Domyn chief says Europe needs more AI investment despite technical expertise

The comments came as Domyn, founded in Milan in 2016 as iGenius and renamed last year, is building Colosseum, a supercomputer in southern Italy, and leading a European consortium developing an open-source model. Domyn has raised more than $1bn, mostly in debt, according to Sifted's reporting cited by The Next Web.

Funding and the machine

Domyn took roughly $1.1bn in total, split about 10% equity and 90% debt, according to that reporting. That structure is what you raise when you are buying hardware rather than hiring researchers, according to The Next Web. Mistral took the same route on a smaller scale, raising $830m from seven banks to build its own data centre outside Paris. European AI companies have found that lenders will finance a building full of GPUs on terms no venture fund would offer for a model, the report said.

The Next Web also cited scepticism about the wider European push: renting capacity through a GPU-as-a-service layer reinforces the illusion of sovereignty rather than delivering it, it said. That is one reason Domyn's decision to own the metal is more interesting than the headline number, according to the report.

The investor list, the valuation, the identity of the lenders and the terms attached to the debt are all undisclosed. Domyn has not issued an announcement about the round. The Next Web reported its ambition to invest $10bn over three years and reach €1bn of revenue in the same window, alongside Sharka's April comments to Semafor. Trade reporting puts its current base in the tens of millions of annual recurring revenue.

Sharka was candid about the scale of the framing. "$1 trillion is the new unicorn right now," he told Semafor. "If you don't talk about $1 trillion, you are out of touch."

Domyn last raised in 2024, taking about €650m of project financing at a reported €1.7bn valuation after a €70m Series A. The new round is more than that entire history, and it has been raised against a machine rather than a market, The Next Web said.

Colosseum

Colosseum is being assembled in southern Italy with Nvidia and Vertiv. It runs 80 Nvidia GB200 NVL72 systems, close to 6,000 Grace Blackwell chips, draws about 7MW and is rated at 115 exaflops, which Domyn describes as purpose-built for models above a trillion parameters.

Domyn sells to banks, insurers, pharmaceutical companies, defence contractors and governments, on the argument that those buyers want to own the model rather than rent access to somebody else's. Customers get the codebase and can pre-train on their own data, and the models are distributed through Microsoft's Foundry platform. Domyn has shipped two models: a small one aimed at general enterprise work and a larger mission-critical system it benchmarked against GPT-4o and Claude 3.5 Sonnet. Those are the company's own comparisons and have not been independently reproduced.

Owning the machine also means carrying it. A supercomputer depreciates on a schedule set in Santa Clara, and the useful life of a Blackwell rack is shorter than the term of most infrastructure debt, The Next Web said. In a separate reference, the report said Domyn has been working with Nvidia and the Abu Dhabi group G42 on the Italian supercomputer.

EUROPA

The European Commission selected the Domyn-led EUROPA consortium in June 2026 as winner of its Frontier AI Grand Challenge, EU Today reported. The competition, launched in February, sought a European general-purpose model with more than 400 billion parameters. EUROPA's planned model is expected to hold about 500 billion and to support all 24 official EU languages.

The consortium brings together Domyn and Germany's Fraunhofer research organisation, including the Fraunhofer IAIS and IIS institutes. Its work will rely on the EuroHPC network of European supercomputers. Under the competition arrangements, the winning consortium can obtain access to as much as 2.5% of EuroHPC's total computing capacity for one year. The final system is intended for release under an open-source licence.

EU Today's account said parameter count alone does not show that a model is competitive. Performance will depend on training data, architecture, computational efficiency, evaluation results and reliability across tasks, according to that account.

European backdrop

The EU has launched a call for tenders for up to seven AI Gigafactories. The Next Web described the effort as a €30bn programme. The Commission's page says the initiative is supported by up to €10bn in EU and national funding and is expected to draw at least €20bn in private investment. AI Gigafactories are large-scale facilities for training models with trillions of parameters, and the programme is intended to bring together more than 100,000 advanced AI processors.

Currently 19 AI Factories and 13 Antennas are being set up. At least 9 new AI-optimised supercomputers will be procured, which the Commission says will more than triple current EuroHPC AI computing capacity.

Italy has been moving separately, with a €1bn national AI fund and a set of penalties for misuse legislated alongside it. The Next Web described Domyn as the most visible private beneficiary of that positioning.