How is AI changing the way we store data?

- Good morning, Michael.

“- Good morning, good morning, good morning.”

How do you like to store information at home? Like, do you have a hard drive? Do you store stuff in the cloud? Are you sort of a print out and find it kind of person? - I think it kind of depends on what it is.

I put some stuff in the cloud, but I have some things home movies and stuff like that on the hard drive. They think even today, like going back to like writing stuff down, is almost a little bit more safe if it's like really, really important. What about you? What do you use?

Because I feel like you probably have multiple hard drive backups of literally everything. - I feel so seen. I have a NAS networked touch storage, which has a four hard drives in a raid.

I live by the rule of three, which is your three copies of any file you don't want to lose. So now, Sam, while the sort of backup methods might work for regular everyday people like us, when you scale it up, the quantity of data

“that you have to store in organizations,”

and the methods you have to use to store it properly, can become significantly more complicated. So this week, we are diving into the topic of data storage, and in particular, data storage in the world of AI. - I'm Michael Bird.

- I'm Sam Gerald. - And welcome to Technology Now from HPE. (upbeat music) We've talked a lot about storage on this podcast over the years.

How you store data, where you store data, that sort of thing. But do you know how much data we are creating nowadays? - I'm sure it's going to be like some sort of made-up term plus the verb byte. I'll be fine, and we'll call it a Zeta beta byte.

(laughs) - No, Decahedron byte, that's sort of fair. - Yeah, Decahedron byte, I like that one better. - Decahedron byte, which is a copywriter. Now, according to research from Petrogtaila in 2025 and published

on statistic, by the year 2020, we were creating and utilizing over 64 Zeta bytes, there's the word. So Zeta bytes of data, and in 2022, that number had risen to over 100 Zeta bytes, and predictions put last year's data at over 180 Zeta bytes.

Now, to put that completely arbitrary word into perspective, the latest smartphones can have one terabyte of storage and to get to 180 Zeta bytes.

You would need over 100 billion smartphones.

- Wow, that's a lot of smartphones, and we have, of course, linked to all of these stats, as well as all of the other sources in the show notes. - Call that big data, my goodness. That's sort of phones, it's a lot of data.

I'm sure not all of it is like high quality data, either. Most of the data we create and use isn't designed to be stored long term. However, as the use of things like artificial intelligence continues to rise, this is beginning to change,

because as you know, AI both uses and can also create huge amounts of data, which needs to be stored, and then processed, and organizations often don't want to dispose of this data until they can confirm they no longer need it.

So, to find out more about data storage in this era of AI, I spoke to Jim Odorecio, Senior Vice President and General Manager HP Storage. - But before we speak to Jim about modern day storage technology, I want to take a look at how we got here,

and some of the designs which didn't quite make the cut. It's time for technology then. The art of storing data goes back to the birth of civilization itself.

We've always needed waste record information.

Books, libraries, and clay tablets were always in which people could preserve knowledge, but digital memory is obviously a much more recent concept.

“Michael, do you know what was the first fully electronic memory?”

And do you want to guess when it might have turned up? - Okay, I'm gonna say 30, something, 30s. - Not a terrible guess. Most computers were beginning to come into their own post-World War II, but the mechanical memory, which is made of moving parts,

was causing progress to stall until two researchers, the Freddie Williams and Tom Kilburn, in Manchester, England. - Yeah. - Created a tube, but this tube was special.

Inspired by cathode ray tube storage, which Williams had seen during a trip to America,

the Williams Kilburn tube became the first

entirely electronic memory, and it worked by storing bits as dots on the surface of the tube. Like with a cathode ray tube and a TV, the Williams Kilburn tube required constant refreshing

by the cathode ray to stop the dots fading and disappearing. Information on the tube could be read using a metal pickup plate near the screen, which would detect a change in the electrical charge when the cathode rays swept across the screen.

However, the Williams Kilburn tube faced a few issues, which would eventually turn out to be pretty insurmountable,

have been hand-tuned frequently.

“They were also notoriously sensitive to the environment,”

and could be interfered with easily. So when they were excellent, when they worked, their unreliability was kind of a huge issue. So even though they were so good that they were used at, for example, Los Alamos,

the Williams Kilburn tube finally fell out of favor

and was replaced by a magnetic core memory. - Yeah, I think storage is something that you want to be somewhat reliable. The idea that you had to want to be hand-tuning things. - No.

- Now, today we use semi-conductor memory, and you might have heard of things like DRAM or Dynamic Random Access Memory, but the world is still changing faster and faster. So to find out more about where we currently stand,

I spoke to Jim Odericio, senior vice president and general manager HP storage. And the first thing they asked him was how storage has had to evolve in the past decade. - It has been an interesting ride to be honest.

If you think about 10 years ago, there was a major move of workloads into the public cloud. So there's a major migration of workloads along with its storage into the public cloud.

At the same time, there's been this explosion

of data on structure data in particular. It's estimated that about 200 zeta bytes of information would be created in 2025 along. It's mind boggling. But as that's happened, and storage is becoming more intelligent,

and it's having to become much more performant, workloads are being moved out of the public cloud and back on premise because there's data sovereignty issues, there's security, there's performance. And then with the age of AI,

“it's much more important for our enterprise customers”

to have sovereignty over their own data. - And you talked about petabytes and petabytes of data. Has the type of data we're creating changed? Like are we creating more unstructured data? - Well, if you think about customer data,

if you think about audio and video, all of the data they get generated from a social media perspective, but increasingly, it's a richer data set that we interact with our customers with,

all that data is getting stored. Everything that might have been paper transactions 20 years ago is now digitally transacted. So all of that data is available, accessible, and really available for insights to be extracted from.

- And presumably the rise of AI has had quite impact on just a sheer amount of data we're creating. - Is sure it does, and it has a huge impact on what we say, because obviously storage has become much more efficient and much more intelligent.

So you can automatically store more. Solid state devices flash devices, they're faster, they're more reliable, they're getting larger and larger. The ability to store massive amounts of data efficiently is increasing every year.

And as such, customers can store more data and they can contemplate what they want to do with the data and how they can leverage it for competitive advantage. - And how did the migration to cloud, and I suppose, perhaps the migration in science

and since it's away from cloud storage,

“how is that impacted the way we think about storage?”

- You know, I think initially everything was going to the cloud, and I think there's a bit of an equilibrium now, but AI in particular is forcing customers to consider pulling data back into on-premises, because there's the ability to operate on the data

with the security, the performance, he has placed new requirements. And so you are seeing some elements of data moving from the cloud back into on-premise locations. - I want to dive into AI in a bit more detail.

It's still the buzzword at the moment, isn't it?

And so why, when we think of AI, is storage so critical?

- Well, certainly the sheer amount of data that's being generated is critical, but the ability for storage to evolve and become much more intelligent and supporting these AI workloads

is becoming much more critical. From intelligence perspective, we're talking about being able to add metadata as the data stored on the storage layer. - Right, okay, okay.

- And certainly, you know, the ability to do things like calculate vector embeddings, as the data's being stored, because then the data's ready for AI operations as it lands on media, and that's a huge advantage. - So what we're saying is actually,

you're taking that piece of data, you're loading load of metadata, it lots of additional information, which the AI algorithm can use to do something with. - Any left, yeah, fast.

- And they get it faster because the idea of the intelligence storage layer allows the data to be AI ready. In the past, if you had a less intelligent storage layer, you would have another post-processing activity,

you would store the data, and then you would run it through a GPU, and you would calculate vector embeddings, and you would do those sorts of things. They would allow you to do inferencing,

and rag across your data. Those sorts of things can be done as the data stored with an intelligent data storage platform, like the ones we have here at HP. - All the different categories of data,

which we discussed before, all being used here, or just AI focus on just one or two of them. - We're talking mostly about unstructured data, and it could be anything from contracts to analysis of contracts, PDF files,

multimodal data that has audio and video.

would be very manual job to make sense of. - Extremely manual, and limiting in terms of the kinds of value that you would get out of the data, and the speed at which you could derive those insights. - So storage is sort of the foundation

that AI has built on.

“- Well, increasingly, it's super important”

because we started off with GPUs that compute nodes and servers, and that continues to exist. But now, companies like HP, or building, intelligence into the data storage layer,

it's becoming essential because it allows the processing

of the information and the value of the data being extracted to happen that much more efficiently and that much quicker. So it's very foundational at this point, and as we talk about zedabytes of data being created yearly,

it's important that we have a flexible infrastructure that it exists. We talked about data has gravity, that we can deploy that infrastructure, where the data exists, where the data is being generated,

that we have the processing of that data, local, to where the data exists. So having this flexible, powerful infrastructure and really relying on the intelligence that the storage layer can deliver is absolutely essential.

- So we sort of took about storage, help him with AI, but how can AI be used to sort of improve storage? - At HPE, we literally use a Gentic AI in our storage products, and it's used to improve the operational efficiency of our storage. It's used to do things like improve

“the proactive support experience for our customers.”

It's used to solve problems before customers even experience the problem. I like to say some of the best AI that we've written, a language models we've written here at HPE, actually running in our storage layer.

- And that's sort of doing the jobs that previously would be again, pretty time consuming. - Let's proactive less efficient resolution of challenges for customers, and increasingly, customers want the self-healing,

self-managed infrastructure, and having AI built into the storage layer allows that to become more of a reality. - Yeah, and don't just go destroy his about to fail, more intelligent than that.

- Yeah, right, exactly right. Yeah. 'Cause that's what he used to be. He used to be a, your dish job is about to fail, and you got,

"Oh, this is great." - Or it fails, and you log a case, and somebody shows up, but now, much more proactively, we're looking at a lot of data, minute by minute, to predict what's gonna happen

as opposed to waiting for it to happen. - And I suppose, is there looking at trends in how, say, I don't know, a storage array is being filled up. For some reason, there's one site that's generating lots of data, so maybe actually on that site,

what increased the storage in that. - Well, we built into our systems as the ability to identify performance hotspots, to do trending analysis, to understand what this noisy neighbor problem where you might have certain workloads

that are impacting the performance of storage, and then making automated recommendations to customers to how to modify it, to get better efficiency out of the storage. - Right, right, okay.

- Basically, make the most of the investment, you feel exactly. - Really, giving the customer the information that they need to be as proactive with their infrastructures, they possibly can be. - 'Cause I think maybe two decades ago,

there was a sense of just bias, just bias, much as you think you might need. It wasn't particularly intelligent. I realize we're just talking pure storage here, but I suppose actually, if you could be more efficient

with it, you could make more of your money, but it's all about efficiency, because enterprise budgets are not increasing. We have to carve out funding to go do AI projects.

And so it's that much more critical

for the infrastructure to run efficiently, and to really provide the kind ROI customer requires. - Planteau, date to sovereignty. How are store agents, sovereignty linked? It feels like a very obvious question.

- Well, you know, sovereignty is becoming much more important especially in the age of AI, and especially in Europe, where rules and regulations at a country level prevent data from leaving the country that it was created in.

And so having a storage infrastructure that's flexible, that can be local to where the data's generated is absolutely critical. And from an HP perspective, we're doing a really good job of that.

- Do you expect to see more conversations around sovereignty in the future? And if so, how will we respond in terms of the way that we create things, you know, that the products that they're creating,

how will that maybe perhaps change? - From a sovereignty perspective,

“I think we can expect more concerns and more regulations”

around sovereignty as the world becomes much more complex, and countries are much more concerned about data leaving their borders. So I think from that perspective, our ability to deliver our management capabilities

that run locally, it creates a particular challenge for some of the public cloud providers, because their infrastructure exists in virtual places, and it's very difficult sometimes to ensure the data doesn't leave a particular country.

- And I think it has the effect of driving

critical content back on premises,

because organizations can control it there. They know where it was created, they know where it's stored, they can put compliance around it. - And will that be more emphasis on, say, reporting on that data?

- I think based on the way you built your infrastructure, it's very auditable, you know, where your infrastructure is,

or they sit within multiple data centers,

all within your country. I mean, if it's under a customer's control, then it's very easy to determine. When you get into a scenario when a public cloud you're pushing data that you think is in a data center,

but it's replicating to another country. That's a fairly common thing. - All right, I wanna ask the question that every technologist hates to be asked, which is,

“where do you see the world of storage in SIF5 or 10 years time?”

- Over the next five to 10 years, you're gonna see faster, solid-state describes, you're gonna see more intelligence built into storage. You're gonna see larger and larger capacities over the next five to 10 years.

And you're also gonna see a lot of advanced development around things that probably won't come to market in 10 years, but you'll see a lot of advanced development around holographic storage, around DNA storage. You'll see a lot of development in those dimensions, yes.

- 'Cause Moll is low, has I an equal?

I'm guessing with storage,

with miniaturizing miniaturizing miniaturizing, with talking flash storage. - You know, obviously, as you think about the efficiencies of man and the ability to store more and more data and smaller and smaller footprints,

you're gonna see capacities continue to accelerate. So you're gonna see performance, capacity, grow, which along with the intelligence that will continue to build into the storage layer to support these AI workloads.

- Well, we start to see what's a cool storage, do you think? - Oh, absolutely, I mean, certainly, we have HP, we have water-cooled racks, we have water-cooled everything, and I think, as you build more intelligence into the storage layer,

you know, you'll see GPUs running in storage, and you'll get to a scale where, you know, water-cooling will likely become much more prevalent, even at the storage layer. - Jim, thank you so much.

Field time. Thanks for coming on, Statorgia now. - Well, thank you, I enjoyed it. Thank you very much. (upbeat music)

- Wow, the idea of holographic and DNA storage is pretty crazy to me. - Yeah, I thought the same thing. We did an episode, what we talked to a research of who was doing DNA storage.

I wanna come on to different storage methods in a bit,

“but my big takeaway, which I think we all know,”

and we've talked about many times before, is AI is all about the data. We've talked about good data in, good data wrap, bad data in bad data wrap, but fundamentally, you need somewhere to store all that data,

and you need to process all that data, and we did an episode a few weeks ago, what we talked about networking, and talks about how networking needs to be running really efficiently, and this conversation,

the same is true of storage. - Yeah, that's true.

Basically, it seemed as though storage with AI

was getting closer and closer to the time of the creation of data to leveraging it efficiently, and so that, to me, just sounded a lot like the edge, which we normally think of in the context of networking. So, it's all kind of starting to become the same thing

to some degree, especially once you start talking about using GPUs in storage, it all seems like networking, storage, and AI, and compute are all just one kind of cohesive thing at a certain point. - I remember storage arrays, like one of the big selling points,

it will let you know before a component fails, and I think Jim said, "I'll pull up the supplier, "and it'll say, my hard drive is about to fail, "and my mom will get shipped out to you." And that was sort of the extent of the intelligence,

but actually, that's just like one element of it, there's so much more that could be optimized, could be tweaked, could be fixed. And again, just like, as you talked about, actually managing that infrastructure can be quite a burden

to identity departments, and again, as Jim said, the budgets haven't got any bigger, so we have to do more with less. - Yeah, that's true. I love what you brought up though about it,

letting you know about support, and at the time I could see where that felt very, very useful, but any more, like, the standard changes, right? Like, now I imagine, as they have these different self-healing products and storage arrays,

that it's, like, don't just tell me, do something about it, right? Like, do it right now yourself. - Yeah, just going into it. - Yeah.

“- I think the conversation that's up with the show”

that we talked about, storage needs to be more intelligent in the world of AI, and essentially, like, loading metadata onto the data, he used the phrase vector embedding. It's a metadata, it's a piece of information

that will go with whatever it is to make processing their data even faster. And again, we've talked about this in the show before. AI is about speed, AI is about making the most of the resources that you have.

- That makes me happy, actually, because it makes me think of maybe, like, there's a whole sustainability component to all of this, too, because in the past, you would just buy lots and lots and lots, lots of physical storage,

or you'd expand endlessly into the public cloud, which was also costly and crazy, but now you're at least optimizing everything, so that you're using exactly what you need and not necessarily going overboard with what you don't.

- Yeah, exactly. Now, the same as you said at the top of the show, technologies can arise in full as times and systems changed. And actually, that does touch on one of the things that I really wanted to ask you about.

- Oh, what's that? - Well, I wanted to let Jim thinks that hard drives will disappear one day, 'cause we're sort of, it's all flash storage now, isn't it? - It's interesting question, because 25 years ago,

a few of us, me, would tape still be around. I would have said, probably not, but still is. It is still is, I've heard it said, that tape is dying at a glacial pace, right?

from spinning discs, but at some level,

it'll become too inefficient to make spinning discs, because there's a lot of technology there, and is the footprint shrinks down from an SSD perspective, and you could store more and more information.

“I think the prices will start to converge,”

and at some point it won't make as much sense.

- Interesting, interesting.

Do you think it will die at a glacial pace today? - I think it will take some time. - It's not going to be this year or no. (upbeat music) - That brings us to the end of technology now for this week.

Thank you so much to our guests, Jim O'Dorecio. And of course, to our listeners, thank you so much for joining us. - And if you've enjoyed this episode, please do let us know, rate, and review us,

wherever you listen to episodes.

“And if you want to get in contact with us,”

send us an email to [email protected]. Subjectline, zeta-bikes. - Do-deck-a-heedron-bite. - Do-deck-heedron-bikes.

- And don't forget to subscribe so you can listen first every week.

- Technology Now is hosted by Sam, Gerald, and myself, Michael Bird,

“and this episode was produced by Harry Lamput”

and Izzy Clark with production support from Alicia Kempson Taylor, Becky Bird, Drew Leatso, Alyssa Metrye, and Renee Edwards. Our theme music was composed by Greg Hooper. - Our social editorial team is Rebecca Wissinger,

Judy Ann Goldman, and Jacqueline Green. And our social media designers are Alejandra Garcia and Ambar Maldonado. - Technology Now is a fresh air production for Huller Park at Enterprise.

We'll see you the same time, the same place next week. Cheers. - Bye, y'all. (upbeat music)