KYIV, Ukraine (US News): Ukraine is developing its own sovereign large language model using Google’s open-weight Gemma framework, a move aimed at establishing full national control over a critical artificial intelligence system that will serve both defense needs and civilian services across the country.
The Ministry of Digital Transformation and Kyivstar, Ukraine’s largest telecommunications provider, unveiled the initiative this week. The project will initially rely on Google-supplied cloud computing resources for the resource-intensive training phase before shifting entirely to data centers inside Ukraine, ensuring the government retains complete ownership of the finished model.
Oleksandr Bornyakov, Deputy Prime Minister for Digital Transformation, emphasized that the effort is part of a deliberate strategy to avoid multimillion-dollar licensing fees for foreign proprietary systems and to eliminate dependence on closed platforms such as ChatGPT from OpenAI.
During the selection process, officials evaluated several leading open-source alternatives, including Meta’s Llama, France’s Mistral, and Chinese models DeepSeek and Qwen, before choosing Gemma for its strong multilingual capabilities and favorable licensing terms, according to Kyivstar executives.
A key driver behind the project is the need to overcome persistent accuracy problems that plague existing AI tools when processing the everyday language spoken by millions of Ukrainians. Many citizens fluidly mix Ukrainian and Russian, along with regional dialects and minority languages such as Crimean Tatar, resulting in frequent mistranslations, legal errors, and unreliable outputs in current systems.
Trump Issues Executive Order on AI: Key Impacts for Illinois Businesses and Workers
To address these challenges, developers are assembling an extensive Ukrainian-language dataset drawn from more than 90 public institutions, including court records, educational materials, parliamentary transcripts, and wartime archives. Four specialized advisory councils with binding oversight authority have been formed to guide technical development, legal compliance, cultural sensitivity, and linguistic accuracy.
Because of ongoing Russian attacks on Ukraine’s energy infrastructure, the most compute-heavy training will occur on secure GPUs located abroad. Once complete, the model will be transferred to fortified domestic facilities. Officials expect the system to face immediate cyberattacks upon deployment and are building protections against threats such as prompt injection.
Kyivstar has already deployed thousands of backup generators and hardened network infrastructure to maintain service continuity during blackouts and intensified bombardment.
Ukraine has become one of the world’s most active users of artificial intelligence in combat, employing systems for battlefield reconnaissance, autonomous drone operations, and damage assessment, including software provided by U.S. firm Palantir. The new national model is expected to integrate with and gradually supplant some of these foreign tools.
The completed AI will first power upgraded digital government services and Kyivstar customer platforms before being opened to Ukrainian businesses and developers.
Government officials say the initiative demonstrates that mid-sized nations can harness open-source technology to achieve strategic independence in artificial intelligence at a time when global superpowers are tightening control over foundational models.
