AI ambition is high, but cloud reality often tells a different story. At TechQuartier’s ÄpplerTime in Frankfurt, Dr Red Boumghar, CEO of Celestical – The European Serverless Cloud, warned that many organisations overestimate their readiness to scale artificial intelligence (AI).

In an interview with Ivett Ódor and Johanna Eickholt, Dr Boumghar explained why cloud costs so often spiral out of control, why sovereignty goes far beyond labels, and why system-level thinking is essential if AI is to move from experimentation into production. 

 

At ÄpplerTime’s panel discussion, you spoke about the tension between moving fast and taking control. Where do you think organisations most often believe they are further ahead with cloud and AI than they really are? 

Dr Red Boumghar: You have a lot of organisations that actually don’t know what the cloud is. That’s a perfect illustration of what’s happening in decision-making processes when companies move to the cloud and have to decide which one to choose. Decision-makers are often told, “You need to go to the cloud”, but they don’t really know how to make that decision because they have too little information and too little technical knowledge. What they mostly hear are the big stories from NVIDIA, AWS, Google, or others. 

That’s why we focus much more on developers. They’re the ones who will end up in those meetings and can say, “I’ve deployed this already. I know how to do it. It’s easier.” That’s what makes the difference. 

On AI, a lot of companies think they are far ahead because they use tools like Microsoft Copilot. They say, “We are doing AI.” But that’s pretty much using a model made by an external company that shapes your emails and the way you communicate. The real question is: are you sure you’re in control, or is it the other way around? 

I think we are just at the beginning of this. People are still exploring. I still see many events where we go through the basics – and that’s useful. 

“Sovereign cloud” is high on the agenda for many European organisations. In your experience, what’s one decision that often determines whether it delivers real value or just adds complexity? 

Dr Boumghar: In the political domain, sovereignty is super important. When we talked to VCs, though, last year, it was too much. Some were very honest and said, “We’ve heard that already.” For many companies, sovereignty had become mostly a marketing story. 

The strongest feedback we had was to focus on performance. That’s the best angle we have taken. If we can beat performance, then we gain independence. We can choose our hardware, our providers, and our infrastructure. That’s how we build sovereignty in reality. We don’t talk about it. We build it. 

AI is changing cloud costs faster than many teams expected. What do leaders most need to rethink about how they manage and explain cloud spending in an AI-driven world? 

Dr Boumghar: We see a lot of these stories because we’re right in the middle of it. At first, we wanted to be very cheap, and I don’t like that word. The real issue isn’t cost, it’s predictability. 

People pay €30 a month, try something, and suddenly they have a €3,000 bill after one week of usage. They weren’t expecting that. 

So, a lot of people go back to the cloud providers and say, “Whoops, I didn’t know it was going to be that high.” That’s why we’re trying to help people understand what costs are going to look like and simplify the models. 

AI is changing costs very fast, and many applications still don’t control that well. For me, the main point is simple: if you need AI to understand your cloud bill, something is wrong. It’s too complicated. 

Compliance is still widely seen as something that slows innovation down. What actually changes when companies build trust and regulation into their platforms from the start? 

Dr Boumghar: That’s actually our case. We’re trying to be compliant by design. I’ve worked in different European institutions, so I’ve had to read the GDPR and the EU AI Act, and I saw the AI Act being brought to life through open discussions. 

There’s often this perception that regulation prevents companies from doing anything. But when you actually read the AI Act, you see that it mainly applies to high-risk AI systems, those that impact everyone, such as health or other critical areas. In those cases, it’s normal to have stricter rules. 

If you’re building an internal AI system for your own company, the requirements are much lighter. Today, the key questions are simple: what data are you using, and do you have the right to use it? That’s why people really need to read these texts. 

They’re not all easy to read. DORA is probably the hardest, because it’s linked to many other documents. But apart from that, the others are pretty manageable. 

You spoke about AI at Tech Show Frankfurt back in 2019. As we look ahead to this year’s edition, what’s the one capability you believe will separate organisations that successfully scale AI from those that remain stuck in experimentation? 

Dr Boumghar: The organisations that succeed are the ones that understand MLOps and AI operations. You need people who know how to train and operate models, but also people who understand how the entire system works and how users interact with it. 

If you don’t understand your system end-to-end, you won’t know how to reduce the barriers that allow AI to scale. That applies to both sides: the back end – how systems scale technically – and the front end – how usage and adoption increase. 

It really comes down to real familiarity with the use cases. And a strong MLOps team. 

For a broader perspective, read the full panel recap from our TechQuartier discussion with AI and cloud leaders across Europe here.

 

Dr Red Boumghar is speaking at Tech Show Frankfurt, 6-7 May.