When choosing the wrong AI setup becomes a costly mistake 

What if launching powerful AI models became simple and affordable? Researchers are exploring practical ways to help companies navigate AI deployment with more clarity and fewer costly mistakes. 

Research scientist Dr. Mats Brorsson is working on a technology that could remove one of the biggest hidden barriers to modern adoption of artificial intelligence (AI). His new venture, Infratailors, aims to simplify access to AI by helping companies select the right AI models and the right hardware without specialised expertise or expensive trial-and-error. 

1. Why the wrong AI setup becomes so expensive

For many companies, the ambition to use AI quickly runs into a practical barrier: even the smartest model will fail if it sits on the wrong infrastructure. The result can be slow performance, soaring costs, or a project that never reaches production. Mats Brorsson sees this problem often. Companies want chatbots, predictive tools, or generative AI features, but few have the in-house knowledge to select the right model or hardware. 

Even major cloud providers struggle with this. Brorsson explains that solution architects, who design complex digital systems, may often spend two or three days finding the right configuration before a single line of code runs. Without specialists, companies often pick configurations that don’t fit their needs. One of Brorsson’s demos shows how costly that can be, with the wrong hardware adding more than €30,000 a year.  

The result? Wasted time, wasted money, more energy consumption and, in many cases, AI projects that never move past the idea stage. 

2. Designing a practical tool for choosing the right AI setup

At the University of Luxembourg’s Interdisciplinary Centre for Security, Reliability and Trust (SnT), Brorsson and his team are building Infratailors, a platform that guides cloud providers and end-users to the optimal setup for their AI models. The tool combines benchmarking data, prediction models, and domain expertise. In simple terms, it learns how different AI workloads behave on different types of hardware and predicts which configuration will deliver the required performance at the lowest cost. 

‟ In simple terms, it learns how different AI workloads behave on different types of hardware and predicts which configuration will deliver the required performance at the lowest cost. ”

Dr. Mats Brorsson

Founder of Infratailors

View profile

The idea grew from PhD research on predicting AI behaviour on modern hardware, which revealed how unpredictable model performance can be without precise measurement. That insight eventually became a startup concept pitched in June 2024 to the SnT Venture Programme, where it secured FNR funding and hands-on entrepreneurial coaching. The team later brought on an entrepreneur-in-residence, Romit Choudhury, giving the venture the momentum it needed to turn academic research into a scalable product. 

Infratailors’ approach is well-rounded. Instead of pushing generic recommendations, the platform tailors suggestions to each client’s performance needs and budget. “We take the strengths and needs into account coupled with budget restrictions. I have never seen that from any competitor,” Brorsson says. The platform already works with catalogues from around ten cloud providers, including both global giants and smaller European providers.  

Looking ahead, Brorsson imagines an even simpler workflow. A company would describe what it wants, like letting customers ask questions in natural language, state its performance expectations and budget. With one click, Infratailors would generate and deploy a model that meets those constraints. OR would recommend a model, choose the right hardware, and deploy it automatically. As he puts it, “We democratise the ability to launch AI models in any company’s products.” 

3. Why this moment matters for AI deployment

The timing for Infratailors couldn’t be better. As generative AI becomes part of everyday digital products, demand for tailored deployments is soaring. Companies want chatbots, recommendation engines, and automated assistants, but all these features require careful tuning and significant computing power. With AI workloads rising and Nvidia reporting record valuations in the trillions of dollars, the challenge is no longer convincing companies to adopt AI but helping them afford to do it well.” 

Infratailors’ value stretches across the AI ecosystem. For businesses, it lowers the barrier to entry by removing guesswork and controlling costs. For cloud providers, it automates a process that currently demands hours of manual consulting.

As he moves into entrepreneurship, Brorsson has learned how important it is to stay anchored in real problems. Founders often drift between confidence and doubt, he says, but steady conversations with industry partners kept proving the idea’s value. Those discussions helped the team avoid early assumptions that often derail young ventures. 

With a growing engineering team and strong entrepreneurial support, Infratailors is positioning itself as a key player in Europe’s AI infrastructure landscape.