So maybe worth a longer note. 1. Domyn is reportedly building a 6k Blackwell cluster in Italy. Meanwhile the call grants about 2.5% of all EuroHPC capacity (yeah it’s bizarre accounting) about 1k h100 cluster for a year, so a tiny fraction of Domyn’s capacity for next year which just… doesn’t make sense. Well capitalized companies hardly ever ask for public compute because it add up frictions, as you cannot use the in house stack (from what I know in France, the allocation of this type went mostly unused). That is if the promised cluster really exists this year and in the promised shape. It has already been postponed multiple times. 2. Domyn (formerly known as iGenius) never pretrained a large model from scratch. The 355b Colossus is a continuous pretrain of Nemotron with not much to show for it (the headline is just an MMLU score). Almost nothing published, hence why hardly anyone in EU AI research knows them. No public deployment story either. Still it does match almost too well the call requirements: an EU company, that "trained or deployed" >100b model, no need for it to be a full pretrain. 3. Financials are fuzzy as well. The 650m headline raise is part debt for cluster buildout, annual revenue seemingly running in the 10-20m range. What we have seems closer to a midsized company creatively rewriting itself to either a lab or an hyperscaler thanks to a sovereign push, and maybe the public compute was actually needed to keep it on.
I must say my first reaction when I heard of this call was: who was it made for? At least I have an answer.
So another mid vintage LLM in two years no one will use. Great.
