Ukraine says no to OpenAI. Builds sovereign AI on Google infrastructure

Ukraine is focusing on digital sovereignty and, in cooperation with Google, is beginning to build its own language model, abandoning commercial solutions such as ChatGPT in favor of the open Gemma architecture. This strategic move is intended to give Kyiv full control over key military and administrative systems, making the country independent of external technology providers in the face of the ongoing conflict.

3 Min Read
Ukraine
Source: Freepik

Ukraine has taken the strategic decision to build its own Large Language Model (LLM). The project, implemented by the Ministry of Digital Transformation in partnership with operator Kyivstar, aims to create a system based on the open Gemma architecture from Google. The initiative sends a clear signal that Kyiv is seeking to make its critical systems – both civilian and military – independent of commercial solutions such as ChatGPT or Chinese technologies.

The decision to choose Google came after an analysis of competing solutions, including the Llama model from Meta and Europe’s Mistral AI. Deputy Minister of Digitalisation Oleksandr Bornyakov stresses that a key factor was the need to avoid dependence on proprietary systems from external suppliers. Indeed, the Ukrainian military plans to deeply integrate AI into its battlefield management systems, which excludes the risk of relying on foreign ‘black boxes’.

The architecture of the project involves a hybrid infrastructure approach. Initial training of the model will take place on Google’s secure computing clusters outside Ukraine to speed up the learning process. Ultimately, however, the finished system will be transferred to Kyivstar’s local infrastructure, providing Kyiv with full data sovereignty. This is particularly important in the context of the planned use of the model for court records, state archives and services for 23 million citizens.

Misha Nestor, product director at Kyivstar, points to pragmatic reasons for building an in-house tool. Current global LLM models struggle with the region’s specific linguistic context, often generating errors in translating legal documents or failing to cope with local dialects that are a mix of Ukrainian, Russian and Bulgarian. The new model, fed by data from more than 90 government institutions, is expected to accurately handle these nuances, including minority languages such as Crimean Tatar.

The venture is not without risk. The developers expect massive cyber attacks from Russia as soon as the system is launched. Nevertheless, the Ukrainian project could set a precedent for other smaller countries, showing how to use open source technologies to build strategic independence in the age of artificial intelligence.

Share This Article