The Sovereign Disruption: Why AI Requires a New Approach to IT Architecture

Hey everyone, we're back here at SUSECON, day one coverage. Keynotes ended. Well, we've been doing interviews, but it was probably closer to 45 minutes, an hour ago.

Want to introduce you to someone. If you watched our SUSECON coverage from last year, I thought he was one of the stars of our coverage, so I was thrilled to see him on the list back this year. Mark O'Regan.

Mark is the CTO for EMEA for Dell Technologies. And look, this show's in Brussels. This show is in Prague, so we're obviously in Europe, but EMEA, there's so much going on here, right, in terms of an involved legislature and people really thinking about issues like sovereignty and AI and these things.

Mark, it's great to have you back here. How have you been? Thank you.

I've been great, yeah. Thanks a million for having me back. My pleasure.

It's one of my favorite trips of the year. Ah, me too. Coming out to SUSECON, I love it.

Me too. Me too. Thank you.

Before we jump into what I wanted to go over with you today- Sure ... give people a little idea of what you've been up to lately, and what are you seeing in... You have a catbird seat, right?

Sitting here. On one hand, you have the worldwide Dell Technologies, which is, I don't know, what is it, 98% of the Fortune 500 or something like that. Yeah.

And on the other hand, you're here in EMEA as we're grappling with issues around AI and sovereignty and open infrastructure and so forth. Talk to us about where your vantage point is and what you see. Sure.

Yeah. Okay. Well, from an EMEA perspective, obviously Dell, we're plugged into that, as you say, the big Dell world, right?

At the office of the CTO, it's a little more focused on emerging technologies and what we're seeing and what we're doing in that space. Mm-hmm. So it's very much here and now what we're seeing in industry, what's happening, the things that you mentioned earlier on, sovereignty, artificial intelligence, what's happening in the quantum technologies- Sure ...

architecture space, and how we're making sure that we're not missing an inflection. That we're doing the right level of research on those technologies, that we're making moves that get us around the corner and ahead of them when they hit prime time. So there's that work that we're doing in the here and now and what we're seeing.

Get back to that in a moment's time. And then there's the work that we're doing three to five years out, which is that sort of forward thinking, and then five years and beyond. So yeah, we're in strange times, I think it's safe to say.

There's a lot of fractionization going on and a lot of polarization in certain areas, and we're seeing that bounce into tech, which is really interesting if you're a technologist like myself, I guess. We're obviously seeing this big explosion, right? We spoke about this last year in artificial intelligence.

This happened November 2022. Then we're having conflicting opinions on the adoption, the uptick of AI, who's doing it, who's not, and why. And that's an interesting thing to look at, right?

So we had your guys in MIT, we have a very close relationship with MIT. They released a paper late last year, and shocking, I think it was somewhere around 95% of people who were assessed did AI pilots, and they failed. Yeah.

Right? " Right? The world is heavily, heavily invested in artificial intelligence, and not just the models, but the functions.

We see that with the emergence of agents and agentic architectures and techniques now starting to make their way into industry. So why are we seeing this failure? And I think there's a broad misconception, and in some cases, even lack of understanding of what artificial intelligence really actually is, and what the expectation of AI is and should be as it makes its way into your specific ecosystem, and how you think about your return on investment.

The conversation about cost and all of this on-prem, off-prem, that's a sidebar conversation. Yeah. The conversation today is around return on investment, and that return on investment isn't just monetary, right?

No. It's down- Dollars and cents that's measurable. Absolutely.

But if you look at how you do things, optimization, you look at process and processes, how they can be improved. If you look at confidence, so this goes back to the MIT piece. We believe that the failures there were really probably due to the fact that a lot of those pilots just weren't thought through correctly.

They were science experiments. They were science experiments. Yeah, exactly.

I feel compelled to take the counterpoint. I don't know if I'd classify them as failures as much as they just didn't have an impact on the bottom line. There was no significant revenue change associated.

Yes. I think there was learning that went on. And then there was another study done by the Wharton School, which is also not a bad university- Yeah ...

or a bad school, that showed almost a mirror image of that. That when you didn't measure it by impact on the bottom line, but you measured it by whatever positives they did, there were a lot of positives. But here's where I'm at with it.

The train's left the station Absolutely. And laying down on the tracks isn't going to stop that train either. There is the momentum, and I'm sure Dell sees this, right?

With the inference computing, where are we getting the... You can't buy memory chips, right? Right.

I just read, I think Apple is delaying their next MacBook Pro and so forth because they can't get memory- Yeah ... for it. So ready or not, here it comes.

Which- It's here. It's already here. And I totally agree with you.

Failures aren't necessarily a bad thing if there's learnings to be taken from them. Right. And I agree with that.

Fail fast, fail often. That whole- Iterate ... whole- And reiterate ...

reiterate and reiterate and learn- Yep ... and apply. I totally agree with that.

As does Michael, you'll hear Michael, Michael Dell talk a lot about fail fast, fail often, move quickly. No, sure. Have that level of agility to be able to adapt and so on and so forth.

That's really a cornerstone of what AI culture is all about. I also agree, and I'm sorry if I wasn't clear enough, success doesn't just turn up on the bank statement, right? It just doesn't turn...

Yes, you can measure it through... But looking down into process, looking down into optimization, looking into your return on investment in bringing artificial intelligence correctly, and understanding what it is you want it to do, and then applying it to that function, and measuring the success and improvement of that function as a result of bringing AI in. So this is thinking through what it is that you're expecting from the AI.

And the failure is MIT's words, it's not yours, it's not mine. Right. But that paper that was written, was written to pilots that were spun up- First generation, yeah ...

we need to be in the AI game. Tell me where we need to spin up an AI, and we'll just do it. We'll create a digital human.

We'll create a coding platform, whatever it is. Well, it's that typical Crossing the Chasm model. You get 15% of people are early adopters.

Yep. I used to say this, I have two sons. My one son, before he would dive in a pool, would go check and stick his toe in and see if the water was warm or cold, or what have you.

Let me get- Right ... the second guy, straight in. He just dives in- High toes ...

wouldn't care whether the water was in there or not. Now you have to worry about him. Yeah.

Yep. I think there's some of that. There's huge...

There is, there is. Yeah. And there's courses for courses, and there's different personalities.

There's different- Absolutely ... people. There's different digital personalities- Yeah ...

as well, right? You're absolutely correct. But here's where I am though, too.

We saw in the last two, three weeks, this Mythos project- Yeah ... from Anthropic, from the Glasswing project, to battle it. Certainly, I think we're starting to say, we're not going to come to Skynet where AI is going to kill us all or whatever, but there are real pitfalls.

There are real downsides to using AI unfiltered. The least of which may be that 95% of them fail. Some of them may do worse than fail, they may really be detrimental.

And maybe, just maybe, we need some regulations here. Until best practices emerge, the technology stabilizes, we need guardrails. Yeah.

This is a huge, huge problem. When you look at the United States, going back 15, 20 years, you would've seen the adoption of cloud. Right.

Very easy to spin up a cloud, like as your AWS or Google Cloud platform in the United States. Why? Because everything was federated under one hood.

You had 50 states, everybody went up to those policies and needed to abide by the rules and regulations that you just called out. In Europe, let alone Middle East and Africa, and Europe, which is my domain, very different story. Yes, we got cloud adoption.

Yes, it was much slower than the adoption that we'd seen in the United States, and we helped at Dell. We helped build those public cloud platforms. We still continue to supply and design some of the computer architecture in there.

But the point I'm making is this. When you look around what we have to deal with in Europe, we don't have 146 countries, but we do have 146 iterations that we need to consider from a regulation perspective. And then saying, oh, so there's all of that, and then there's what's going on with the EU AI Act.

The fight between the European Commission and what's going on with the US. And then on the European side, there's us as Europeans looking to Ursula von der Leyen and her teams where they're stealth against- Sovereignty, yeah ... yeah, sovereignty, and they're stealth against no big US tech companies aren't coming in and ruling the roost in this region.

We've seen that openly with texts on X to Elon Musk when he bought X. Very interesting to look at. But the point I'm making is, there's also been a genuflection of sorts back to the US, because back to that thing, that dollar thing as well, there's so much investment down into this technology.

So here at SUSECON, I think what I've been... discussing certainly yesterday, I was part of the Sovereign Summit that they'd set up here. A huge, huge focus on how do we build true sovereign capability in your- I like the word you used, capability, because it's not just, frankly, it's not a sovereign data center alone.

No, it's not It's sovereign AI, it's sovereign infrastructure from the chip design and manufacture to the networking gear, to the electricity, to the cooling. It's the entire stack, and not just of compute. So last week I was at the European Energy Summit in Bilbao.

Mm-hmm. I did three talks and two panels there. And really, a big focus on water.

There's a funny story, I think you'll like this. I was being mic'd up like I was with the guys earlier on for this. Just before my first talk, I was going on to talk about water and directed chip and direct cooling in data centers and so forth, to your point.

I had a small glass of water, I drink a lot of water, a small glass of water with about a quarter full. And I went to put it down so the guy could mic me up, and I went and it was near his laptop and he got- A little nervous ... pretty animated.

Yeah. " I get up onto the stage, and I'm on the stage for 45 minutes to talk about water in a data center. So my first address is- Without water ...

Adrian, who just put on my mics, got very animated about me taking a couple of mils of water and bringing it over close to one piece of silicon. Imagine bringing- Hundreds of- ... tens of thousands of tons of water into- Hundreds of thousands of silicon chips ...

of silicon chips- Yeah ... in a data center. So it's not sustainable.

I believe that "We need it right now," quote unquote, we need it as an industry because you're not going to be able to power the infrastructure to drive the models that we want to drive at scale. And people say it's the AI that's causing the energy crisis that we're having. It's not.

It's the compute. So there's a lot of change coming from a futurist perspective. What are the models going to look like in five, 10 years' time?

What's the infrastructure underneath the model it's going to look like in five to 10 years' time? How do we handle energy? Yeah What's going on in the battery space and so on and so forth, modernized batteries and so on.

And then the whole thing back, it's coming full circle around, okay, what is sovereignty? So right down into the data center, yes, we need to understand what needs to be sovereign from a data center perspective, an ISO point of view. And then we're going into domains, we're into finance, PSD2, PCI, all of those things on the healthcare side, HIPAA, and then we're going up into the regional policies and so on and so forth.

And a beautiful word that you used, phrase, guardrails. Right. Guardrails are great.

They can give us a direction where we can actually pull an AI instance and do inferencing, and do it safely enough for many instances, but not for all instances. Sometimes guardrails are good, but oftentimes they're not good enough. They are not at the fidelity that they need to be in order to deploy.

They're low common denominator. Lowest common denominator. Absolutely.

So there's so much in that space. So last week, Bilbao, this week, here with you talking about all things sovereign. I fly out tonight to Helsinki, where I'm partaking in a talk around robotics.

So this is where we're- Which is right behind us here. You know what I mean? I know.

So this is all... You've got energy over here, you've got infrastructure and sovereignty here, and now I'm going out to talk about how we take AI and drop it into- It's a physical ... a robot.

Yeah And have agentic functions operate and inter-operate. Autonomously. Autonomously.

There you go. So very, very wide field. And you know- And regulation ...

not to be a henny penny. Henny penny. But the fact of the matter, government has never been known to act quickly.

No. Right? And I went to law school 50, 40 years ago, 45 years ago.

One of the things a professor told me is, almost inherently, it takes us three to five years to catch up from a law perspective, from a regulation perspective, to technology. So think about where we are on this AI journey. Hmm.

We're barely three years into it, so we're just starting to see the beginning of regulation here. Absolutely. It's moving so much faster, though.

I don't think that three to five years is true anymore. I think it's probably closer to five to seven years. Because, and through no fault, but if I start something today and I think about it for six months or nine months, you see what's going on.

Nine months from now, who knows where this is, right? Exactly. So I just feel like it's a moving target and we can't use the kind of response that we've done in the past where we're going to put a commission, we're going to study it.

By the time all that gets done, it's obsolete. It's moved from here to there. And I do, I worry.

Robotics. You mentioned quantum. Quantum's coming quick, right?

"It's not that far. We're in 2026. Yeah, and I think if you look at RSA, I think we discussed this last year, right?

Yeah. RSA probably theoretically is already broken, so this one, two, three years ago- Yeah ... industry is saying 2036, 2037, this is when we see cryptography, RSA getting broken.

I was always averse to that- Getting, yeah ... in the sense that, no, peer back 10 years because it's around 2026 and I believe that. But you're right.

It's a fast pace, fast-moving target, and all the more reason why regulation and governance and compliance and controls need to be wrapped around this thing. We talked about failure a little bit earlier on. If you look at social media, from a governance perspective, failure doesn't get any more dramatic than this, right?

No. So we have polarization of- Stakes are high ... teenagers now.

These are people who have lived in a world where they've never been without this technology. So it's all that they know, and they're relying on people like us to say, well, there was a time when there was governance around- It's scary, yeah ... you couldn't walk into an Irish bar and say what you wanted without getting arrested.

Yet you can get up onto a social platform and say- And say anything ... exactly what you want. Yeah, I agree with you.

So there's a lack of regulation in around this. And now when we kind of punch forward and we're looking at what artificial intelligent agents do, they autonomously go out and converse with each other. They go out in swarms at mass, and they can scale very quickly.

And this, for me, is the huge risk. And I share your sentiment. I do worry also.

There needs to be a level of control over that. If not, catastrophic results- It could be ... will happen.

I agree with you. There's no two ways about it. Mark, I'd like to spend the rest of the day talking, but they're giving me all kinds of sig-- I feel like I'm at one of those award ceremonies and the music's playing.

We got to... I'm going to reach out. We'll continue this conversation virtually on Techstrong TV, but- You bet ...

always a pleasure to see you. No, it's my pleasure. Have a safe trip to Helsinki.

Thank you. Thank you. Mark O' Reagan, CTO Dell Technologies for EMEA.

If you want to talk, always a pleasure speaking to him. But we're going to take a break. We've got our next guest lined up.

You're watching Techstrong TV. We're at SUSECON.