36:23 Webinar
The End of the 5-Year Forecast: Navigating the Storage Crunch in the AI Era
Fred Lherault and Savas Nicolaides explore a critical shift from a “Buy and Build” mindset to an “Outcome and Service” model. Learn how adopting the Evergreen architecture can eliminate the risk of making the wrong hardware bet, prioritizing business agility and operational resilience over simple asset ownership.
This webinar first aired on 07 April 2026
The first 5 minute(s) of our recorded Webinars are open; however, if you are enjoying them, we’ll ask for a little information to finish watching.
Click to View Transcript
Hello, Register audience. We are live. Live from London, I suppose we could say, couldn't we? You're very welcome. We've got a great session today because we are talking about the end of the five-year forecast.
What does that mean? Well, a decade ago, we weren't talking about ChatGPT or gen AI, and for that and other reasons, then I- IT teams, they're pretty comfortable forecasting what infrastructure demands would be three or five years out. Well, something's changed, hasn't it?
We can't do that anymore. Let's find out what has changed and what we might do about it instead. So who is going to be helping me to find out about that? Well, we have some good friends of The Register today from Everpure. Savas, welcome back. Hello.
Hi, Tim. Great to be here. Now, Savas, remind me, what do you do at Everpure? I, Hi, everyone. I am lead principal technologist for EMEA at Everpure. I work with our product teams and customers across the region, and my background's always
really been working with large orgs on their data and infrastructure projects. Before we go any further, STaaS, Everpure, that wasn't your name a couple of months ago. No. We were formerly known as Pure Storage, and we have made a rebranding, to really one, our commitment to Evergreen, and we'll talk a bit about that- today, I guess.
And the other, the fact that we are not just storage, we also, work with data and customers' data projects. Great. So also from the company formerly known as Pure Storage, not known as Pure Storage anymore, Fred, hello. Hey, Tim. Good morning. And Fred,
what do you do at Everpure? So I'm, what we call field CTO. I basically represent the office of the CTO and the engineering organizations with our, largest and most strategic customers. Great. Now, let's get straight on into this. I did say at the beginning that a decade ago you could model demand a
few years out, and that you can't do that anymore. Well, first of all, we've gotta find out what has broken that model. What has broken that model, Savas? I think, part of it was is the expectation that users have within organizations. So if we think back to when public cloud was introduced, it created a different paradigm,
which is then fascinating because it then imposes different expectations on internal IT as well. So there is an expectation of instant availability of services, from users that consume IT, and then that creates a challenge of predictability for, IT organizations. A- and, this is something, Fred, that AI in particular has made very important and very
difficult to forecast, isn't it? Yeah, it has. I mean, so first of all, no one really how the world of AI will be evolving, over the next six months, one year, and so on. And for a lot of the, a l- a lot of the organizations we talk to, they've got an AI project.
They know, roughly what they're going to be, deploying in terms of AI compute, GPUs, but they don't really what they're going to need in terms of data storage and data storage access, reqs. So it's, it's become really difficult for, all of those organizations to say for sure, "This is what I need to invest for the next five years," or even, even one or two years has
become really difficult to predict. If you think about it, the world of AI itself has been evolving way faster than anything else we've seen, at least in the past 25 years. But, Fred, there is an aspect that's always been like this. It's never been possible to be 100% sure of what you're gonna need to provision.
Why are we particularly worried about the needs of AI projects? Yeah, because if you get it wrong, it's, it can be very costly. And moreover, because again of that change of, in the way, i- in the way people do AI, right? So just the, not just the hardware, but also the software, the, the, the, the methodologies that exist around it.
You may actually end up with a requirement that is 10 times more or 10 times less than what you were planning. So it's the It can be a re- it can have a really, really big impact on everything else that you're doing and competitiveness. So, Savas, when you go to the When the IT team, the business goes to the IT team and says, "We've got an I- we've got an AI project, and we'd like you to provision it for
us," and they can't say yes, is that the problem that we've got here? What's the consequence here? What happens? Yeah. I guess the, the problem is the time to market of purchasing traditionally, right?
You could say, "Well, get to the public cloud." That'd be a quick way to start. But then the challenge with is not all use cases and workloads are applicable to public cloud. You might have sensitive data. You might, you know, when we look at AI, it is working on the organization's data and needs
to be in close proximity. So-You know, public cloud that would've been a, an answer doesn't necessarily work. So then, yeah, time to market. Maybe Yeah. Or, or maybe not at very large scale. Yeah, at large scale. Yeah. Yeah.
Public cloud is great to, start an AI project, but the minute you start scaling it, the, the cost starts scaling, at the same rate, so it can very quickly become expensive. Fred, is there a problem as well that the business ends up routing around the IT department if it perceives the IT department as a blocker? Well, of course. I mean, shadow IT has been a thing for, that,
that we've seen, organizations, doing for, a long time, and now you've got shadow AI. Fair enough. And, and I think it all stems back to what Savas was talking about. It's the expectation. In the world of the public cloud, you want a resource. Well, first of all, you, you don't have to try to explain to the, that IT team, tell them
exactly what you need. You get, offered a, a, a catalog of services. You go and pick one, and then it's available pretty much instantly. If you contrast that with the way traditionally people have been doing their, forecasting at, at the end of every year, you've got the IT team that goes and talk to
every business unit and say, "How much do you think you'll need in the next year or so?" and then they try to concoct a, their, their, their budget for the, for, for the coming year. The So for the, you know, the people that are, used to that cloud mentality and that cloud, level of agility, it's, it's a completely different way of, of, of addressing things.
So if they That's the reason why they, people were, doing shadow IT in the cloud, and that's the reason why they're looking at, they're doing shadow AI today. Yeah, STaaS, when you think about the cost of getting this wrong, we, we can think about the cost in terms, the financial cost in getting this wrong, but there's an There is a reputational cost.
There is a cost to, the credibility of the IT department as well, isn't there? Yeah. And, and I think it comes down to risk as well. So there's the risk of getting it wrong, is risk for your, you know, having the wrong equipment or having too little or too much, and then there's the risk to the business of the delay in starting their projects.
So yeah, we, we like to think about, as like, how do you de-risk, the data in the same way public cloud de-risks new projects as they start up in the public cloud? And, and, and by the way, that because of that risk, what have people been doing? Well, they've been buying more than what they needed. Because if you, if you come to me and tell how much you're going to need, and I think
going to need, say, 10 TiB, I'm not going to tell you 10 TiB. I'm going to tell you 20, and then the, whoever does the budget for, at the IT level is going to say, "Well, okay, maybe pad that up a bit more, 25 or 30." And then you come at the end of, what's being deployed, and it tends to be two to than what is actually needed this year.
Yeah. The, I, I, I'm sure there's plenty of people watching this who had the, the, yeah, who, who's had this problem of just not knowing what they want, having people being at the number of trucks that are showing up, or just having a look and seeing what they've used and finding it's, in any direction, completely out of scale to what they thought it was gonna be a few months ago.
So let's think about a way that we can get this right. Now, you have a particular way in which this can, that people can get this right. It's your in- It's your business, and, it's, Evergreen//One, isn't it? Now, we've discussed this before on, our Reg Cast, and, it's, it's often described as a sort of an OPEX model for storage rather
than a CAPEX model for storage. Does that capture what Evergreen//One is? No, no, it doesn't. It's Well, not just. So, so the, It's really about, the consuming as a service on demand versus, purchasing and owning assets and, again,
having to guess how much you're going to need, in the next three, four, five years. Now, the, the, I guess the, the type of budget that you use to consume it actually will vary, depending on, on customers. But we, we have had customers, using CAPEX type of budget for, for Evergreen//One, depending on the, the, financial rules of, of a given country.
Just maybe I can add to that. So just to underline what Evergreen//One is, it's a STaaS offering from Everpure that is built on an Evergreen architecture, which means it's not something you can do with a, traditional like finance wrap or something built on a lease. It, it relies on having a storage infrastructure architecture which can be
non-disruptively upgraded and has no end of life. And therefore, we can offer that, really de-risking that to the, to the organization that's subscribing. A- And, and to be clear, this is something that we deploy in the data center, so our customers' data centers.
But, but the, their Evergreen//One subscription also allows them to consume our, Everpure Cloud, so previously known as Pure Storage Cloud offering, in AWS and Azure. But the, the, ju- just, again, to, just to clarify, what we're talking about is something that is actually also designed to fix those type of challenges in the data center with
assets that belong to Everpure that we are deploying in our customers' data centers, but they can consume on demand.So if I'm one of your customers and I've got your asset in my data center, and I'm consuming a certain amount of storage, and suddenly in this example that we've been given, that you've been given at the beginning, where you've got an AI project and the demand goes way up, what happens?
How do I manage to get the storage that I need? You just, use more storage and it's our responsibility to make sure that there's always enough buffer in the system to accommodate it. I've got a slide maybe you can, you could, you could show at that time- Yeah that shows the, some of the various SLAs, but that's what it is, right?
It's an SLA-based service, where, so y- literally you've got SLAs in terms of, how much capacity is available, what performance levels, you're going to get on a per TiB basis. The SLAs in terms of availability, even SLAs in terms of how much, power we're going to be consuming.
And it's, it becomes our responsibility as the provider of this service to make sure that whatever we deploy in the data center matches those SLAs. And it's actual SLAs, meaning that if we, if we don't deliver on those, there's financial penalties, that, that come with it. Um- Ah, right. The Yeah.
And yeah, Sever, as the demand sort of zips up, are you sure that you can keep to these SLAs? Yeah, so we have So once you deploy the service, we actually have customer service managers whose job it is to run the service. They will look at the performance, make sure we're meeting the SLAs. Okay. You know, it's on us to upgrade
needed to make sure you are deli- we are delivering the SLAs. And similarly, to have enough buffer in there so that you can, either grow your workloads or up your, your reserve amount if you want to. So a bit like the cloud, you can, you can predict your working rate and reserve that for a discount.
And then if you've got spiky workloads and you wanna burst into using some capacity, maybe there's a temporary project, you can use that, and then when you don't use it, you don't pay for it, right? So you're essentially paying for just what you consume, and Everpure takes on the risk. Yeah, this is You said something really important here, Sever.
It's you pay for what you consume, and the way we measure this is in terms of real written data. So unlike actually, the public cloud or a lot of, other offerings, where what they care about is how much you're provisioning, how much you, you think you're going to use, here we on- we only charge our customers for the real
and so i- in the example I was, using earlier, of the 10 TiB that became 25 or 30, you would only be paying for the 10 TiB, and actually only 10 TiB when you need it. It's very likely that you're going to start, using less than that and then grow into it. So we don't really care about how much capacity is being provisioned by customers,
the size of the data volumes or file systems that they create. We only care about how much real data they've written on, on, on those volumes systems and, and buckets. And, and, and I suppose one point of that it stops your customers having at notice to go scrabbling around trying to find extra capacity, which we all know is not
always available, certainly not at the prices they want to pay. Of course, and the, that's the So, so of all, one of the benefit when you about written data is it's a lot more predictable than provision capacity. If you look at provision capacity, you've got those steps, effect in there. Whereas real written data tend to, to, to grow in a l- in a more linear, type of fashion.
But one other thing you get, also with Evergreen//One is the STaaS, so we were saying the, the, there's a buffer of capacity available in there, free capacity, and it's our responsibility to go and replenish that buffer as needed. And, so, and, and, it's something that we're doing, with, with the, within a four to six weeks, depending on location.
So it's, it's also a really good way of ensuring that the, you actually get the capacity that you need when you need it, instead of having to plan a long time in advance, f- for capacity that you may only need, at the end of the year or next year. And interestingly, we do that within the constraints we've set out here. So energy's really important at the moment, probably getting more important
That energy SLA means we can't just put any old equipment in. It has to satisfy the right density of, you know, watts per terabyte, plus the buffer, so that, you know, you can predict your power requirements as you grow as well. And the other one that on this slide that I think's really fascinating is the paid power and rack-and-stack. So, so that you can compare this service
that's in your data center to the public cloud, we pay for the power and rack space, so that then it's, you know, you can make that comparison more easily. Right. Now, if I am correct, there is a particular offering for AI projects, AI applications. Am I correct, and how does that work?
Yeah, yeah, you're right, Tim. So the, indeed, we, So if you j- if we just take a step back. So the way, Evergreen//One works is we've got various levels of service classes that our customers can choose. So they'll say, for example, "I want a performance level, class of service." And the way we measure this is by saying, "You'll get this performance per terabyte."So by the way,
this is aligned with the WEKA, public cloud storage is being consumed. You know that if you deploy a specific amount of capacity, you'll get the specific amount of performance that goes with it, and as you increase capacity, you also increase performance. So that's the way most of our, Evergreen //One services as Are working, and we've got,
10 or 12 different classes depending on, on, on what people need, in terms of performance density, right? Performance per terabytes. Now, when we looked at the challenges that our customers had when came to When it comes to, to AI, we realized that they had a different type of challenge.
So they genuinely knew what they would need from a performance point of view, but not from a capacity point of view. The reason they know what they need from a performance point of view is they know how many GPUs, they've purchased, type of GPUs, and it's fairly easy to say, actually the GPU, vendors will tell you, "For this type of GPU, you need to be able to drive
this many MB/Sec times the number of GPUs," right? So the, the sizing there is fairly easy and, and if you wanna be certified, you have to be able to You have to demonstrate that you bring that level of performance. But the big unknown is the capacity. So, we, we've seen cases where, initially some organization think they are going to need
maybe a TiB of data per, GPU. But then as they evolve their use cases, they realize that they're going to need maybe 10 or 100 terabytes, of, of, of, of data per GPU. So we've built that Evergreen//One for AI offering, which is really designed to answer that. So we come with a guaranteed level of
performance, measured in GB/s. And again, based on, the, the type and number of GPUs that you have, we'll, tell you, "This is what you need to get there," and it'll be a certified, architecture. And then the capacity you use is 100% on demand. So if you only need a TiB per GPU, and that's what you're using, you only pay for
that TiB per GPU. But as you start using more, y- the, the capacity's already there, and you'll only pay for that capacity when it becomes used, because again, there's so many unknown in the world of, of AI. Does that make sense? Yeah.
It, it does make sense. Now, the thing that we are mentioning all the way through is the cloud. And many people will think, I know you have your companion AWS service for it, but many people will be evaluating this compared to just saying, "Let's not do this on-prem at all." why stay on-prem?
It- data sovereignty is big in the news at the moment. Is that big for you? Yeah, yeah- Go on, Silas. Yeah. So yeah, for a number of our customers, especially if they're regulated, data sovereignty is something they talk to us about, in the sense that they wanna
You want your Sometimes you want your data to not be in the cloud for that reason, so that you have control. One customer I went to see recently was, you know, part of some critical link infrastructure. They wanted complete autonomy in the sense that if something goes wrong or if there's a disaster, they can control all the services
that apply to their data. So, so that means having it on-prem. And I think for, for them, what's interesting is when you look at what the What you get with public cloud, and encourage people to think about what are the benefits you're looking for, i.e. Agility, flexibility, being able to deal with
the, you know, unpredictable demand, and then compare that to an offering like Evergreen//One. Can you get that? Sorry, Fred, you were gonna add something. Yeah, there, there's a cost aspect, but, the To double-click on what Silas was saying, the So it's likely that the most interesting use cases you'll build
around AI will actually require your most critical data. So the, the Really your crown jewels. So the, the ones that are actually the, the least likely to end up in the public cloud. I mean, certainly there's regulated industries where, it's just never, it's, never going to happen.
But just in general, the Again, those interesting use cases are going to require access to some of your more important data. So just for this reason, and for a lot of our customers, they are building those, well, those AI private clouds. And this is the perfect consumption model to go with any form of private cloud
because again, the I know we keep going back to the way people used to do things, and, and previously. Now, when you're building a private cloud, you're basically building a Yourself a service. You're defining some, offerings, some, defining your own SLAs that you want to offer to your own internal customers.
The It becomes really difficult to predict what the developers, the data scientists are going to want to do in even a year's time. So instead of, purchasing and then amortizing it, and only being able to really recoup the usage of these over time, something like Evergreen//One is the perfect, consumption model to back that and avoid investing
upfrontWhat do you do about resilience, STaaS? Because that's what a lot of people worry about on-prem. What happens if there's a disaster? What happens if there's a successful attack? Yeah. So we have, obviously there's the
SLA, and you can obviously have, protection across multisites or, in fact, replication across three sites. But if the worst happens, either a, a cyber attack or a disaster, we have an add-on to Evergreen//One which is, we call a cyber resilience and recovery add-on, where effectively, again, it's a, it's an SLA that if you need clean infrastructure to
recover to, we will ship that within 24 hours. So, there are some, you know, high-profile organizations that were hit with cyber attacks. Typically, you know, you could recover to the same infrastructure, but you might wanna quarantine that infrastructure and investigate, in which case, you know, the best place to recover to is to clean-room infrastructure, that's been shipped.
Um- But there could be cases where you don't have a choice. So we- Yeah we've, we've seen some of our customers in the past, where the, their existing IT infrastructure was declared, a crime scene, and therefore, they were not able to recover over there. So they, needed to really quickly, stand up a new, a new platform to, to recover their IT infrastructure from.
Let's be practical about what success might look like. How might this change the way in which the IT department functions, and also if their, their, an IT department wants to migrate to your service, how that happens? Yeah, so f- first of all, if I am in that IT department and I am using Evergreen//One, how does this change my day-to-day working life, what I can offer to the business?
What will I notice? Thank you. The first thing is to think about service levels. So in, so with Evergreen//One, the conversation we have very much more is about what are the outcomes you want to achieve? What are the service levels you want to offer to your users?
And then we effectively, you know, our systems engineers will build or design the service to match that with the appropriate SLAs across the sites that are needed, which is a very different and more productive conversation than turning that into a product, right? Or, "Here's the number of devices you need, with capacity, and here's the IOPS." You
know, it, it's, it's a much more straightforward conversation. And then allows the IT department to, to not be in that forecasting, game. And they can offload that risk to us because then we are building the infrastructure to meet those services to deliver those outcomes. Yeah. I think you, you, you say the important thing
here, STaaS, which is risk. The, if you, you know, traditionally you'd have, well, once you've cooked up all that you're going to need in terms of capacity and performance, you'll go and do, do, do a tender process and end up buying something. And then if it turns out that you needed more capacity or performance,
then it became your problem. With, Evergreen//One, the, those capacity and performance SLAs, if we don't deliver, it's our problem to come and fix them. But I think another way, things change is And it, it actually links, links back to what I was talking about in terms of those private cloud
is we've seen a lot of the infrastructure team in our customers, their role has changed. They are not the people that are provisioning, say, storage, volume, file systems, or bucket anymore. What their role has evolved to is they're now sort of a product manager, so they define services, based on what their own customers are asking them, and that's also where
something as flexible as Evergreen//One is interesting, is you can match it to, their exact requirement. And then once those services have been defined and there's clear SLAs, they through a cloud portal or something that will go and automate the infrastructure consumption. And then their, their role becomes to just monitor, are we actually in
line with our predictions? Are, is this service used more than this one? Why do we need to Do we need to review our customers are actually using versus they told us? And so on. So it's, it's a lot more of a proactive,
service-type designer than a day-to-day provisioning type of role. Mm-hmm. What does, a potential customer for you, what would they do if, they're thinking, "I really want to move on this now. This sounds much more like the sort of IT department that I want to run," but they've done a five-year forecast and are a couple of years into it, and they're
in that mid-cycle position. How can they move from where they are now to, well, where you would want them to be? I think it, I, it really depends on their specific situation, but you could There's no, with, with Everpure really, there's no downside to starting off small. So maybe they've done a forecast for their existing infrastructure.
That's fine. But if there's a new project, you can start off with a, a small Evergreen//One deployment. And because it can non-disruptively grow in any dimension, performance or capacity-Or change the service levels. There's really no downside to starting off there.
Now, having said that, with our ex-existing Pure Storage customers who've bought arrays can convert into the Evergreen//One service, so we have that option for them. Yeah. A-and when STaaS is talking about starting small, the, the, the minimum amount of capacity, the, the, that, f- to, to, to commit for, from an Evergreen//One point of view is 50 TiB, and a no more than 12 months, as
the minimum contract is 12 months. And even, even for some of our customers that are, that have been Really the way they've been doing things Forever is more around the ownership of asset. What, what happens is they constantly have unplanned requirements are coming, some new project just pops up between, in the year that no one knew about it, and then
they're scrambling to find capacity. Having, something like Evergreen//One as a buffer so that when the business you actually end up saying yes rather than, "Oh, I need to figure out what I can do in maybe two months before I can accommodate your re- your request," is much better. So, basically what we're saying is obviously, they W-we've got, some
100% Evergreen//One, but it's actually also okay to have Evergreen//One, just for the unplanned requirement. Yeah. I, I guess, Savvas, it's a question of behaving differently internally as well, learning that, you budget for, for projects and you provision projects in a slightly different way.
Yeah. It's a, it's a bit like, I think have got used to it in the way they budget for cloud, so it's very sim- It's thinking about on-prem in a similar way. But interestingly, in today's climate, Evergreen//One is a way to lock in that pricing.
So your small project that maybe isn't consuming that much today, you then have a predictable cost of expansion. Whereas if you had to go and buy new infrastructure in six months, you know, who knows what that price is gonna be. Okay. So as we near the end, it's time for the
advice on what anyone who's watching this and thinking, "Yeah, I do want to go in that direction," what they should do next, what they should be doing today, what resources they can look up. What do you suggest? S- so I think there's a, The, the first thing I would say is start by thinking about, if you were to build a new service for your customers, what would it look like?
And actually, a lot of organizations, they have something already defined, right? So maybe it's a bronze, silver, gold class of service, that's, that they have for some application. Well, just start thinking about what would you say is required there in terms of capacity, performance, and, you know, how much
you're going to keep data for. And then start putting together a S- a simple matrix of, what those, data storage classes of service would look like, and how you would make them available to customers. Is that something that, your own internal customers would want to consume, and in which manner?
Um- Savvas? Yeah. We, And then so once you've put that together, we have a service catalog that describes the SLAs. So in addition to the guarantees we have on the slide, you, there are the different, service levels with the performance and the capacity, amounts there.
So I'd say compare that, and it's And then it becomes a conversation with the Pure, Pure team, and to designing what does that service look like. And in terms of getting started, we have a very predictable, four to six start time. So from when we contract a new service, it's four to six weeks to, that we have an SLA to, to get that SLA, that service up and running
in your data center. Okay. And I note that we have the omnipresent QR codes for you to, go to some of the resources, the Evergreen resources that you offer there. Yeah. Indeed. Okay. Guys, that's it.
That's all we got time for. That's fantastic. Thank you very much. That was very clear. And, so, thank you. Thank you, Savvas. Great. Thank you. Thanks for having us.
Thanks, Fred. Thanks, team. Thank you for having us. And, thank you to, all of you who tuned in to watch this. As I always say, hope you found something interesting, and if you do want to find out more about it, let us know, and we'll put something on for you.
But good luck in the future. I know you've got a lot of challenging projects coming up. And, if you go this route, let us know how you get on with it. Thanks to everyone at the Register behind the scenes for putting this on. It just remains for me to say, from me, Tim Phillips, see you soon.
Goodbye.
Savas Nicolaides
Lead Principal Technologist EMEA, Everpure
Fred Lherault
Field CTO, Everpure
In an era defined by AI-driven demand spikes and volatile market shifts, the traditional 3-5 year storage forecast isn’t just challenging - it’s a significant business risk. Modern IT faces a fundamental crisis: the inability to accurately predict future demand, especially for rapidly evolving AI workloads. While organisations grapple with unpredictable lead times, project timelines continue to move faster than traditional procurement can accommodate.
Fred Lherault and Savas Nicolaides explore a critical shift from a “Buy and Build” mindset to an “Outcome and Service” model. Learn how adopting the Evergreen architecture can eliminate the risk of making the wrong hardware bet, prioritizing business agility and operational resilience over simple asset ownership.
Key Takeaways:
- The Myth of the 5-Year Forecast: Predicting the future is risky; respond to your environment in real time.
- De-Risking the Data Centre: Achieve cloud-like agility on-premises with guaranteed service levels for availability, performance, and capacity.
- Solving for the AI Unknown: Deliver a certified AI factory with guaranteed performance without overpaying for storage before it’s needed.
- Outcome-Based Resilience: Transform cyber-recovery from a “tool” into an SLA-driven service capable of shipping clean arrays within 24 hours.
We Also Recommend...
Personalize for Me
My Updates
No updates at this time.