Chris Carreiro, Park Place Technologies: Bridging the AI readiness gap

Penny Horwood (0:05)

Welcome back to Ctrl Alt Lead, the podcast from Computing where we talk to IT leaders at the sharp edge of technology and leadership. I'm Penny Horwood.

At Computing, we know from our audience that turning AI ambition into production reality is proving quite a lot harder than many anticipated. My guest today is Chris Carreiro, CTO at Park Place Technologies, which is a global IT infrastructure services specialist, and as such has visibility across a huge cross-section of enterprise infrastructure, ranging from legacy estates through to the liquid-cooled environments, which we need to support the next generation of AI workloads. As CTO, Chris has a real insight as to why AI projects are stalling and who actually really owns AI infrastructure readiness and why cooling has rapidly become one of the most important conversations in enterprise IT. Chris, welcome to Ctrl Alt Lead.

Chris Carreiro (0:59)

Hi, thanks. Thanks for having me.

Penny Horwood (1:02)

Good to see you. Good to see you. So Chris, AI is the only talking town right now, but it is also clear that there's a big gap between AI ambition at board level and genuine progress.

Why is this happening? Huge starting question for you there.

Chris Carreiro (1:17)

Definitely. Yeah, I would say first off, I mean, it's just ambitious goals are set during the demo phase. A lot of companies get right into it and just turning something from a demo into production. There's all kinds of challenging pieces that are popping up, leveraging AI, compliance challenges, data fragmentation, all kinds of things across the business.

Penny Horwood (1:39)

Okay. Good stuff. So you've identified that goals initially may be a little bit overambitious, for example. So why is that happening, do you think?

Chris Carreiro (1:51)

I feel like, so at any new technology, a lot of people are still trying to understand exactly how to leverage it, not having clear defined utilization. You have the board pressuring you to get AI implemented, but not really understanding other business challenges. AI works really well when you have the right data, when you have data in a central place, clean information that it can leverage, make decisions and drive into there. I think that's one of the big pieces. The other is new products, not having those clear defined goals, it's really hard to track ROI. So you have different issues trying to put a number on this, on how much savings you had or the overall spend on a project.

But having that problem tracking the ROI makes things kind of difficult to quantify an application.

Penny Horwood (2:43)

It does. Is that because costs have escalated, which obviously they have in the last quarter, really, especially? Or do you think the cause is more fundamental?

Chris Carreiro (2:55)

It really, so you've got to look at the application of the AI in a business. If you're using cloud cost, let's say you're just building, you're doing a POC, you're leveraging the cloud. Yes, those costs are huge. You can have, just like it was when we shifted from on-prem infrastructure to cloud, shifting your entire workload into the cloud. That was very expensive to do something like that. Not understanding the full usage, usage pricing, all of those things. It took a while for that to settle down. I believe the same thing's happening now with AI, and that's on the cloud side. On the on-prem side, there's a lot of, these systems are extremely power hungry. It could be very expensive hardware that you're trying to roll out, and then that hardware has a lot of dependencies, such as cooling and other power infrastructure, network infrastructure, that you might not have already figured into the initial project.

Penny Horwood (3:48)

So do you think it's an infrastructure readiness issue, or do you think, you know, because you've alluded to data readiness as well, or is it, are there elements of all of that feeding into the problem?

Chris Carreiro (3:59)

I would say elements of all of that. Data is probably number one, regardless of where the infrastructure lives. If it's in the cloud, you're still going to have a data problem. When you start looking at where all of the data islands kind of reside in a company's IP or inside of their database structure, the infrastructure problem is more so for on-prem infrastructure. If you're going to, let's say you want to have data sovereignty, you want to have nothing placed on a cloud-shared AI model, you want to run your model locally, you introduce a lot more challenges there with, is the infrastructure ready? Do you have, did you budget enough accordingly? Not to mention the shortages with memory, just to try to get new servers, memory and hard drives. That's a whole other problem.