CASE STUDY
Advisory Support for the European Commission to Establish a Fair Hardware Evaluation Criteria for AI Gigafactories under the EuroHPC Joint Undertaking
TechnoLynx delivered a benchmark methodology report and advisory guidance to support a fair, evidence-based evaluation process.
Challenge
The European Commission’ faced a high-stakes challenge with establishing hardware benchmarks for AI Gigafactories under the EuroHPC Joint Undertaking. The core issue was how to fairly and reliably compare competing AI accelerator solutions across multiple vendors in an intensely and rapidly evolving market. The Commission needed an evaluation approach that was objective, vendor-neutral, and reproducible, providing transparent and defensible grounds for comparing proposals and supporting future-proof decisions. Key concerns raised by the Commission included how to ensure legal defensibility, result validation, fairness for all bidders, and adaptability to future technology changes within the same process.
Legal defensibility
Result validation
Fairness for all bidders
Adaptability to future technology changes
TechnoLynx’s Role
TechnoLynx (a European AI R&D Consultancy, based in Budapest, known as the developers of LynxBenchAI) was engaged as an independent expert advisor on AI performance and benchmarking, to support the European Commission in defining state-of-the-art evaluation guidance for AI accelerator proposals in the context of AI Gigafactories. The engagement focused on defensibility: objective, vendor-neutral, and reproducible evaluation principles, and how vendor-reported results could be assessed in a way that remains credible as the technology evolves.
Objective
Vendor-neutral
Reproducible
Outcome
TechnoLynx delivered a benchmark methodology report and advisory guidance to support a fair, evidence-based evaluation process. The European Commission retained full discretion over its final approach and the internal evaluation choices made as part of that process. TechnoLynx’s contribution ensured that defensibility, validation, and long-term value considerations were explicitly addressed.
Defensibility
Validation
Long-term value
