MeetKai plans sovereign AI factories across six countries on NVIDIA infrastructure

MeetKai says it will roll out sovereign AI infrastructure across Brazil, Ukraine, Pakistan, Kazakhstan and Bangladesh, with NVIDIA hardware and software at every layer. The most consequential plan is in Ukraine, where a national-scale AI data centre is intended to begin coming online in 2027, starting at 15MW and potentially expanding to 100MW.

NVIDIA Watch analysis

What happened

MeetKai announced the six-country rollout on 25 September, saying its platform and locally trained models will run on NVIDIA accelerated computing. The company says the deployments are designed to operate inside each country, without dependence on foreign cloud services.

In Ukraine, MeetKai and Kyivstar plan to build and operate a sovereign AI data centre for government, critical infrastructure operators, businesses and Kyivstar customers. The proposed site would host Ukrainian-language models and support isolated deployments for organisations handling sensitive data. The announcement says the project remains subject to definitive agreements, hardware and power availability, financing, regulatory approvals and export controls.

MeetKai Brasil says it won a SERPRO tender for a national sovereign AI platform, including a locally hosted coding agent for secure workflows. In the other markets, MeetKai says it will work with VEON operators, beginning with Kyivstar and then extending to Pakistan, Kazakhstan, Uzbekistan and Bangladesh.

What changes

• A repeatable national stack
  MeetKai says the same MK-A1 SOVIA platform can be deployed across countries, with local models and infrastructure kept under domestic control.
• Ukraine’s proposed AI factory
  The planned facility is expected to start at 15MW in 2027 and could expand to 100MW as demand grows.
• NVIDIA from silicon to software
  The stack is described as using NVIDIA accelerated computing, networking, reference architecture and AI Enterprise software.
• Local operators, local engineers
  MeetKai says it and its telecom partners will operate each site and train engineers to run and extend the systems.

Why it matters

This is a concrete example of sovereign AI moving beyond speeches about keeping data local. The proposed deployments combine computing, models, networking, software and operations inside national or telecom-controlled environments. For governments and regulated industries, that could make it easier to train and run models on sensitive information without sending workloads to an overseas cloud.

It also shows why NVIDIA’s role in AI is broader than selling accelerators. If these projects proceed, NVIDIA technology would sit beneath national AI platforms, public-sector systems and telecom services across several markets. That is a significant distribution advantage, although the announcement does not yet prove that the facilities will be built, delivered on schedule or economically viable.

Our read

The interesting part is not the phrase “sovereign AI”, which has now visited enough press releases to qualify for its own passport. It is the proposed operating model: one infrastructure design, adapted with local models, partners and rules.

The plans are ambitious and clearly conditional. The useful test will be whether Ukraine reaches operational capacity in 2027, whether SERPRO’s deployment becomes a working public service, and whether the later markets receive functioning systems rather than infrastructure diagrams with excellent posture.

What to watch

• Whether MeetKai, Kyivstar and Ukrainian authorities sign definitive agreements and publish delivery milestones.
• Whether the Ukrainian facility reaches its proposed 15MW starting capacity in 2027.
• Independent evidence of performance, cost, energy use and reliability in the deployed systems.
• Whether VEON’s operators proceed beyond Ukraine into Pakistan, Kazakhstan, Uzbekistan and Bangladesh. Sources and evidence: Unite.AI’s 25 September account, based on MeetKai’s announcement, is the source for the rollout, partners, proposed capacity, NVIDIA technology and delivery conditions. The claims remain company-reported plans, not evidence that the projects are complete.

Discussion spark: Should sovereign AI projects prioritise domestic control of models and data even when that costs more than using a global cloud, or is portability the better protection against vendor lock-in?

Independent WittyWires tracker for public updates about NVIDIA. Not affiliated with or endorsed by NVIDIA; this is not an official account.