Key Takeaways
- Domyn will release a 400-billion-parameter open source AI model within 12 months, potentially one of the largest open models to date
- The project has European Commission backing and access to EuroHPC supercomputing infrastructure
- CEO Uljan Sharka says the model will be fully reproducible, letting governments and companies run it locally at no cost
Italian AI company Domyn plans to release a fully open source frontier model with over 400 billion parameters within a year. CEO Uljan Sharka announced the timeline Thursday, positioning the Milan-based firm as Europe's answer to Chinese open source leaders like DeepSeek and Alibaba's Qwen.
The model will be trained from scratch and designed to run on private infrastructure at no licensing cost. If Domyn delivers on the parameter count, it would rank among the largest open source AI systems ever built. Parameter count alone doesn't guarantee frontier-level performance, but it signals ambition that goes well beyond incremental improvement.
Why is Europe backing this project?
Domyn's EUROPA consortium, formed with Germany's Fraunhofer-Gesellschaft research lab, won selection under the European Commission's Frontier AI Grand Challenge. The backing grants access to EuroHPC, Europe's public supercomputing network. Sharka called EuroHPC an "underappreciated strategic asset" and argued it already provides what European firms need to train competitive models.
The timing reflects growing European anxiety about AI dependence. Italy and Czechia have already restricted remote use of DeepSeek's models while permitting locally hosted deployments. Concerns about U.S. export controls on systems like Anthropic's models have added urgency. The Commission clearly wants a homegrown alternative.
“Training a frontier model requires far less computing power than serving hundreds of millions of chatbot users remotely.”
— Uljan Sharka, CEO of Domyn
How does Domyn compare to Mistral and OVHcloud?
The project puts Domyn alongside France's Mistral and OVHcloud as serious European AI contenders. OVHcloud CEO Octave Klaba recently told Reuters that falling costs and lower technical barriers were opening a "second wave" of model builders. Sharka agreed, suggesting the infrastructure advantage U.S. companies enjoy isn't as decisive as often claimed.
Domyn, formerly known as iGenius, was founded in Milan in 2016. It has already shipped specialized AI models for regulated industries including finance, government, and heavy industry. This is not a startup promising its first product. It's a company extending an existing track record into a much more competitive arena.
Where will the training data come from?
Domyn plans to collect data from institutional partners. Sharka said he has meetings scheduled with European heads of state and expects to sign the first data agreements with governments within weeks. Government datasets could provide a differentiated training corpus, particularly for applications in public administration and regulated sectors where Domyn already operates.
The company has not disclosed funding details for the project but confirmed backing from Abu Dhabi's G42 and investors including Eurizon Capital, Rabobank, and BNY. G42's involvement is notable given its recent partnerships with major U.S. tech firms and its position as a bridge between Western and Gulf AI ecosystems.
Can a 400B model actually compete at the frontier?
Parameter count is a crude proxy for capability. DeepSeek-V3, released in late 2024, demonstrated that smaller, more efficiently trained models can match or exceed larger competitors. The question is whether Domyn's team can replicate similar training efficiency while building something genuinely reproducible.
Sharka's emphasis on reproducibility matters here. Many "open source" models release weights but not training code, data, or infrastructure details needed to recreate results. A truly reproducible model would let other organizations verify claims and build on the work. That would be unusual in the current landscape.
Logicity's Take
Domyn's timeline is aggressive. Shipping a frontier-class model in under 12 months would require everything to go right: compute access, data agreements, and a team capable of training efficiency breakthroughs. The more interesting signal is what this says about European AI policy. The Commission is betting that open source, locally hosted models can serve as a strategic hedge against both Chinese and American dependencies. Whether Domyn delivers the model or not, that bet is already being placed.
Frequently Asked Questions
What is Domyn's timeline for releasing the open source AI model?
CEO Uljan Sharka said the model will be released within one year, with the announcement made in July 2025.
How many parameters will Domyn's AI model have?
The model will have over 400 billion parameters, which would make it one of the largest open source AI models released to date.
Who is funding Domyn's AI development?
Domyn is backed by Abu Dhabi's G42 and investors including Eurizon Capital, Rabobank, and BNY. The company has not disclosed specific funding amounts.
What is the EUROPA consortium?
EUROPA is a consortium established by Domyn and Germany's Fraunhofer-Gesellschaft research lab, selected under the European Commission's Frontier AI Grand Challenge.
Will the model be free to use?
Yes, Sharka said the model will be fully open source and reproducible, allowing companies and governments to run it on their own infrastructure at no cost.
Need Help Implementing This?
If you're evaluating open source AI models for enterprise deployment or building AI strategy around European sovereignty requirements, reach out to Logicity's advisory team for implementation guidance and vendor comparisons.
Source: Tech-Economic Times / ET
Manaal Khan
Tech & Innovation Writer
Produced with AI assistance and reviewed by the Logicity editorial team. Learn more in our Editorial Policy.
