Road to AI Episode 2: What really happens when you start building enterprise AI?

What really happens when you start building enterprise AI

Global, Nov 10, 2026

Author: Ben Walker, Marketing Director, Logicalis Group

As part of The Road to AI, Ben Walker, Marketing Director, GTM, caught up with Ulrich Vierkotter, Lead Data Center Networking Logicalis Germany and Alex Spirin, Solution Architect Logicalis Germany while the team was building the AI Pod in Cologne. A few months earlier, the project was little more than an idea. Now the infrastructure is taking shape.

They discussed what they’re building, what has already gone wrong, why AI infrastructure requires a different kind of collaboration, and what they’re learning by doing it for real.

Ben: When we last filmed you guys, none of this existed. Where are we now?

Ulrich: We’re currently working on the storage system. The servers are configured and included in the AI Pod, but storage is something we're having to work through because we haven’t used this particular system before.

With most parts of the infrastructure, we're confident. But AI introduces some additional technical requirements that you don't necessarily encounter in a normal server deployment.

GPUs are a good example. They have specific requirements around how they communicate with each other, so we need to understand those requirements and make sure the infrastructure supports them properly.

Alex: Last time, we had mounted the systems into the racks. Now we’re actually configuring everything and starting to bring it to life.

Ben: Does building AI infrastructure require different skills from a traditional infrastructure project?

Ulrich: Definitely. An AI Pod consists of several layers and you cannot have one person who knows everything.

We are infrastructure engineers, so we're focused on areas such as networking and servers. Then other specialists come in to deploy the operating system, containers and the platform. Then you have another group specialising in the AI applications themselves.

You need all of those people working together.

Alex: That's really important. We've divided the responsibilities according to who has the right expertise. Our first job is to prepare the infrastructure. Then the application developers can come in and do their part.

Ulrich: Everyone has to pull on the same rope.

Ben: The idea for this started long before the hardware arrived. Why did you want to build the lab in the first place?

Ulrich: We knew that if we wanted to get into AI properly, we needed servers with GPUs and the NVIDIA components required for things like AI inference and model training.

So we approached management and said, “Can we build a lab?”

And, of course, the first question was: “How much does it cost?”

It’s expensive.

So we had several conversations about what we could actually achieve with the lab and why we needed it.

The first reason was simple: we can learn.

We can learn how to deploy AI infrastructure ourselves.

The second was that we could use it to show customers what an AI deployment could look like within their own on-premises environment.

Ben: Why is proving that important for customers?

Ulrich: Because most customers are still at the beginning of their AI journey.

They're asking: How should I start? Should I go to the cloud? Should I create something on-premises? What infrastructure do I actually need?

And AI is always about data.

That data can be extremely sensitive, so some customers may not want their applications or data running in the cloud. They start thinking about an on-premises environment, but then another question appears:

How do I actually build it?

Our goal is to be able to show customers how they can run sensitive AI applications on their own infrastructure in a way that is performant and secure.

Alex: And once we have real applications running, we can also start comparing the options. Not only technically, but financially.

We can look at the costs of an on-premises solution and compare those with consuming AI services from the cloud.

Ben: We’re showing the real journey with this project, not just the successes. What has gone wrong so far?

Ulrich: We had some challenges getting the initial server management running, and then we ran into issues when we started building the OpenShift cluster.

We would add the servers into the environment as workers and encounter an error. We'd look through the logs, understand the fault, change something and run the discovery again.

Then we'd find the next problem.

So we went through several rounds before we really got everything running.

Alex: In the documentation, everything works on the first attempt. In reality, you need some effort to get it working.

That is part of what we're learning.

Ben: Was there ever a point where you thought: we've taken on too much here?

Ulrich: No. We were always confident that we'd get it running.

Alex: Because we have the people around us.

We have a team made up of experts from multiple parts of the business. If our own knowledge isn't enough, we can immediately ask somebody who has that expertise.

That has been really important.

Sometimes we deliberately get everyone into the same room. If one system isn't reachable, for example, instead of creating a ticket or organising another call, I can just turn to someone and ask them to check it.

Especially at the beginning, that makes everything much faster.

Ben: Can you give me an example of something you discovered by actually building the environment?

Ulrich: We found a good example involving the GPUs.

For demanding AI applications, you might need to distribute a workload across several GPUs. To do that efficiently, those GPUs need to communicate with each other, and the network has to support some very specific requirements.

We discovered that the switches we'd originally bought didn't support one of the features we needed.

That's the type of thing that can go wrong. You can buy something and then discover it doesn't completely support what you need.

Fortunately, we found a workaround and have other switches that can support it.

But these are exactly the things you learn when you build the environment for yourself.

Ben: You've both volunteered to work on this project alongside your normal customer work. Why?

Alex: Because we need to be part of this change.

There are unexpected and unknown things because the technology is so new. That's exactly why it's important for us to start working with it now, rather than waiting until later.

Ulrich: I've always been interested in new technology.

As infrastructure engineers, we're not developing the AI models ourselves. Our challenge is different: how do we create the infrastructure that supports what AI workloads need today?

That's what makes this interesting.

I've worked with technology for more than 20 years, and there have been many innovations during that time.

AI is one of the most interesting.

Learning by doing

Perhaps the biggest lesson from the conversation wasn't about GPUs, switches or OpenShift. It was the importance of experience. The team doesn't expect every decision to be right first time. They're building, testing, finding problems, bringing in different expertise and trying again. And that's precisely why the AI Pod exists.

Because before we can confidently help customers navigate their own road to AI, we need to travel it ourselves. That journey is still very much underway. 

The Road to AI: a Logicalis docuseries

Catch up with episode 1: The idea

Catch up with episode 2: The build

 

Topic

Related Insights