Czechia’s Alpha Industries demonstrates LUMI’s power and sustainability in their initial test

A new story from the European AI ecosystem has been published by Jan Tyl, one of the Czechia’s prominent AI popularisers. Tyl is the founder and CEO of Alpha Industries, an Czech technology company established in 2018 that develops AI products, educational software that aims to expand human thinking, not replacing it.

Writing about his firsthand experience on the LUMI supercomputer—accessed via a 5,000 GPU-hour grant through the IT4Innovations National Supercomputing Center in Ostrava—Tyl shared the results of “Phase 0” of his latest AI evaluation project.

-It is almost poetic: a supercomputer calculates climate models, scientific simulations, or artificial intelligence, and the heat from these computations is then returned to the town. Not as waste, but as useful energy, says Jan Tyl, Founder and CEO of Alpha Industries.  

Key highlights

The Power of Bulk Processing: Tyl’s testing highlighted that running individual queries or single prompts on a supercomputer underutilises its massive infrastructure. The real economic and performance breakthrough occurs when a client switches to large-scale batch processing—processing thousands of queries simultaneously. This unlocks the true capability of the system.  

Smart Testing and Preventing AI “Hallucinations”: When tested with tricky questions, the smaller AI model sometimes made up false information—a common AI issue known as “hallucination”. The goal of Tyl’s project on LUMI is to map out exactly when small, cost-effective AI models are enough, and when it is the right time choose a larger, more expensive model.  

A Model for Green Computing: Beyond the technical benchmarks, Tyl praised LUMI’s environmental footprint. He highlighted how the supercomputer runs entirely on renewable energy, and how its waste heat is directly redirected into the local heating network to warm the homes in the Kajaani area. It proves to emerging startups that cutting-edge AI development can be both high-performing and environmentally sustainable.

The original comprehensive article can be found here.