One Stack, Six Nations: MeetKai and NVIDIA Ship Sovereign AI to 700 Million People

Sovereign AI has mostly been a slogan with a slide deck. On September 25, 2026, it got a delivery schedule. MeetKai, a Los Angeles-based sovereign AI company, announced a six-country rollout of its full national AI stack — built entirely on NVIDIA accelerated computing — across Brazil, Ukraine, Pakistan, Kazakhstan, Uzbekistan and Bangladesh. These six countries are home to more than 700 million people, and in each one the platform, the models, and the silicon will run inside national borders, with no dependency on foreign cloud services.

The announcement matters because it converts the abstract idea of “AI sovereignty” into a concrete, repeatable engineering design — one platform, called MK-A1 SOVIA, deployed six times with local partners operating every site.

What was actually announced

MeetKai is deploying its sovereign AI platform and locally trained models in six markets, working through two anchor partners:

• Brazil — MeetKai Brasil, working with SERPRO, the Brazilian federal government’s own IT company. MeetKai Brasil won SERPRO’s public tender for the national sovereign AI platform with the highest overall score, and is now launching a sovereign coding agent for secure workflows with SERPRO.
• Ukraine and four more VEON markets — MeetKai working with VEON (Nasdaq: VEON) and its operators, beginning with Kyivstar (Nasdaq: KYIV), Ukraine’s leading telecommunications and digital services provider, followed by Pakistan, Kazakhstan, Uzbekistan and Bangladesh.

The choice of partners is the strategy. SERPRO already operates much of the Brazilian government’s systems, and VEON’s five markets cover more than 150 million mobile customers and over 225 million digital customers. MeetKai is not selling software into these countries — it is embedding with the incumbents who already run national infrastructure, and acting as what VEON calls an “AI lab-as-a-service.”

Ukraine gets the first national-scale AI factory

The centerpiece of the rollout is a sovereign AI data center in Ukraine, built by MeetKai and Kyivstar — the first national-scale AI factory in the country. First capacity is planned to come online in 2027, and the facility’s design is expected to expand from 15 to 100 MW as demand from the Ukrainian public sector and industry grows.

The site will use the latest NVIDIA accelerated computing and NVIDIA reference architecture, running MeetKai’s sovereign AI platform with Ukrainian-language models trained and hosted inside Ukraine. Because the platform is designed for isolated environments, Ukrainian ministries, energy and transport operators, banks, and defense-adjacent civil agencies are expected to be able to train, fine-tune and run AI on their own data in fully air-gapped deployments.

For a country at war, air-gapped, domestically controlled AI is not a compliance checkbox — it is an operational requirement. A 100 MW AI factory is roughly the scale at which sovereign compute stops being symbolic and starts being able to serve an entire government’s workload.

The NVIDIA layer is the whole story

What makes this repeatable six times over is the decision to build on NVIDIA at every layer of the stack:

• National language models trained, fine-tuned and served on NVIDIA accelerated computing
• The sovereign AI platform running on NVIDIA accelerated computing, NVIDIA networking, and NVIDIA AI Enterprise software
• An AI factory reference design for sovereign sites specified on NVIDIA infrastructure

“Sovereign AI only works when the entire AI stack, infrastructure to models, sits inside the country,” said James Kaplan, co-founder and CEO of MeetKai. “That is a hard thing to build once and a very hard thing to build six times. Building on NVIDIA turns it into a repeatable design: the same accelerated computing under every site, our platform and national models on top.”

Calista Redmond, VP of Global AI Initiatives at NVIDIA, framed the deal as evidence of a broader shift: “Every nation wants AI that runs on its own infrastructure, in its own languages, under its own rules.”

The standardization argument is the quiet punchline. A model built for one country and a deployment pattern proven at one site can both be repeated at the next. Sovereignty used to mean bespoke, expensive, one-off builds. A reference architecture makes it a product line — which is exactly why NVIDIA has been so aggressive in this market. Sovereign AI already accounts for roughly 14 percent of NVIDIA’s revenue by some estimates, and deals like this one extend that lead.

Brazil is the regulated-market beachhead

While Ukraine is the flagship build, Brazil may be the more commercially significant win. MeetKai Brasil’s SERPRO tender victory opens one of the Western hemisphere’s largest regulated markets. Brazilian banks, insurers, energy companies and state-owned enterprises carry workloads their regulators expect to keep secure under Brazilian control — and SERPRO operates many of the systems that economy already runs on.

The same air-gapped platform design serves both markets: institutions train, fine-tune and run AI on their own data, in-country. The sovereign coding agent launching with SERPRO is a first concrete workload, aimed at secure government development workflows.

Skills stay in-country too

A detail that is easy to miss: in every market, MeetKai and its operating partners train local engineers to run and extend the sites. The explicit goal is that AI skills diffuse into each country’s workforce and stay there. That addresses the most common failure mode of national AI programs — a foreign-built data center that becomes a foreign-operated black box after the ribbon-cutting.

MeetKai is also an NVIDIA Inception member and is expanding related work in Egypt with Smart Africa, building its stack on NVIDIA technology as part of its plans for that market.

Why this matters

Sovereign AI has been the industry’s favorite conference theme for two years, but most announcements have been memoranda of understanding, tax-break announcements, or single-country pilot programs. The MeetKai rollout is different on three axes:

1. Scale — six countries, 700 million people, one platform architecture.
2. Specificity — a named tender win in Brazil, a named facility in Ukraine with a power envelope (15→100 MW) and a timeline (first capacity 2027).
3. Full-stack commitment — not just GPUs in a local data center, but locally trained language models, an AI factory reference design, and workforce programs.

The risks are real and the company discloses them plainly: definitive agreements still need to be negotiated, hardware and power availability can slip, and regulatory approvals, export controls, and political conditions in each market can all delay or derail timelines. A 2027 first capacity date for Ukraine is a plan, not a fact on the ground.

Still, the direction is unmistakable. The frontier-model race gets the headlines, but the quieter contest is over who supplies the AI infrastructure of mid-sized powers — and this deal is the clearest signal yet that NVIDIA-and-partner turnkey stacks are the emerging default answer. For the 700 million people in these six countries, the AI they encounter from governments and telecom operators over the next few years is increasingly likely to be trained, hosted, and governed at home.