Ukraine has officially taken its biggest leap yet toward AI sovereignty. Kyivstar, working with the WINWIN AI Center of Excellence under the Ministry of Digital Transformation, has tapped Google’s Gemma as the base model for Ukraine’s first national large language model—a homegrown LLM built to understand the language, culture, and context of the country it serves.

It’s a significant alignment with U.S. tech, and a strategic follow-up to Kyivstar’s Nasdaq debut earlier this year. But more importantly, it’s a signal: Ukraine wants AI that speaks Ukrainian—not just literally, but contextually, securely, and on its own infrastructure.

Building a Model That Actually Understands Ukraine

The national LLLM will be trained on extensive Ukrainian-language datasets covering dialects, domain-specific terminology, and historical context. The goal is simple but ambitious: an AI model that can behave more like a Ukrainian expert than a global generalist.

By keeping all sensitive data within national borders, the system aims to meet the security thresholds of government agencies, health institutions, and financial organizations—requirements most off-the-shelf global models struggle to satisfy.

As Danylo Tsvok, Chief AI Officer at the Ministry of Digital Transformation, put it, the team prioritized two things during model selection:

1. Gemma’s multilingual strength, including strong baseline performance in Ukrainian.

2. Model controllability, ensuring the nation can mitigate linguistic and ethical risks during fine-tuning.

In other words, Gemma didn’t just win on tech—it won on trust.

Kyivstar Takes the Lead—and the Responsibility

Kyivstar, one of Ukraine’s most influential digital service providers, will operationally lead development. VEON Group CEO Kaan Terzioglu framed the initiative as both strategic and ethical, arguing that countries deserve AI models trained on more than just datasets—they should reflect local context, culture, and priorities.

That philosophy already fueled VEON’s earlier projects: KazLLM in Kazakhstan and a national Urdu LLM in Pakistan. Ukraine is the next key node in VEON’s push to close the “AI language gap,” where smaller or non-English-speaking nations rely heavily on foreign-built AI systems.

Why Google Gemma?

Google says Gemma offers an efficient balance of performance and resource usage—ideal for a country building sovereign AI under both technical and economic constraints.

Krzysztof Kaziów of Google Cloud noted the model’s strong multilingual support and track record with Ukrainian LLM deployments, emphasizing that Gemma wasn’t chosen as a gamble but as a proven foundation.

Next steps include:

• Optimizing Gemma specifically for Ukrainian

• Improving the tokenizer to better capture local linguistic nuances

• Training on carefully curated datasets

• Establishing Ukrainian benchmarks for future fine-tuning

This isn’t a quick tune-up; it’s a nation-scale AI engineering project.

What It Means for Ukraine’s Digital Future

Once completed, Ukraine’s LLM will serve as the backbone for AI-powered services across multiple sectors. Think:

• Government automation and regulatory tools

• Legal and compliance analysis

• Education platforms and tutoring systems

• Banking and fintech innovations

• Healthcare triage, documentation, and diagnostics

• Enterprise automation and customer support

The real advantage is local relevance. Global LLMs may understand Ukrainian—but they don’t deeply understand Ukraine. This national model aims to flip that script.

If successful, Ukraine could become a blueprint for how emerging markets can build secure, culturally aligned AI without reinventing the foundational technology.

And in the broader geopolitical landscape, selecting Gemma strengthens Ukraine’s fast-evolving digital ties with the United States—another data point in a year where AI and geopolitics are increasingly inseparable.