those products. Incredible. The other thing that's happening though is there's kind of, it's kind of shaping, you know, we call it like a K shaped economy today. Well, it's a K shaped evolution of what's happening in storage where there's one vector of very high performance or sorry, Very high capacity like we talked about, but also very high performance. And so this accelerated compute infrastructure is driving a level of performance that the CPU systems really could never drive. And so we're having to architect kind of in both vectors and innovate in both vectors, both for exceptional performance, to really almost scale out what I call the memory of the system, extend what HBM and DRAM can do into the storage. And then on the capacity side. Yeah, just keep the higher capacity. The more power efficient that storage can be, the better for that network attached storage. So it's kind of both innovating on both vectors. Yeah. Is it getting close to, I mean, if you're going to double or triple storage, is it getting close to being kind of like on the atomic scale for some of the components that go into these solid state drives? I mean, like, I'm just trying to imagine how do you assemble a solid state drive that can fit 250 terabytes on it that's the size of a stack of cards. And it's just, I mean you're talking about microscopic sized components and pieces, right? Like how well the atomic level comes in the memory chip. And yes, we're scaling the memory chips with higher layers, more bits per cell. So that means that I can store additional data bits for every memory cell that there are pretty crazy stuff that will evolve. The stack up of memory, what we call memory layers is actually just more rows of, you know, more layers of memory cells. And so that's happening. So each individual memory die is getting more dense. And then the innovations. Yes. On the form factor side are to make sure that we can continually integrate more componentry into a smaller amount of space. Yeah, it's pretty crazy. It's really remarkable engineering. Yeah. And I'm assuming it'd be fun to like even look in the factory where all these things are assembled in your lab where you have all these machines putting these things together. It's pretty crazy stuff. Yeah, it's exciting actually. If you like mechanical things. Watching these things get built and seeing the designs and the thought processes and the innovations are pretty cool. Yeah, I can imagine. Is there anything that we didn't touch on that you think we should as far as either storage or data centers or future of work or anything that you're doing internally that you want people to know about, even if it's how you're using AI? Yeah. I mean, again, we're a storage company, we like for people to think about storage. This is really almost the year of storage, where it's now it can't be an afterthought. It has to be designed into your system and architected up front. And we're seeing that again at all customer sizes and types where AI is. Data is integral for AI. AI doesn't work. GPUs don't work without SSDs, AI doesn't work without data. And so it's really now a critical piece is to make sure that your data infrastructure is modern, but also your employees and your future workforce. As a senior leader in the company, we are wanting to make sure that the education system itself are educating people in a way that will be useful for them. Coming into our company and learning how to not only be an engineer or a marketer or a finance person, but how to do that in an AI first way. And so that takes actually re engineering of the education system too. And we're engaging with our university network to begin to do that. Interesting. Can you share a little bit more about what you know, like if you and Solidigm had. Because I hear this a lot too. So if you and Solidigm had your way of kind of changing the education system so that when people graduate, they would be prepared to work and live in an AI first world. Is it just a lack of skills in certain areas, is a lack of fluency with the AI tools? Like what's the big gap that you're seeing for a lot of people who are graduating from college as far as entering the workforce? Well, you know, first is there's a varying level of comfort within the education ecosystem to be able to adopt AI. And oh yeah, it was frowned upon for so long you were told not to use it. And if it was banned completely across the board. Yeah, yeah. And you can't say that it's going to be an easy transition. Right. Because you want people to learn the skill and critically think and be able to do things themselves. But if they can't do that in a way that leverages AI, they'll be behind and frankly will be much less useful to employers. And so you kind of have to have both of these skills coming in. Getting educators comfortable with that is. It's an interesting challenge. Getting students to be able to be interested in that also is kind of a new mindset. And so bringing both of those things in is really critical for us as a large employer. But it can be done with partnerships with both the local government, State, local, local governments, with educators and with us as we think about what types of employees we need, we need to actually know that ourselves as well. I think a lot of these educational institutions are going to get really disrupted over the coming years. I mean, like I was saying, I have a 6 year old and a 9 year old. By the time they go to college, you're going to be able to get your own Optimus robot, you know, that is, you know, 100x better than whatever you see now. And you're gonna be able to feed it with all the information that you possibly need. Of course it's gonna have storage, probably a couple of those decks of cards in there and then you kind of have to take a step back and say is it worth even going to college at that point? You know, in 10 years from today, is college gonna be worth it to take out a, you know, spend $200,000 or more for a four year degree when you can get access to any anything that you need and have this personal teacher teaching you anything that you want. So it'll be very interesting to see what the next couple years looks like. I totally agree. And my kids are a little bit ahead of you, but I still wonder about that for them as well. Yeah, next few years will be pretty crazy. Greg, where can people go to learn more about you? Solidigm? Any of the cool stuff that you're working on? Where should people go? Well, you can access our website@solidim.com or if it is easier to remember, storage4ai.com also leads you to our product pages. Also we have many case studies with how we've been working with partners in the industry. Of course you can learn about our products as well as myself and my peers on the management team. Very cool. Greg, thank you so much for taking time out of your day. Really appreciate it. Thank you Jacob. If you're ready to scale AI and data intensive workloads without letting storage become the bottleneck, solidigm is built for. What's next? With enterprise SSD solutions designed for high capacity, performance and efficiency across the data center and edge, solidigm helps organizations store more, move faster and make every watt count. Learn more@solidigm.com that's S O L I D I G M Com Solidime. Com.