Domyn plans open-source frontier AI model launch within a year
The Italian AI company formerly known as iGenius is building a 400-billion-parameter model backed by NVIDIA hardware and a growing roster of heavyweight partners
An Italian AI company most people have never heard of just announced plans to release what would be one of the largest open-source AI models ever built. Domyn, the Milan-based firm formerly known as iGenius, says it will drop a fully open-source, 400-billion-parameter model within a year.
What Domyn is actually building
The company’s focus is not consumer chatbots or AI-generated art. Domyn is targeting regulated industries, specifically financial services, government, and defense, where data governance and compliance are not optional extras but table stakes.
To power this ambition, Domyn is constructing what it calls AI gigafactories. The first facility, located in northern Italy, is expected to be operational by early 2026. It will run on thousands of NVIDIA Grace Blackwell GPUs.
The company has lined up strategic partnerships with NVIDIA, G42, Microsoft, and Cisco. Domyn is also reportedly aiming to secure a funding round of around 1 billion euros to expand operations.
The European sovereign AI play
The company’s rebrand from iGenius to Domyn appears designed to signal a shift toward a more ambitious, infrastructure-level identity. In 2025, former BlackRock Managing Director Stefano Pasquali was brought on to lead a newly formed financial services division.
The open-source angle is strategically interesting too. By making the 400-billion-parameter model freely available, Domyn could accelerate adoption across European institutions that might be wary of locking themselves into proprietary American AI platforms.
What this means for investors
Domyn has no reported connections to cryptocurrency, blockchain, or digital assets. This is a pure AI infrastructure play.
The partnerships are worth watching closely. NVIDIA’s involvement, Microsoft’s presence, G42’s Gulf sovereign wealth fund connections, and Cisco’s networking expertise are each instrumental in supporting the deployment of a 400-billion-parameter model.
The risk side of the ledger is equally important. Training frontier-scale models is brutally expensive. The compute costs alone for a model this size can run into hundreds of millions of dollars. If the 1-billion-euro funding round falls short, the timeline slips, or the gigafactory faces construction delays, the open-source launch could be pushed back significantly.