June 26, 2026, (Inside AI) — Italian AI company Domyn will release a fully open-source frontier model within a year, CEO Uljan Sharka said Thursday. The model will contain over 400 billion parameters and be trained from scratch, positioning it among the largest open-source systems to date.
Domyn leads the EUROPA consortium, selected under the European Commission’s Frontier AI Grand Challenge. The project includes Germany’s Fraunhofer-Gesellschaft and aims to reduce Europe’s dependence on foreign-hosted AI, especially after Italy and Czechia restricted remote use of DeepSeek’s models.
Sharka confirmed the model will be fully reproducible and free to run on users’ own infrastructure. He said the first data agreements with European governments could be signed within weeks, following meetings with heads of state.
Europe’s Sovereign AI Infrastructure Advantage
Sharka argued that Europe’s public supercomputing network, EuroHPC, is an underappreciated strategic asset. He noted that training a frontier model requires far less compute than serving millions of chatbot users, a task dominated by U.S. hyperscalers.
OVHcloud CEO Octave Klaba recently described a “second wave” of AI builders enabled by falling costs. Sharka concurred, adding that Commission support provides access to EuroHPC resources, bypassing the need for massive private infrastructure investments.
Domyn, formerly iGenius, was founded in Milan in 2016. It already offers specialized AI models for regulated sectors like finance and government. The new model will be fully open-source, unlike most leading U.S. models that remain proprietary and accessed remotely.
Sharka said the model’s 400 billion parameters would place it among the largest open-source AI systems. However, size alone does not guarantee frontier capabilities. The effort joins European players like France’s Mistral and OVHcloud in challenging U.S. and Chinese dominance.
Chinese firms DeepSeek and Alibaba’s Qwen currently lead open-source AI, while U.S. export controls on Anthropic’s models have heightened European concerns over strategic autonomy.
Funding and Strategic Partnerships
Domyn declined to disclose funding details but confirmed backing from Abu Dhabi’s G42 and investors including Eurizon Capital, Rabobank, and BNY. The company plans to gather training data from institutional partners, with initial government agreements expected soon.
Sharka’s announcement underscores a broader European push for sovereign AI capabilities. The EuroHPC network, often overlooked, could become a cornerstone for training large-scale models without relying on foreign cloud providers.
The model’s open-source nature may attract enterprises and governments wary of vendor lock-in. Yet, the true test will be whether a 400-billion-parameter model trained from scratch can match the performance of established frontier systems from labs like OpenAI and Anthropic.
