Bridging the AI Production Chasm: SUSE’s Rhys Oxenham on the New AI Factory

Hi, everyone. We're back here at SUSECON, continuing our coverage of day two, Wednesday. It's in the afternoon.

Our next guest, Rhys Oxenham- Perfect ... is with SUSE. Rhys, welcome.

Thank you so much for having me. Thank you for coming on. Rhys, let's talk about you.

What do you do? Yeah. How'd you get here?

Yeah. So, today I'm the general manager of SUSE's AI business. It's actually a role I've only been doing for the last, I think, just shy of two months.

Well, a lot of AI roles that they're advertising for are looking for 10 years of experience in AI. 10 years? Goodness.

But, yeah. It's okay. No, no.

So I've been with SUSE for around about three years now. Okay. So I ran a lot of the telco and edge engineering teams- Mm ...

prior to my time at the company. So, when you look at AI, it's also an infrastructure play. Yes, it is.

And so I think my background there really helps- Helps out ... with this new position. Exactly.

Right. I don't think anyone in AI has any more experience in their AI roles. Sure.

Maybe. There's a lot of AI roles out there now, right? Yeah, exactly.

But, with all of this ambiguity around AI and AI roles, it's because underlying it, we're still kind of figuring it out. I think that's right. And it can mean a lot of different things.

So beyond the title- Mm ... what is it you're really doing? So I think what we're really trying to do is actually solve those problems for customers, right?

I don't know if you saw my keynote earlier on this morning, but we're particularly talking about what we call the production gap or the production chasm. Enterprises are under a huge amount of pressure to actually do something with AI, right? Look at the board level directives, accelerate through AI, do more with what you have.

And I think a lot of enterprises don't really know how to kind of deliver on those mandates. How do they actually accelerate the business? How do they actually reach those kind of tangible business outcomes and a return on investment?

These aren't easy things for these guys to answer really. No. They're certainly not easy.

Also, there's a nomenclature problem, right? Mm. One of the terms I hear bandied about a lot, Rhys, is the AI factory.

Yeah. Now, back in the US, everybody's talking about data centers- Oh, yeah ... as long as you don't build it in my backyard.

Sure, yeah. Right? Because they make noise, they take my water, and they raise my electric bill.

Yeah, exactly. But they're not even data centers, they're factories. They're AI factories.

Yeah. And some of them are the size of small towns. Mm-hmm.

It's crazy in scope. But not every AI factory is this behemoth. No, it's really not.

So when it comes to what we're doing with SUSE AI Factory, this is one of our announcements, especially with NVIDIA, right? SUSE AI Factory with NVIDIA. We think of the factory terminology more in line with the assembly line.

Mm. So we're not talking about physical factory or- Right ... physically building infrastructure.

Many of our customers, Switch included, they were here, they are actually building AI factories. Yes. They're building these enormous data centers to really power the AI boom.

Yes. But when it comes to our software and what we're doing, it is about providing that assembly line. How do organizations build?

How do they assemble them? How do they deploy workloads that run on the infrastructure? And at the end of the day, it's those workloads or those agents that can actually help with some of the autonomous actions within the enterprise.

They need a solid base, a solid framework to run. So our AI factory is really about how customers onboard and manage AI workloads and providing that assembly line to build them, as opposed to SUSE's not getting into the data center game. Right.

We're not going to be building these factories. Can you say that definitively? Who knows what the future is?

Never say never. Right? Oh, no.

As far as I'm aware, SUSE has no plans to build any data centers. As part of this whole sovereignty thing, someone's going to have to build them. Sure.

Absolutely. Absolutely right. And Switch, a great partner of ours, they're certainly leading the way with that, and we're incredibly excited- Good ...

with what they're doing. Absolutely. But these AI factories, it's interesting.

I was talking to one of the SUSE partners- Mm-hmm ... Traffic Labs- Yeah ... right before I interviewed you, and we were talking about sort of the integrated stack.

Mm-hmm. There was a time... Actually, I was also speaking to the CTO of Fujitsu- Udo Witz?

yes. Yeah, Udo. Him and I are of an age, let's put it like that.

And there was- None of us are getting any younger. No, I know. But him and I are of an age.

We were laughing. There was a time where we'd go out and buy servers. Mm-hmm.

Real servers. Yeah. Physical servers.

With a tool belt, we would- Yeah, rack mount ... rack mount those servers. Yeah.

Get the rails in, yeah. And that was a server. That was infrastructure.

Mm-hmm. Yeah, sure. Today, when we talk about servers and infrastructure, it's not that rack mounted server, and it's an integrated- Oh, you mean from like a density perspective?

Not only density, it's more than hardware. Sure. Yeah, okay.

When we talk about our infrastructure, software is part of that infrastructure. Absolutely. And it's not, frankly...

There was a period when we said, "Okay, I need a server with an OS," and we put SUSE Linux on there, and that was infrastructure. Yeah, sure. Today, we'd laugh at that because, well, what are you doing for hypervisor?

What are you doing for containers? There's many more layers in there, that's for sure. We've built the software stack, that software infrastructure, but now it's an integrated hardware, software package.

That's right. And to me, that's where the AI factory has to be. How can we- Correct ...

where do we put our AI onto that stack? Yeah, you're entirely right. So as I was explaining to you, the AI factory is really about the kind of the assembling of the applications- Yeah ...

to run on top. But to your point, we can't actually forget about all the layers that comprise that underlying infrastructure. Yep.

So SUSE AI, what we're really doing is we're helping by providing kind of opinionated and prescribed configurations for customers to get started. But what we're doing every time we release that is kind of defining and providing the guarantees that every underlying layer has been proven, tested, secured, hardened to really make it, yes, there may be more layers in an infrastructure platform today, but consuming it doesn't have to be a burden. Right.

Operationalizing it, it shouldn't be your concern. We have this covered. This is our expertise.

We've been doing this for 34 years. Yes, the world has changed, and those components have changed- Yes, they have ... but it's still the same underneath.

Okay. Now, from customer's point of view- Mm ... look, what you just said, as a customer, that's what I want to hear.

Sure. Perfect. But rubber meets the road, reality sets in.

What are you seeing? We hear the MIT study said 95% of AI didn't return ROI or whatever. Yep.

I don't believe that. But what are you hearing from customers? Real-life success, failures?

Yeah. Something in between? Yeah, sure.

I'm sure it runs the gamut. No, it absolutely does. We speak to a lot of organizations that say, look, they've seen some initial success in doing some smaller pilots.

I think it's very quick to show that there is some value in doing certain capabilities through AI. But then there's a problem. How do I do that at massive scale?

How do I make that available to all of my enterprise? How do I make it available in a state where it is consumable, that it is secure, that it is governed, that I can look after it, and I can upgrade it as time goes on? These are the challenges that we are having to embrace now.

And like with every technology, we go through the hype cycle. We see a lot of interest, a lot of pressures we were talking about earlier to really do something with AI. " AI has so much promise, but actually doing it in the real world, I think that's a challenge.

Over the last few years, I think we've been in that world where a lot of pilots are going on, a lot of testing, a lot of evaluation. But I think now is really where we're starting to see real projects, real implementations- Yeah ... real value being derived.

I think a lot of that has to do with technology maturity- Yeah ... availability, stability, vendors that are offering support for these technologies. SUSE are, of course, included.

The ecosystem that brings this alive, our friends at Nvidia and other partners, of course, bringing their capabilities to the front as well. I think now we're starting to see these implementations really start to take shape. But again, who knows what happens in the world of AI and what will happen over the next few years, but it's certainly an exciting place to be.

Absolutely. Next few years, I'd like to know what happens in the next few months. My goodness me, yes.

Absolutely. Every day, well, we're on our own agentic- Yeah ... exploration at Techstrong- Yeah ...

internally. Drinking our own champagne, eating our own dog food- ... I'm not quite sure.

Yeah. But every day brings a new adventure. It absolutely does.

And new possibilities, too. Exactly right. I was having a conversation with a member of the press earlier on, and we were talking about just the rate of acceleration thanks to AI.

I mean, I remember multiple different kind of evolutions in the tech world. Yeah. So there's going Unix to Linux, virtualization, cloud, containerization.

Yes, these happen, but they, at least for me- Not at this rate ... not at this rate. Yeah.

It certainly felt like they were taking years. So I was a dot-com baby, right? Yeah.

I started my first company, sold it, then took another company public during dot-com. Yeah. And that was quite the revolution.

Absolutely. And it was a fast rate because it happened, the internet went commercial, Netscape browser- Yeah ... in '96.

Sure. By '98, we were rocking. By 2000, by 2001, the bubble burst, right?

Right. But nowadays, it just seems like the rate of change is- And the change is- ... incredible ...

so radical. Yeah, absolutely. So radical.

Absolutely. No doubt about it. But it's also the scale of the application.

And what I mean by that is, in cloud, yes, everyone's kind of using cloud. They don't necessarily know it, but with AI, everyone knows about AI. Well, there was the IBM commercial where the grandma said her- Yeah ...

pictures were stored up in the cloud, and she would point up. Yeah. But she didn't really know what cloud was.

Right. Exactly. But here's an interesting thing.

At best, cloud only represented about 20 to 25% of people's infrastructure. Yeah. The majority of infrastructure was still in private data centers.

Yeah. Sure. Where here, you've got to ask yourself, within the next year, what isn't AI going to touch?

Absolutely right. Yeah. We're obviously at a tech conference.

We're an enterprise company. Yeah. We have to talk about what is going on in the world of enterprise AI, but you just have to look at the connection between AI and the real world.

Just look at what's going on with physical AI, robotics, self-driving or autonomous cars. AI is touching everything we do on a daily basis. Absolutely.

And that, I think is-It's interesting, for sure. But it's going to be disruptive, and I think that what we're starting to... Because we obviously report on this a lot.

Yeah. So now we're starting to see a bit of a backlash, and that's the typical hype cycle, right? People are saying, "Well, it's not as good as we thought it was.

" It's this, it's that. When you look at where we are, it's pretty remarkable right now, and it's getting better every day. I completely agree.

I think we are surpassing expectations at this stage. Still. Absolutely.

And we are very much in the early days. No doubt about it. " Exactly.

That's a great way of... I'm going to steal that if you don't mind. By all means.

By all means. Yeah. So we're here now, day two.

You did your piece up on the keynotes- Yeah ... stage. One of the big things...

Well, I've been writing some articles in addition to doing the videos. Right? The ecosystem- Mm-hmm ...

that SUSE has put together here- Yeah ... is pretty remarkable. You mentioned the Nvidia relationship.

Yeah. Who are some of the other partners? The data center provider, other partners involved in the AI Factory project.

So right now, we're just making an initial announcement of the SUSE- Okay ... AI Factory. SUSE AI Factory with Nvidia.

First partner, their tightly integrated solution. Not a bad one to start with. Yeah.

Sure. Exactly. Fujitsu is our hardware partner.

Yes. Good for them. They announced today with us.

And of course, SUSE is a big advocate of choice. We know that our customers are going to want flexibility in the underlying infrastructure, kind of- What about the underlying frontier models? Yeah.

Absolutely. And there are so many different models out there for lots of different capabilities. You look at what we're doing in the edge space, or you look at computer vision, for example.

There are optimized models that are specifically built for that, and our customers want choice. Right. They want the ability to consume models, have flexibility in them, and even fine-tune models.

You have the concept of retrieval augmented generation. Right. Awesome technology, but for many use cases, simply fine-tune a model with all of that data built in.

So there's lots of flexibility there, and I think a platform like SUSE AI Factory enables that flexibility. So I think what you just mentioned right there, though, about the models and specialized models, that's the next wave here. Yeah.

Right? So many people now, they know Claude, obviously, they know OpenAI, Gemini, these frontier models. But I think the next wave is literally dozens, hundreds, thousands of specialized models.

You go to Hugging Face now, you get all- Sure. You do ... like a million.

I think it's also the proliferation of technology and making it accessible to people, be that in the tools- Yep ... but also the specialized hardware. You can build a model on a consumer-grade GPU today.

It'll take time. You know what? I- But it's possible to do it, right?

I think the new CEO coming in at Apple is part of that wave. Possibly, yeah. This is a guy who comes from the hardware world.

Yeah. Ran SVP of hardware there, and I think the fact of the matter is Apple had a great edge piece of hardware to begin with, as it turns out. Sure.

But now we'll see where that goes. Exactly. Rees, for people who want to stay up to date on this- Yeah ...

AI Factory product offering from SUSE, what's their best bet? So right now, obviously, we're in the early days. com.

On the homepage, there's the SUSE AI Factory with Nvidia. It links to a blog post and a whole bunch of other materials. Go check it out.

We're going to be releasing a lot more materials in the coming weeks. Blog posts, demo videos, integrations we're doing. So we're super excited about it.

Okay. Thank you for coming on Techstrong TV. Thank you so much for your time.

Hey, check out SUSE AI Factory. It's an interesting piece, too. We didn't get into it as much here, but you've got SUSE virtualization- Yeah ...

SUSE Rancher, now SUSE AI Factory. Yeah. That's a stack.

Correct. Absolutely. Silicon to solution.

Yeah. We're going to take a break. We got more SUSE Con coverage coming at you.

Stay tuned.