What happens when the data your AI project needs is spread across systems built for other jobs? On Across the Tech Pond, David Marshall, Anthony Savvas, and I welcome Molly Presley and Floyd Christopherson from Hammerspace. We discuss why buying compute is only part of the work, particularly when an organization wants to use live data without disrupting the applications and teams that already depend on it.
Molly explains Hammerspace’s approach to giving users a unified view across storage locations and automating where data is placed for a workload. Floyd describes why a catalog or a separate storage tier may solve only one part of that problem. The conversation follows the practical questions: what can the team find, what can it access, and how does the data get to the compute selected for the job?
We also discuss Hammerspace’s Data System announcement at Supercomputing, including a more integrated deployment option, richer metadata, and its OpenFlash Platform. The guests explain how those additions fit with the company’s existing software approach and its emphasis on using infrastructure customers already own. Their argument is that a simpler deployment should help teams spend less time assembling separate tools before they can run useful work.
For leaders assessing another AI investment, the discussion offers questions to take into the buying process. Does the proposal require a separate environment? Which data must move, and why? What remains available to existing applications while the AI workload runs? Before adding more capacity, which part of your data access problem would you want a supplier to demonstrate in your own environment? Share your thoughts with us.
Connect with Molly Presley and Floyd Christopherson from Hammerspace.
[00:00:03] Hi there, welcome to Across the Tech Pond, the US European podcast covering big technology events and announcements, hosted by David Marshall, Neil Hughes and myself Anthony Savas. In this show we have some very special guests. From Hammerspace, a leading player in the hybrid cloud storage market, Emma will have a very special announcement for them to boot. There is a change in cloud storage market as organisations strive to meet their needs around AI-driven cloud storage management.
[00:00:33] In this podcast, we will look at what continued market change means for customers and how, for the industry, technology evolution has become ever more important for delivering what customers need. Before we move on, let's introduce ourselves. David. Hello everyone, I'm David Marshall. I'm the founder and executive editor of VMblog.com.
[00:00:56] I'm also the co-founder and co-host here at Across the Tech Pond and I'm located in Texas here in the US. Neil. Hi everyone, I'm Neil Hughes based in the UK. I run a range of enterprise tech podcasts under the company Tech Talks Network. I also write for Cyber Moves.
[00:01:16] I'm Anthony Savvas. I'm the editor at large at IT Europa, which is a pan-European channel title, writing about all different technologies, including AI, data storage, all the hot areas, Internet of Things, what's happening in the cloud space generally, etc. And I write for IoT now and also the Agile Telco amongst others.
[00:01:37] As mentioned, we have some very special guests and they're the senior vice president and chief marketing officer of Hammerspace and the VP of product marketing at Hammerspace. Please introduce yourselves. I'll get started first. I'm Molly Presley and I'm responsible for global marketing here at Hammerspace. And I'm Floyd Christopherson. I'm responsible for product marketing, which I always like to say is not product and not marketing,
[00:02:04] but it's that glue that binds them both together. Very good. Well said. Welcome, Molly and Floyd. I'll just come to one of you here. Molly, for the members of the audience that aren't so familiar with Hammerspace, you're probably sick of explaining this, but it's essential in this case. Could you first explain what Hammerspace does? I will. And to give kind of a colorful version of the story,
[00:02:29] the term Hammerspace, if any of you are into comics, Japanese anima, that idea of Bugs Bunny pulling many large things out of a very small hat, or Mary Poppins pulling things out of her purse that shouldn't fit otherwise. That's where the name of the company came from. And it's really fitting in helping you remember what Hammerspace is, where that idea that you can go to one location and access essentially infinite amounts of data. So how do we do that?
[00:02:59] It's essentially the idea that you, from a single mount point, can see all of your organization's data that you have access to, that you've been granted access to, whether it sits in an on-prem NAS filer, an HPC storage system, a cloud object store, whatever it might look like. And then we make it that you can automate moving that data. So whether you're moving data to save money, you want to use compute that's in a lower power region, or a less expensive storage tier,
[00:03:29] whatever it is that you're looking to do with your data, you not only have that unified view, but you have the ability to automate the movement of that data without disrupting access to it as it moves into new storage locations. So we are a data platform, and I do that entire prelude because lots of data platforms exist out there. They all do very different things, but our goal really is to logically unify your data and automate the placement of it where you want to store and use it.
[00:04:00] Excellent. Thanks for that. David, you've had a question. Yeah. First, thanks for being here, both Molly and Floyd. Floyd, I'll throw this one out to you. So the data management and storage market is really going through a lot of changes right now, and that's likely to mean some upheaval for customers. How unique is Hammerspace in this space? Can you describe for the audience really how you differentiate yourselves and what you do that others don't? Sure, and that's an excellent question
[00:04:29] because where you're seeing the rub that occurs, particularly in the enterprise, is the fact that the introduction of AI into the mix of what was traditional application pipelines within the enterprise. And what AI is doing is exacerbating the problem that Molly just described that we are the solution for, and that is if you've got your data distributed
[00:04:57] across multiple different storage types and locations, and if the compute, especially if you need to now use high-performance computing for AI-type pipelines, if that compute maybe isn't in the same place, now you've got your data scattered everywhere. And a typical infrastructure-centric view of the world is, from an IT standpoint, is, you know, well, then we have to move the data to somewhere else.
[00:05:26] That doesn't work in AI, you know? And this is where if you elevate the solution out of the infrastructure and get it to, and again, it's back to the Hammerspace analogy, get it back to the metadata layer where everybody is interacting with the metadata, which is the file system, which is the view of your files, and the actual placement can then be opportunistic. Because don't forget, these application environments in the enterprise,
[00:05:55] they already have a relationship between a front-end set of applications and use cases and the storage that that data goes onto. It's typically hierarchical. Now you're introducing another application stack that not only needs to see that hierarchy, but also distributed data that might be anywhere. And so what we're seeing now is a forcing function in the industry where just a data catalog or just a data orchestrator
[00:06:24] or just a tiering solution alone doesn't do it if it doesn't have that global view and management over the data, regardless of which storage it is in today or needs to be in tomorrow. Thank you. Neil, you wanted to come in there. Yeah, I'd love to bring Molly back in here, especially with what Floyd has just said there. What growth has Hammerspace seen in this market? And has AI-driven data demand and the resulting specific technology demand?
[00:06:55] Has that even surprised you? And when I say that, I mean in terms of GPU, traditional storage, and cloud capacity demand. Because from the outside looking in, it feels like quite an exciting time. It is. And AI certainly has created, I'd say, a forcing function that led to much faster adoption of Hammerspace than we were seeing pre-AI. Now, granted, we were a very new company at that time too. But in any case, I think what you can really see is that three or four years ago,
[00:07:25] doing hybrid cloud in a more automated fashion was kind of nice to have in enterprise IT. The fact that your S3 object store in Amazon and AWS maybe wasn't real tightly integrated with a couple of different storage technologies in your data centers, it was a bit annoying, but IT teams largely were able to put up with some manual copying, some third-party data management tools. That has really changed with AI because we now all of a sudden
[00:07:54] have a many-to-one relationship of applications and users of the data. So the business unit who created the data is not the only ones using it now. Now there's AI researchers, there's models that they're trying to train, agents they're trying to train. And so having a single unified copy of data or kind of your bold data set that can be accessed and maybe even modified by different applications requires doing things differently. So what we have really found
[00:08:22] is this forcing function of data needs to move more liberally outside of the storage system where it was created. And it needs to be used in ways maybe not originally envisioned when it was created has really forced that Hammerspace is a much more urgent item when procurement comes around. It used to be kind of a nice to have if there's a bit of budget left, let's go ahead and try that. Now it's more that we need to have our data estate organized, mobile. It needs to be governed. It needs to meet
[00:08:52] our security requirements. And we've become an urgent need in this 410 modern AI era. Just building on that slightly is that part of the problem is that the data in those application environments has a day job, right? Where now it's got a function that those businesses are operating on. Well, now we're bringing in an orthogonal view of that data and needing to access it
[00:09:22] and needing to use it for something else, but you can't disrupt the existing environment. And so how do you do that? How do you do that on live data? Especially if your only option is to copy it, which raises all kinds of other cost, complexity, compliance, governance problems. And so it's really this ability to plug into your existing data estate without disrupting your existing applications and extend that into an AI context
[00:09:51] is really the key differentiation. And back to your original question, that's the one thing that Hammerspace does that nobody else really can do as well. And I know that you have other questions, but I want to build up one more thing you can tell we're passionate about what we do. The other piece that we didn't explicitly say and it's probably obvious, but worth calling out is not only was the data designed with a specific intent, you know, for finance or research, whatever it is, but that compute system that they designed
[00:10:20] to be used with wasn't designed for AI. So what we're seeing is now you need GPUs and, you know, a lot more memory and there's a different kind of workload in AI that the data needs to be orchestrated along with the workload to the GPUs that folks want to use. So to take advantage of NeoClouds, to take advantage of GPUs that aren't local to the data storage system, data orchestration has become a key technology to enabling that. And I just want to touch on that too, that we're not just
[00:10:50] about storing the data and keeping it secure and accessible by the right people at the right cost, but about putting it in the location where we want to use it for things like training agents. Yeah, I mean, what do you think about that, Neil? Because obviously, Hammerspace, you know, sometimes gets put into the storage company basket. And a lot of these companies, I suppose, are they doing it, they've always done other things, but a lot of companies like Hammerspace just don't want to be simply known as a storage company, do they, Neil?
[00:11:19] A hundred percent. And I also think you experience a lot of growth at the moment. And I think it'd be really useful for people watching and listening to understand what that growth that you're seeing here actually looks like in practice. Is it the particular types of customers? Is it the workloads? What's driving the strongest demand that you're seeing right now? Yeah, I'll touch on then if Floyd wants to add to it, by all means do. But the top line revenue growth over the last 12 months came in large
[00:11:48] part towards the early part, like early 25, 26, was in building foundation models. So we had our kind of run rate enterprise business, which was healthy and growing at kind of a more normal rate. But this hockey stick growth was from the model builders. And what we were looking at was you needed to have incredibly fast data storage systems next to compute and you needed to be able to use that storage efficiently. You know, you didn't
[00:12:17] want over provision memory, CPU and things like that because there was just so much of it. So we got our slice of that pie. But now that we're moving into the world where organizations are trying to run their enterprise on AI, whether they're doing inference or trying to access many data sets to train agents, whatever it is they're trying to do, it's a very different market. So now we're seeing more volume of customers instead of a couple massive ones. It's a much larger volume of ones that are looking to use data as being
[00:12:46] created each day, analyze what would be useful in their AI workloads, automated, get some early results where they're actually using AI in production. They're not stuck with kind of a siloed pilot, but they're actually using it in production. And that's where we're seeing that growth. And it really is just, you know, enterprise AI. And some of the companies are banks, some of them are research institutes, some of them do video production, but that's the kind of piece that runs through all of them.
[00:13:16] Floyd, in many ways we're talking about both data scalability and data security here. As we speak, I understand that Hammerspace has made a major technology announcement at this year's supercomputing show in Chicago. What a key show that is in the present climate of what you do. Could you tell us about that, please? Sure. And what we are doing and what we announced at supercomputing expands on what Molly just said. And that is, you know,
[00:13:46] Hammerspace's core capabilities as a data platform solved a major problem in the industry. The problem is also, though, when people are consuming and they're looking to expand into AI in particular, they actually don't want to have to integrate. They want to buy it as a system. When they access their compute, they're accessing systems. And so what we've done is actually expanded, in a sense,
[00:14:16] Hammerspace's core capabilities taking its foundation of the data platform software that we had before, but instead of just being a software defined platform that could run on any hardware with any storage, we also provided a much simpler way to deploy this. So which means as a complete system, end-to-end with hardware, either hardware that we've created, and we can go into some of that, or integrated hardware with
[00:14:45] partners, or if the customer still wants to build it in their own environment on VMs or whatever infrastructure, we will help them with that. Because again, what we're trying to do is reduce the friction for people to be able to actually take advantage of this unifying capability, to unify and orchestrate their data, especially as they're expanding into new AI-type pipelines. But to complete that
[00:15:14] picture then, in addition to just bundling everything into pre-configured hardware, we're also adding key capabilities, such as data intelligence. And this is the ability to get richer metadata out of your existing data estate, to be able to search for files, to be able to operationalize that in a way that a data scientist would expect, but a storage administrator doesn't really care about. A data scientist
[00:15:43] needs a plain language view, and that's what we've introduced here. In addition to that, we also announced the Hammerspace OFP storage platform, which is a system extreme high-dense DPU-based storage. I think you have slides on that. And what that does is it also fills a gap where for tier one storage or as an extended context storage, a very high-performance, extremely dense form
[00:16:13] factor that basically can disappear into the top of the rack. So to wrap that up, essentially that's what we're doing, is we're wrapping Hammerspace into a full system that can be still configured to your individual use case, still configured based on the type of hardware preferences you have and your performance requirements, but can be done so as a complete system. The key difference here is that you don't have
[00:16:42] to create a net new parallel environment. This system stitches together with your existing storage, your existing data, not only stitches together with it, but enables you to bind it together yourself across multiple sites, locations, and clouds. So it's really, you know, we're very excited about it because it makes it much easier for our customers, especially enterprise customers who you just simply
[00:17:12] want to plug it in and see it work. And I think it has a very sensible name as well, the Hammerspace data system. That's right. Yeah, we thought a long time ago. Did you want to come back there, Molly? Yeah, so I think that what I would just add to that is, you said earlier that Hammerspace is often kind of grouped in the storage category, category. And it's interesting, what we've seen over the last couple of
[00:17:40] years is because we really offer a data layer, a data, you know, we call it a data platform, now we call it a data system. But we've, we really came at this problem on really good data intelligence, having the ability to, you know, we use a global namespace, but we can assimilate or bring data sets into the Hammerspace namespace logically without moving it off the storage system. There's a lot of, you know, techie stuff, which most of our audience is probably techie understands all
[00:18:10] that well. But my point is more that we actually would not sell the capacity, the storage itself. We've always leaned heavily on use what you already have in place. Maybe you have 20 petabytes of capacity in your data center today and you're not using four of them. No problem. Hammerspace can use that, whether it's sitting on Vast or NetApp or Cloudian, whatever it might be. And as customers, customers started to deploy AI, it became clear that just like what happened on the
[00:18:40] AI factory or the compute side of putting together your AI compute environment turned out to be quite complex. It was a lot of different tooling, a lot of different procurement projects, and it was difficult to get everything configured, optimized, purchased, and up and running. That's where everyone in our industry was focusing by and large for the last few years. Now they're going, oh my gosh, my AI factory, my compute environment is data starved. And they have data, it's just in a lot of different places, and
[00:19:09] they don't know what the data is and is something that would be useful for their job to run, they don't know. And Hammerspace solves that really elegantly with our data system. But there are cases where a customer also is buying more capacity, and they want to do it all at once, and they don't want to have to buy data management and data orchestration and data storage and data metadata, all those different things, and configure it just like they didn't with their compute environment. So kind of summarizing what we're doing is we're doing for the
[00:19:38] data what NVIDIA and others are doing on the AI factory side by making it into one unified, integrated, easy to deploy, quote, optimized system. David, you're itching to come in here, aren't you? Yeah, just to kind of tack on to that, you know, Molly, obviously the supercomputing announcement that you all are making is major news. You mentioned your growth earlier in a previous question, but as a fast-growing company that's in the
[00:20:08] data management and storage space, how important is this latest announcement that's coming from the supercomputing, you know, as a growth driver for Hammer space? We think it's critical. There's a lot of things going in our market right now with shortages on SSDs, shortages on GPUs, shortages on power, like all of that we all know very well and are dealing with in our day-to-day lives. And in the midst of this, we're also
[00:20:37] trying to put AI to work to solve research problems and optimize how enterprises are run. And making this just more predictable, making it faster, making it easier to do AI brings those two together. Use your infrastructure efficiently, be able to quote it fast, get it deployed quickly, and get up and running. So we have a massive amount of backlog waiting for these systems to be really out in the market and ready to quote through our partner network in
[00:21:07] conjunction with our alliance partners like Supermicro, Cisco, folks like that. So even the big systems companies that do commodity type of off-the-shelf storage are anxious for this because it designed them in to the data system as well. And think about it, that whichever company that might pop to your mind as being an AI storage company, you know, there's a handful of them are really well known. They came at from first, how do I manage
[00:21:36] my capacity, do it efficiently, or do it really fast, have very high performance storage. And now they're building their data services in. HammerSpace has been in the data service world for years. That's what we're designed to do. But we didn't package ourselves as an appliance or as a system. And I think we're all converging on the same thing, that you need the AI factory for data. I think the industry has spoken, customers have spoken, that's what we need to do. And we're very early in delivering that so
[00:22:06] we can be a leader in this space for AI data management, AI storage, because we've already done the really hard part of building all these complex data services. We're just packaging it a bit more elegantly now and very much in response to customer demand. So we expect we're on a massive growth curve already, you know, hockey stick type of growth. And we suspect that this will help us continue to do that without having to, you know, scale the cost of our company substantially. It's a faster way for us to deliver
[00:22:36] HammerSpace. One thing to never forget is that AI is a moving target. We see that every day, you know. And for an enterprise, when they're making a decision about, well, what are we actually going to use AI for? This isn't for their people to get on to Copilot or ChatGPT. This is, you know, operational processes, which could be in a manufacturing pipeline, customer service desk, whatever it is. The problem is, if you're
[00:23:05] having to build a bespoke environment that is very rigid, that also needs to access that live data, and you don't necessarily know what all those use cases are going to be, you're going to learn. And so one of the advantages of our architecture is this flexibility, the ability to begin with a pilot project, but then graduate that into full-on production, whether the pilot might be in one data center, but production may need a
[00:23:34] cloud or another cloud, or may need to pull in other data. And then guess what? The minute you get into these pipelines, you start realizing, oh, we could also do this. But if your system is not flexible enough to adapt to any compute, any data on any storage, then you're going to be ossified into a fixed use case now. And that's one of the key things here, is that simplicity and
[00:24:03] the agility of a data system like Hammerspace enables people to course correct, to change which hardware they prefer, which compute they prefer, and do this all real time without disrupting their existing workflows. Excellent. Floyd, I think Neil had a question for you. Yeah, another question I'd love to ask you, Floyd, here, is I think that many organizations are quite rightly concerned about the cost of AI right now, especially some of those expensive GPUs
[00:24:32] that are sitting idle while waiting for data. So how can a unified data system help improve maybe the economics of AI infrastructure, address that ROI question that every CFO must be asking right now? And also beyond that, as AI agents, inference workloads, and sovereign AI environments all continue to evolve, how do you see the requirements for enterprise data infrastructure changing over the next few years? I appreciate I've cheekily snuck two questions in there, but what are you seeing?
[00:25:03] Sure. Molly talked about this a little bit before as well, and that is, you know, being able to, you know, sort of adapt to those changes, to be able to, you know, take on any new use case, to be able to be fragile, be agile, not fragile, but agile with this. And part of it is the economics. And when people talk about token economics, they often look, talk about time to first token or time to token. But what they aren't looking at necessarily
[00:25:32] is the entire data cycle that is required. If your ability to stage that data requires you to copy it to a new environment, then already you're not dealing with live data. So there's a lag there. And if your ability is not, if you don't have that global view and global orchestration with access, then guess what? Your GPUs are going to be sitting there waiting for their data. And so we take a more
[00:26:02] holistic approach. We call it the three clocks, you know, so the time to very first token. In other words, what data staging needs to occur across all of the distributed data sets, curating down just the part that you need for an AI pipeline. And then based on that AI pipeline, whether you've got an agentic workflow going, whether it's, you know, a rag pipeline, whichever sort of flavor that you happen to be going, what course correction can you make without now forcing
[00:26:32] again your GPUs to wait for you to gather all the data? So that's why the data system and the AI factory, the holistic view of both sides, while allowing that existing data to do its day job within existing applications, if you don't have all of those pieces together, you're inevitably going to have inefficiencies in your utilization of hardware. You're going to add costs through duplicating infrastructure, especially if
[00:27:01] you're duplicating an all flash array and to do something that maybe only a portion of that data really needs the performance, then you're, you know, it's like cost upon cost upon cost. And so that executive who's trying to say, well, what is the ROI really for this AI project? Guess what? If you're burdening it with costs before you even get to the first token, then that argument becomes a lot harder to sustain.
[00:27:30] And so our view is a data centric approach that assumes that all data is available anywhere across any storage at any time and that you can orchestrate that where it needs to be and add the data intelligence tools, add key, whatever key pipelines you need without disrupting your existing data estate. That's kind of the winning combination and that's what we feel we have. Does that do it for you, Neil? It really does. I think the only thing,
[00:28:00] Michelle, I'd love to try and bring to life what we're talking about here. You don't have to mention any names, but maybe with a practical example of where better data access has actually helped a customer make better use of its GPUs or just reduce the time that they spend waiting for data. Are there any examples you could share there? We'll jump in one. I know Floyd wants to too. I mean, we can give, you know, very practical examples. But one of the ones I thought was really kind of fun is on a video platform, which we
[00:28:29] are currently recording on, we can't use their name publicly, but was looking to do AI in the cloud. They partner with Oracle and what they're using the cloud for, sure, they are, they're generating data all around the world. You know, let's say edge locations as videos are being, video conferencing is occurring. But when the data is identified that they want to use it in their AI, it first needs to be identified, then it needs to be orchestrated up to Oracle Cloud,
[00:28:59] whichever region they're running in. And then they need to use the GPUs that they've, the GPU shapes that they've identified that they're going to use in Oracle. And in that environment, it touches on what we call them Hammerspace superpowers in a variety of ways. We can assimilate the data. So we look at the metadata as it's being generated around the world real time. Once data that's identified that they want to use in the AI project is found, it's automatically orchestrated to the appropriate GPUs,
[00:29:28] wherever that region is. Once it lands there, we place it locally on the NVMe in Oracle that's right next to the GPUs. We call that tier zero. Why that's important is, one, the customer's already paid for them. When they rented their GPUs, they're also renting the NVMe that they're sending in those GPU boxes. They don't have to buy additional storage. So it saves them money. And it's sitting as close as you can get to, other than being in memory, to the GPUs. So you use them very efficiently. There's no network latency and
[00:29:58] anything like that. And then when they're finished with the job, they can delete it, store it in object storage, whatever they're going to do. But I think it's a really practical example of, they saved, I'll get the numbers wrong, but it was around 10x on their Oracle bill simply because they were using it more efficiently and they didn't have to buy storage on top of the storage they were renting with their GPUs. But a massive amount of workflow efficiency as you can imagine for their AI workloads. That's a big company,
[00:30:26] massive data set example where you're talking about dozens of petabytes, lots of active real world data sets. But then when you look at some of the, maybe not as broad scale environments, but the types of environments, we just announced the Keiko case study, which is a scientific research environment in Germany, where we announced that at the show this week as well, where they're doing it on a real-time research scale. They're constantly looking to do research. They have new data coming in off
[00:30:55] of different researchers, different instruments, and they're constantly trying to keep their AI projects updated with the latest information from there as well. Even though it's not done at a global scale, it's still across multiple systems, storage environments, and that have to get to GPUs. So it's very practical that unifying your data, being able to use the storage you already own, and then add capacity as appropriate. It all makes a lot of sense, but there's a lot that had to be built to the technology to make this viable.
[00:31:25] And one thing we didn't touch on yet that I think is so important is what makes a big portion of this all possible is that we're standards-based. So our technology, our software runs on Linux. If you need proprietary client software to speed the GPUs, which you consider parallel file system software, we've built that into the Linux kernel. And why that matters is, as AI moves forward, and as Floyd said, AI is very dynamic. We don't really know what it's going to look like a year or two or three years from now.
[00:31:54] The drive from NVIDIA, AMD, and others towards a standards-based architecture so that AI can change, it can flex, and you're not making an investment today that you have to throw away tomorrow, and that investment risk that every company leader is worried about. The fact that we've built this all on standard Linux into standard Linux helps customers feel much more comfortable that it will move with them to wherever their AI goes next. Awesome. Excellent. Pushing that into the standard kernel, I mean, something that
[00:32:24] maybe, that Molly mentioned that you may not have caught is when we're orchestrating that data, we're not spinning copies everywhere. We're using something that we introduced into the Linux kernel called Live Data Mobility that enables, and we also stitch it into our own. We activate all of these things that we pushed up into the kernel that makes it easy for the customer to get this level of performance, but we've also built it into our superpowers
[00:32:53] within Hammerspace, which means that I could have users working on data, live data, and we could be orchestrating instantiations. It's the same file. It's the same history. You're not forking the data, but instantiations that need to go into a compute environment somewhere else. And so we've got complete control over that, giving the customer complete control to be able to use live data without interrupting their existing workflows.
[00:33:22] That's just one example of the thousands of features that we've contributed into Linux that we also take advantage of to put this whole picture together. Awesome. Thank you. And I'm sure at this week's Supercomputing Show in Chicago, your representatives there are keenly answering questions about what bottom line savings potential customers can generate from this sort of technology. You're very exciting. Coming back to me in terms of questions,
[00:33:52] for Molly, in terms of technology areas generally, obviously what you're announcing in Chicago will branch out in certain areas. But generally, what other areas are you looking to address over the next maybe 12 to 18 months? Where is Hammerspace going at the moment? What's the direction of travel? Can you give us an idea of some of the broad areas you're looking to improve for end users? Yeah, absolutely. So this week's announcement really touched on expanding our capabilities and to buy it
[00:34:22] as the system, have increased amount of data intelligence, understanding of what your data is, enriched metadata. But we also introduced, I don't think Floyd mentioned it by name specifically, but KVCache solution, that kind of ties into the data intelligence side as well. So we'll continue to see those types of advancements coming into our platform. But if you look 12, 18 months from now, we do believe workflow integrations are critical. Data orchestration should tie with compute
[00:34:50] orchestration, metadata enrichment, and knowing what your data are, that if you're in healthcare, you have different requirements of what data you track and how you use it versus maybe a sovereign cloud in Korea. You know, very different. But the technology is, there's a lot of similarity of how to do it, but the why and what your rules are, we'll be continuing to tie our technology into kind of the governance side of the environment as well.
[00:35:18] And we do see there's possibility for continued innovation on the hardware side. We introduced OFP, our OpenFlash platform, as an additional option for storage this week. Right now, we're focusing it on performance, on context storage. But there's some really cool stuff. And if you check out the OpenFlash platform initiatives website, I think we give a tip of a hat to where we're going, but I'll talk about it here since you asked. You know, we have the idea that every top you can take out
[00:35:48] the data path makes data faster, but it also saves power and it saves money. And the more we can move the data, the Linux storage server itself closer to the SSD or even inside the SSD is where we see it going. You can imagine the efficiencies you can get because now you have less network ports to buy and power, like less storage servers to buy and power all the way through. So we're working hard on compressing and hardware as much
[00:36:17] innovation into the SSD as possible. And the SSD vendors are quite keen on doing that. So there's a lot we're looking to do on, you know, one side is data intelligence. The other side is infrastructure and storage efficiency. And we have some pretty cool stuff we'll be talking about in the coming year about that as well. Excellent. So it's a case of what's this spaced in? So we all be. That's what I'm trying to do. So David, he has a question for both of you. Yeah, whichever one wants to jump in this.
[00:36:46] I'm trying to get, you know, my head around. So data orchestration, data management, and now data storage, they're all right at the center of the AI market. Out of curiosity, when potential customers think about solving these problems, where do you think Hammerspace sits in their minds? And is there more you need to do to be at the top of their list? I'll take a first start. I mean, yes, there are some more things to do. So brand awareness is part of it.
[00:37:16] And, you know, you think about, you know, you all are immersed in the space of what kind of analyst reports are out there and market sizing. You know, being in the Gartner magic quadrant for AI storage, that's super helpful. And the analog of that with other analyst agencies, certainly those types of things are important. And just to be on the short list, training the answer engines where people are doing the majority of their research these days up in Gemini or ChatGPT and making sure that we're recommended properly.
[00:37:46] There's a lot of just brand and marketing work that Hammerspace is doing, but nothing speaks louder than customers who were early on the enterprise-wide adoption curve talking to their peers about what they're doing. So we have some early, really great advocates with meta and folks like that. But sometimes enterprises were like, that's cool that meta used you to train llama, but I'm not meta and I'm not building llama. And so getting customer stories out there on how we helped them, how much we saved
[00:38:15] them is really critical. So this week at the supercomputing conference, we had a user group meeting where, gosh, I think six or seven of our customers that are more of the general AI space, we're talking about their results, why they chose Hammerspace, how they deployed us. And that was everything from the quantitative trading side to universities to AMD-based NeoClouds, different types of customers. And really what we're looking to do there is have the folks who have already deployed
[00:38:42] us and maybe chosen us instead of an alternative, speaking to why and how we helped them across the hybrid cloud and all the pieces that we're talking about in today's conversation. But I think those are the big things. We have the customers. We have the metrics. We just need to get the result, the noise out there. Yeah. Marketing, you're there. It's a fascinating area. It becomes more and more important in technology.
[00:39:10] For some people, it's very complicated. It needs a good marketing message. And I'm sure you're the two people we need to talk to about that as far as Hammerspace. So we're on the money there. Floyd, did you want to add to that? Yeah. Yeah. I think that part of the issue is that people for 40 years, 40 plus years, when they have a data problem, their kind of knee-jerk reaction is to go to the storage. It's like, well, what storage do I need for this?
[00:39:39] Or I've got to figure out how to bridge these. So what point solution am I going to help me stitch all this together? And so moving from an infrastructure-centric view, that's what we call it, to a data-centric view where you think about the outcomes and the storage, in a sense, becomes a bit commoditized. It doesn't mean that you don't have specific.
[00:40:05] I mean, we just introduced OFP with the very specific use case. But we also know that that's not the only storage somebody's going to need. There is no one-size-fits-all. And so, but the industry, which, you know, storage industry, they make money by selling storage. And so they're going to want to keep you in their storage. Our view is, and this is where the shift occurs, is shift is occurring with the analysts.
[00:40:32] The shift is occurring with you guys, with the media as well, as to even know to ask the question differently. In other words, which, not which storage should I buy, but how do I actually manage that data across where it is today without having to buy all the new storage? And if there's an SSD crisis and I can't even get it, so wouldn't that make even more sense?
[00:40:57] So it's a pattern shift that we have been driving at with the analysts, with the industry, with the consumer, with the customers, and that will continue to go. Thank you. And our hope is people will get what a data system is, that just like the term AI factory is starting to become a known term, something that NVIDIA, who has a lot more money than Hammerspace, spent a lot of time developing.
[00:41:24] But we're trying to kind of ride those coattails of, if you get why you needed an AI factory, that it was really complicated putting it together on your own. You need something like that for your data too. We're trying to draw some simple correlations, and that's the data system. And having people kind of know what to go look for, to help users just get to these answers quicker, instead of trying to come up with their own solution based on how they built things 10 years ago for different reasons. Thank you.
[00:41:51] Neil's going to round out this key conversation with a final question here. Neil. I'd love to try and give everyone watching and listening a valuable takeaway, maybe a bit of homework, maybe something to think about over the holidays. So is there anything else you'd like to leave our audience with regarding how organizations can better prepare for AI-driven corporate data changes that are on the horizon, especially as with that magical time of year where we're planning for next year? Maybe give everyone... I thought you were talking about Christmas.
[00:42:22] Maybe give everyone listening some key areas to address. Anything spring to mind? I mean, what jumps to my mind is before you go down the path of doing what you've always done, this really ties to what Floyd was saying of, okay, I have a project. I'm going to go buy an application, some storage, a network, and make this bespoke environment. Stop and look at what do you have today? And how could you put it to work and operationalize it?
[00:42:48] And that's the idea of unifying your data logically without physically having to buy new storage systems, things like that. What would you do with your compute if you could easily move your data to any compute? Maybe it's in London today, it's in Vancouver tomorrow. How would you maybe do things differently? Stop and kind of sketch that out. And I think that for folks to think about that, it just takes a minute to not go down the path of I'm very busy, I'm going to do things the way I've always done them.
[00:43:16] But think with a kind of white sheet, a blank sheet of paper. And the solutions are out there for a faster, more efficient way. You just have to take time to embrace them. Floyd? Floyd? Yeah, I would just echo that. And I think enterprises of all kinds, organizations of all kinds, should ask of their vendors, how do you bridge these silos? Do you require me to migrate my data?
[00:43:46] Do you require me to set up a parallel infrastructure? Because that will ferret out a lot of answers as to whether this is a data-centric approach that is actually just going to provide more vendor lock, as opposed to, I mean, a storage-centric approach that is going to provide more vendor lock versus this ability to step above the infrastructure and look across the data. And so data-centric is not just a marketing turn.
[00:44:13] It is a philosophical shift of the mindset. And people have to be educated and think, you know, I got to ask my vendors these questions. Can they do this? You know, and how do they play with what I already have? And how do they do this without disrupting my existing application workflows? And so I think, you know, we would advise our, you know, the industry to ask tougher questions of their vendors. And, you know, we're ready to answer those. Thanks for that, Molly and Floyd.
[00:44:43] We hope we've generated plenty of interest among our audience around the key topics around data orchestration, data management, and cloud data storage. It just remains to say goodbye. Thanks for being here, Molly and Floyd. And it's goodbye from me until our next Across the Tech Pond podcast. Very, very soon. Molly? Yeah. Thank you for joining. And Anthony, Dave, Neil, you guys are great hosts. It was fun to get you all in the same conversation. Definitely.
[00:45:12] Yeah, definitely. Great questions. And I enjoyed the conversation. It was good. Neil? I'd just say to everyone watching or listening, remember, we are on Apple Podcasts, Spotify, all major podcast platforms, as well as YouTube. And we are also out and about on tech events around the world. So if you see any of us, please let us know what you thought of the show. David? Yeah. Thanks so much, Molly and Floyd. Great insights.
[00:45:39] Congratulations on your big announcement at the supercomputing event. And if you're a technology company with something important to say, this is the place to say it. Across the Tech Pond puts you in front of an audience of business and technology decision makers on both sides of the Atlantic, just as it has today for Hammerspace. So reach out to any one of the three of us and let's talk about your story. Thank you. Thank you. Thank you. Thank you.

