By Leo Marchandon
June 25 (Reuters) – Italy’s Domyn will release a fully open-source AI model within a year, aiming to build one of the most advanced so-called “frontier” systems available, CEO Uljan Sharka said on Thursday.
The move comes as Europe seeks alternatives to reduce reliance on foreign-hosted AI systems. Italy and Czechia have restricted remote use of DeepSeek’s models while allowing locally hosted deployments. Concerns have escalated over U.S. export controls on Anthropic’s models.
Domyn’s EUROPA consortium, established with Germany’s Fraunhofer-Gesellschaft research lab, was chosen for the task under the European Commission’s Frontier AI Grand Challenge. The project positions Domyn alongside France’s Mistral and recent entrant OVHcloud in the European AI landscape.
Domyn, formerly iGenius, was founded in Milan in 2016 and has already released a suite of specialised AI models for regulated sectors such as finance, government and heavy industry.
The open-source push comes as Chinese firms such as DeepSeek and Alibaba’s Qwen dominate the open-source landscape, while most leading U.S. models remain proprietary and accessed remotely.
Domyn’s model will have over 400 billion parameters and be trained from scratch.
A 400-billion-parameter system would rank among the largest open-source AI models to date, though size alone does not determine whether it matches the capabilities of leading frontier systems.
Sharka said the model would be fully open-source and reproducible, allowing companies and governments to run it on their own infrastructure at no cost.
OVHcloud CEO Octave Klaba told Reuters at VivaTech last week that falling costs and technical barriers were opening a “second wave” of AI model builders.
“I totally agree with that,” Sharka said, adding that the Commission’s support grants access to EuroHPC, Europe’s public supercomputing infrastructure, which Sharka described as an underappreciated strategic asset.
While U.S. companies are spending heavily on AI infrastructure, Sharka said that Europe already has what it needs through the EuroHPC network, arguing that training a frontier model requires far less computing power than serving hundreds of millions of chatbot users remotely.
Domyn plans to gather data from institutional partners. Sharka said he had meetings planned with European heads of state and expected the first data agreements with governments within weeks.
Domyn declined to disclose funding details, but said it is backed by Abu Dhabi’s G42 and investors including Eurizon Capital, Rabobank and BNY.
(Reporting by Leo Marchandon in Gdansk;Editing by Elaine Hardcastle)
Comments