Summary
- Kyivstar and MeetKai plan a Ukrainian sovereign AI data centre with first capacity expected in 2027.
- The design is intended to expand from an initial 15MW towards 100MW as demand develops.
- The facility is planned around Nvidia accelerated computing and locally hosted Ukrainian-language AI models.
Kyivstar and MeetKai plan to build a sovereign artificial-intelligence data centre in Ukraine, with an initial 15MW design expected to scale towards 100MW as public-sector and industrial demand grows.
The companies expect the first capacity to come online in 2027. The facility is being described as Ukraine’s first national-scale AI factory and is intended to host computing infrastructure, networking, software, and locally governed AI models inside the country.
MeetKai said the design will use Nvidia accelerated computing, networking, and AI Enterprise software. Its sovereign AI platform is expected to include Ukrainian-language models trained and hosted within Ukraine rather than depending on a foreign public-cloud environment.
Kyivstar is majority owned by VEON and is best known as a telecommunications operator. The proposed project would extend that infrastructure role into dedicated AI compute at a scale that, if the later stages are delivered, would resemble a substantial data centre campus rather than a conventional enterprise computing deployment.
Sovereignty becomes a physical design requirement
The term sovereign AI is often used in software and policy discussions, but a national-scale deployment makes it a physical infrastructure problem as well. Keeping model training, inference, storage, and supporting services within a jurisdiction requires enough local compute capacity, connectivity, power, cooling, and operational expertise to avoid pushing the workload back into external cloud regions.
The proposed 15MW first stage gives the project a relatively clear starting point. The path to 100MW is more consequential because the infrastructure requirements do not scale solely through the addition of more GPU servers.
Power distribution, substations, backup systems, cooling loops, water or heat-rejection arrangements, network capacity, and physical-security systems all have to grow with the IT load. High-density accelerated computing also places particular pressure on thermal design, making the cooling architecture one of the central engineering questions for any later expansion.
MeetKai has not disclosed a detailed cooling design, site location, grid arrangement, construction contractor, or delivery timetable for the full 100MW. The upper figure should therefore be treated as the intended design trajectory rather than commissioned capacity.
The Ukraine project forms part of a wider MeetKai sovereign-AI rollout covering six countries. The company’s model is to use a common platform and infrastructure design while keeping national models and deployment environments under local control.
15MW is the first delivery test
The staged approach means the first 15MW will carry more significance than the 100MW headline until the project reaches construction. It will establish whether the partners can move from a national AI strategy into an operating data centre with the power, cooling, network, and software layers functioning together.
Scaling afterwards will require each of those layers to remain expandable. A facility whose electrical or thermal plant is designed only around the first deployment could face expensive retrofit work if later GPU generations increase rack density faster than the site’s base systems can support.
The choice of Nvidia reference architecture may help standardise the computing and networking environment, but reference designs do not remove site-specific constraints. Electrical utility conditions, construction availability, redundancy requirements, local connectivity, and operational staffing remain tied to the individual facility.
The project also adds another form of demand to Europe’s rapidly expanding data centre pipeline. Instead of a multinational cloud provider building a conventional region, Kyivstar and MeetKai are proposing nationally controlled infrastructure around language models and public- and private-sector AI workloads.
That model places sovereignty alongside the established drivers of data centre location — power, fibre, land, latency, and regulation. Where governments or nationally important companies want compute and models retained within their own borders, physical data centre capacity becomes part of that policy rather than an interchangeable backend service.
The next meaningful milestones will be the site, power arrangement, cooling architecture, and construction programme for the initial 15MW. Those details will determine how quickly the 2027 target becomes operating capacity and how credible the longer-term route towards 100MW proves to be.
