The Edge Landscape and how Innovative High-Density Storage Transcends Edge Workloads

Patrick Moorhead:
The Six Five Summit is back, it’s 2024, and we are talking about AI. We’re talking about AI and the impact of the enterprise, the continued build out of AI infrastructure, which is, as you know, in the data center, but also on the edge. Daniel, how are you doing, my friend?

Daniel Newman:
Yeah, Pat, it is good to move on from some of the data center conversations. Of course today we have our cloud infrastructure track, but we also are here on our edge track. And Pat, we’ve heard this many, many times about the size of the opportunity. And of course, data centers are exploding. The AI factories and data centers are being built because we know the future workloads and how important this is going to be. But if you think about things, whether it’s vehicles, factories, whether it’s retail stores and operations, there is so much data being created at the edge that it would be a big miss if we didn’t spend some time talking about that.

Patrick Moorhead:
Absolutely. And Dan, you know I love to look back in history and literally over the last 40 years, every major shift has been a movement. It’s like an accordion. We go out, expands, contracts, so we saw mainframes, mini computers, X86 servers, PC, smartphone, tablets, and then the industrial IoT. It just goes back and forth. And I feel like we are in the mode of this moving out. And historically compute storage and memory typically wind up being where the data is created, and that’s the edge. So I’m super excited to have this conversation about storage on the edge. Dave and Greg, welcome to the show.

David Dixon:
Hey, yeah.

Patrick Moorhead:
Welcome guys.

David Dixon:
Thank you.

Patrick Moorhead:
That’s great. Hey, can you just tell us real quickly what Solidigm does for those who might not be aware?

David Dixon:
Sure, yeah. So my name’s Dave Dixon. This is Greg Matson. I’m a co-CEO in Solidigm. And Solidigm was created back about four years ago. We were an acquisition by SK hynix from the old Intel NAND products group. So now that we’re really a company and entity, et cetera, we have a unique focus compared to everybody else in the NAND business. We’re really entirely focused on the data center business, but that of course includes the storage servers there at the Edge as well.

Patrick Moorhead:
I love it. Greg, how about yourself?

Greg Matson:
Well, I run strategy and marketing for the company. So anything to do with products. And from definition to go to market to doing stuff like this.

Daniel Newman:
Let’s talk about the growing opportunity at the edge. Of course, we’ve enjoyed having a couple sit downs with you. People that are watching our infrastructure track may see these gentlemen again in another conversation. But you are really a perfect example of that data center to edge. You heard us teeing it up. How big is this opportunity? Do you see it and talk a little bit about it specifically through the lens of storage.

David Dixon:
Well, yeah, as we did talk about previously, we looked at the scalability of the infrastructure in the data center and how the high-performance, high-density, QLC SSDs are really going to make it be a game changer for data center storage. Now, like you mentioned, edge growth is probably going to be at least 2X faster, what we see in our projections faster than what’s happening in the data centers. So that opportunity is going to be even bigger there for edge servers, et cetera. And they’re really going to have the same growth drivers. If you look at what’s happening on the edge, I think we’re familiar with those magazine articles from not too long ago, data is the new oil.

And all that data, it’s all being created obviously out there on the edge. And one thing we’ve been talking about internally because we’re very familiar, very close to data, it has a lot of what we call gravity associated with it. It’s very costly and expensive to move. It just want to stay there. If you really want to try and shift all that data back to a data center, back up to the cloud, to complete all the analytics and AI training, et cetera, it’s too darn slow, too darn expensive to do it. So we’re going to see more and more trends of analytics, computation, even light training happening more at the edge and we’re seeing that now. We’re going to continue to see that it’s going to be a mega trend over the next rest of the decade.

Patrick Moorhead:
Yeah, so Greg is this classic, we always like to put smart, smart manufacturing, smart logistics, anything done with cameras? Is this the flavor of what we’re talking about on the edge? Retailing?

Greg Matson:
Retailing, there’s also detection for fraud or quality detection network, maybe power grid failure detection and just anything like that. But as Dave said, it’s so expensive to move the data that it’ll have to be that some level of compute gets moved back to make the data useful at the edge because if you just store it there and do nothing, it’s worthless.

Patrick Moorhead:
So it might be obvious to some that there’s a difference between storage related to the data center and the edge, but I think it’s worthy of a conversation. How is it different? Is it different from reliability, performance, capacity, form factors? How is it different and why is it different?

David Dixon:
Well, I think we’ll see some of the same trends that we talked about in the data center itself in terms of power and performance. But you mentioned reliability. I think that’s an interesting one. It’s much harder to have service stuff out in the field and of course than inside the data center. And so you’re going to see longer expected lifetimes and just the reliability requirements of comparisons between SSDs and HDDs. Historically, the trends have been about 10X better annual field rates between SSDs and HDDs because of the mechanical aspects of HDDs of course, when you talk about that more and more deployments out in the field and having to service those higher failure rates out there, that’s going to be a much bigger deal.

Patrick Moorhead:
So at data center, what are some of the constants, raised tile flooring, every one of them. And now we’re even seeing very dialed-in cooling types of systems. It’s literally a science as you go down each one of these floors, but also each one of these aisles. And on the edge, it could be a server bolted to a brick wall that the air conditioning sometimes doesn’t work and it goes out, could be in the elevator shaft of a company, on the shop floor of a major auto manufacturer. So the variability certainly has to be that challenge too.

Greg Matson:
Absolutely. And that’s where the ruggedness of a solid state drive comes in versus a hard drive. And as Dave said, even the reliability 5 to 10X or more, in some cases, reliability differences translate into just ease of deployment of SSDs versus HDDs at the edge.

Patrick Moorhead:
And when that hard drive goes down, business essentially stops.

Greg Matson:
Can stop, you lose your data. Maybe it’s okay, maybe it’s not that critical data. Maybe you’re a big groceries chain and you had your one store lost all its records for the day and you’re not uploading the revenue back into-

Daniel Newman:
Sounds like a big deal to me.

Greg Matson:
That’s a big deal, but it ranges.

Daniel Newman:
You guys could call out maybe some of the opportunities too. Reading on this, you’ve made some commentary about ADAS, on research. Talk a little bit about even more how specifically SSDs can up-level, because really what you tend to be talking about the edge, you tend to be talking about a lot about service providers.

David Dixon:
We talked about the reliability and some of the ruggedness. Now where I think we’re getting into some of these real-time applications of what’s happening at the edge. We’re going to really going to look at the much higher performance requirements right now. We’re going to be doing real-time decision making on the edge. And that really requires much better latencies. You’re not going to be sending the data back up to the cloud to make a decision about what to do in a factory, for example, or in the middle of an operation. Those decisions need to be made real-time between the computation and the storage. And between the random read performance differences, then the quality of service latencies. It’s no comparison between what’s happening with the legacy 90% HDD in storage right now to where SSDs are going to really be the much bigger impact.

Patrick Moorhead:
My company’s done research on AI and the edge probably about five years, and it was machine, not really deep learning, because the [inaudible 00:08:58] a black box. But when I compare that to generative AI, it’s smaller data. Even a small model, a measly billion parameter model that I have to infer against is still much bigger than this random object recognition AI framework that I pulled down off of hugging face. And then what we’re seeing too is the co-mingling of data on the edge brings storage to the forefront. So it seems like a big opportunity.

Greg Matson:
Well, and that’s where the capacity density that you can get with high capacity solid state drives is as much as 4 to 5X more than the highest capacity hard drives available today enabled by both the form factors, but also just the sheer capacity of each drive. Today we’re shipping 60 terabyte drives and the nearest competitor hard drive is 24 terabytes moving to 30 later this year. And by that time we’ll be well ahead of the 60 terabyte capacity point.

Daniel Newman:
So you’re teeing me up very nicely, Greg, and I appreciate you doing that. But portfolio, let’s talk about that a little bit. Because I think we’ve talked at a macro level about what the challenges and opportunities are, where storage fits in. But what about Solidigm? So what are you building, Greg? Talk a little bit about the products and how they’re addressing these challenges that we’ve outlined.

Greg Matson:
Well, for several years now, we’ve been looking at the opportunity to how do we capture more of that cold data that’s sitting in hard drives and transforming it into warm data. And what’s required? What’s required is a broad selection of form factors depending on the type of server used. So at the edge you might have a different form factor than you use in the core data center. Also, flash technology. So we’re on our fourth generation of quad level cell flash technology, very well proven, highly reliable, and we’re translating that into the highest capacity SSDs on the planet’s.

David Dixon:
It’s right there, by the way.

Greg Matson:
Oh yeah.

Daniel Newman:
Show and tell everybody.

Greg Matson:
Here’s one of our 61 terabyte drives today. Those things, I got to tell you, are hot commodities these days with all the AI build outs, both in core data center, but also at the edge.

Patrick Moorhead:
I have to ask, some people might be watching and they’re like, “Okay, what’s the catch here?” What am I missing? Now that I’ve got QLC and the ability to have a ton of error correction, performance is good, reliability, what’s the catch here? What objections would you possibly get from this or are you getting from this? And how are you overcoming them?

Greg Matson:
That’s a good question. First is everyone knows that from a CapEx perspective on a per gigabyte basis or per terabyte basis, flash drives are more expensive than a hard drive off the shelf. But what we’re seeing with building out this new technology type of data centers is that CapEx is quickly overshadowed by the other costs, whether it’s power constraints, whether it’s space constraints, and in many cases it’s the constraint of leaving a GPU unutilized. And all the big data centers today. From a core data center’s perspective, are already moving to high capacity flash. And we’re seeing that trend at the edge as well, where the space is even more at a premium.

Patrick Moorhead:
The transition has been magnificent. And when you’re literally, you have companies that have to outlay billions of dollars in CapEx, if they can possibly shorten that even by a couple months, that’s a tremendous amount of capital and working capital that could be used.

Daniel Newman:
Yeah, I look at a calculus and those that are doing the very, very simple math, to Greg’s point, are going to often land at HD. But if you’re doing the full calculus workup, I think you’re going to very quickly see that rotate.

Patrick Moorhead:
Yeah, I know it’s a little bit different on the edge, but that’s where the comes in and the speed. If you’re trying to run the models out there on a hard drive, that just seems like something [inaudible 00:13:24].

Daniel Newman:
I think if you have a two-foot-tall rack, that’s all you can get in there on the edge. Anyways, we could have the debate, but the TCO debate I think will be very interesting. And I think when you do that long-term CapEx in return on capital, you do that math. I think you’re going to increasingly see what you’ve built becoming more and more of a winner in that formula. But again, it takes time. It does.

Greg Matson:
It takes time. But it’s a little bit miraculous how quickly all of a sudden the sheer compute intensity of AI, the cost of all the other components in AI, all of a sudden switching to solid state, it’s almost a no-brainer.

Patrick Moorhead:
Yeah. And By the way, on the power front, my final volley here is if I move from hard drive to your SSDs, I could have more power available to a GPU.

Greg Matson:
That’s right.

Patrick Moorhead:
Right. And that’s a conversation, I like to call that power sloshing. I don’t know, everybody likes it, but essentially I’m moving this power budget over to here. And by putting SSDs on the edge, you’re giving them that opportunity. That’s right.

Greg Matson:
And maximizing your investments in GPUs

Daniel Newman:
And that my friend was the whole calculus. Greg, Dave, I want to thank you so much for joining us here at this year’s Six Five Summit.

Greg Matson:
Our pleasure.

Daniel Newman:
Thanks guys. Appreciate it. All right, everybody stay with us. We are here in the Intelligent Edge at the Summit 2024. It’s a Six Five Summit. I don’t know why it looked like I might’ve forgotten that, I didn’t. Stay with us for all of our sessions. Another great year, another great event. We appreciate you. We’ll see you back soon.