Ukraine taps Nvidia to build "sovereign AI" for state and defense
Ukraine is moving ahead with a state-controlled AI infrastructure in partnership with Nvidia. The plan centers on "sovereign artificial intelligence" - government-run compute, models, and data pipelines built on Nvidia hardware and expertise. Officials frame it as a national security and data protection priority, not just a technology upgrade.
Mykhailo Fedorov, Ukraine's Minister of Digital Transformation, said the collaboration gives Kyiv access to top-tier tools to build sustainable, independent AI for government and defense. The objective: reduce dependency, speed up model training, and keep sensitive data under national control.
What "sovereign AI" means in practice
- National AI stack: State-owned compute and model infrastructure built on Nvidia platforms.
- Talent pipeline: AI education programs to grow domestic expertise and reduce skills gaps.
- Joint R&D: Co-developed projects to localize models, tooling, and deployment methods.
- Startup support: Resources for local builders to spin up products on the national stack.
- AI Factory continuity: Builds on Ukraine's existing "AI Factory" effort to deploy powerful infrastructure from Nvidia.
First deliverable: the Diia AI LLM
Ukraine's initial project with Nvidia is a large language model trained on Ukrainian laws, public services, and government data. The LLM will serve as the intelligence layer across the Diia ecosystem - powering the portal assistant and a future voice assistant in the app.
Diia is Ukraine's e-government platform used for IDs, permits, and public services. It has already introduced a state AI agent that answers questions and completes services in chat. More on the platform here: Diia (official site).
Why this matters for government leaders
- Data sovereignty: Keep sensitive datasets at home, under public governance and clear legal frameworks.
- Operational speed: In-house compute shortens deployment cycles and reduces external bottlenecks.
- Security posture: Tighter control over models that touch identity, benefits, defense, and critical infrastructure.
- Public trust: Transparent handling of data sources, auditing, and model behavior improves citizen confidence.
- Vendor exposure: Centralized hardware and software bring scale - and concentration risk. Plan for portability and exit paths.
Implementation priorities to get right
- Data governance: Classify datasets, define access controls, and log lineage from ingestion to inference.
- Security & compliance: Align with defense-grade standards; isolate workloads; encrypt in use, at rest, and in transit.
- Procurement: Multi-year agreements for GPUs, networking, and storage; performance SLAs; clear total cost frameworks.
- Hosting strategy: On-prem, sovereign cloud, or hybrid - with strict data residency and disaster recovery.
- Evaluation & safety: Red-teaming, benchmarks on law/public-service tasks, and policy guardrails for high-risk use cases.
- Interoperability: Connect models to registries, case-management systems, and identity services (e.g., Diia).
- Workforce: Upskill civil servants and technical teams; build roles for product owners, MLOps, and model evaluators.
Regional signals
Ukraine has already launched what it calls the world's first state AI agent on the Diia portal - not just answering questions, but executing services in chat. Albania is pursuing a similar path with "Diella," a virtual government member focused on procurement integrity. The direction is clear: more governments will embed AI directly into service delivery and oversight.
Action steps for government teams
- Identify 3-5 high-volume services where an LLM can reduce wait times or errors (permits, benefits, FAQs, procurement).
- Stand up a secure data lake for laws, regulations, forms, and prior decisions; prioritize clean, labeled data.
- Draft technical requirements: model latency/accuracy targets, observability, audit trails, red-team cadence.
- Form an AI review board spanning legal, security, operations, and citizen experience to approve deployments.
- Launch a skills sprint for core teams. Curated learning by role can help: AI courses by job.
Learn more
Bottom line: Ukraine is formalizing AI as national infrastructure - compute, models, and delivery - with clear use cases and a plan to build domestic capacity. For public leaders, the playbook is emerging: secure the stack, govern the data, ship targeted services, and train your people.
Your membership also unlocks:
