MeetKai takes sovereign AI built on NVIDIA to six countries

25 Sep 2026 (ET): MeetKai said its sovereign AI platform, MK-A1 SOVIA, will run on NVIDIA computers in Brazil, Ukraine, Pakistan, Kazakhstan, Uzbekistan, and Bangladesh — more than 700 million people, on the company’s count — after a SERPRO tender it says it won and a Ukraine AI factory whose first power is aimed at 2027.

SOFTWARE desk — a country that wants AI in its own language, on computers it controls, still has to stand up the whole stack, from the chips to the model. A repeatable platform matters because MeetKai says one design, MK-A1 SOVIA, can be copied from site to site without renting a foreign cloud, starting with a Brazil bid it says it already won and a Ukraine factory that is still a 2027 plan.

What the release says is rolling out. The subhead says a sovereign AI tender win in Brazil and the first national-scale AI factory in Ukraine open a six-country rollout with SERPRO, VEON, and Kyivstar. The lede says MeetKai announced a six-country rollout of its sovereign AI stack, built on NVIDIA AI infrastructure, across Brazil, Ukraine, Pakistan, Kazakhstan, Uzbekistan, and Bangladesh, six countries home to more than 700 million people. Sovereign, here, means the computers and the models are meant to stay inside the country. More than 700 million is the company’s count. In each country, the lede says, MeetKai’s sovereign AI platform and locally trained models will run on NVIDIA accelerated computing. Accelerated computing, here, means chips built to run AI work faster than an ordinary processor. The split on the page is MeetKai Brasil, working with SERPRO, and then the spelling “Meetkai” — a lowercase k — working with VEON (Nasdaq: VEON) and its operators in the other five markets, beginning with Kyivstar (Nasdaq: KYIV), called the leading telecommunications and digital services provider in Ukraine.

How the stack is described, and the one name for the platform. MeetKai says it builds the stack on NVIDIA at every layer. Its national language models are trained, fine-tuned, and served on NVIDIA accelerated computing. Fine-tuned means a model is trained further on a country’s own language and data. Served means that model answers requests. The sovereign AI platform runs on NVIDIA accelerated computing, NVIDIA networking, and NVIDIA AI Enterprise software. The page links the words NVIDIA networking to nvidia.com/en-us/networking/ and the words NVIDIA AI Enterprise to nvidia.com/en-us/data-center/products/ai-enterprise/. Networking, here, is the cables and switches that connect the chips. AI Enterprise is NVIDIA’s software for running AI in a data center. The AI factory reference design for sovereign sites is specified on NVIDIA infrastructure. One platform, MK-A1 SOVIA, across every market means a model built for one country and a deployment pattern proven at one site can both be repeated at the next, with no dependency on foreign cloud services. A foreign cloud, here, means renting the computers from a company in another country. MK-A1 SOVIA is the release’s name.

James Kaplan, as a quote, not as a site this desk toured. Kaplan, co-founder and CEO of MeetKai, said sovereign AI only works when the entire AI stack, infrastructure to models, sits inside the country. He said that is a hard thing to build once and a very hard thing to build six times. He said building on NVIDIA turns it into a repeatable design: the same accelerated computing under every site, the platform and national models on top. He said Ukraine will have AI it fully controls, Brazil takes the same approach into Latin America on the back of the sovereign AI win there, and the company will replicate it in every market in between. Those are his sentences on the wire. Fully controls, and will replicate, are his tense. They are not a factory this desk walked.

What the Ukraine plan is, and what is still a plan. MeetKai and Kyivstar, called VEON’s majority owned subsidiary and the leading Ukrainian digital operator, will build a sovereign AI data center in Ukraine, the first national-scale AI factory in the country, with first capacity planned to come online in 2027. Majority owned means VEON owns more than half of Kyivstar. The design of the facility is expected to expand from 15 to 100 MW as demand from the public sector and Ukrainian industry grows. A megawatt, MW, is a million watts of electric power. Fifteen is the starting design. One hundred is the expansion the release expects if demand grows. Expected, and planned, are the release’s tense. The building is not described as already on. It will be equipped with the latest NVIDIA accelerated computing and NVIDIA reference architecture and will run MeetKai’s sovereign AI platform, including Ukrainian-language models trained and hosted inside Ukraine. MeetKai and Kyivstar will operate the facility and deliver sovereign compute and AI services to the Ukrainian government, critical infrastructure operators, enterprises, and Kyivstar’s own customers. Will operate, and will deliver, are the release’s tense.

Who the release says could use an isolated copy. Because the platform is designed to run in isolated environments, it is expected that Ukrainian ministries, energy and transport operators, banks, and defense-adjacent civil agencies will be able to train, fine-tune, and run AI on their own data in fully air-gapped deployments. Air-gapped means the network is cut off from the public internet. Defense-adjacent civil agencies is the release’s phrase. It does not say a weapons system. It is expected, and will be able, are the release’s tense.

The five VEON markets, kept as two customer counts. Across VEON’s five markets, home to more than 150 million mobile customers and over 225 million digital customers, MeetKai acts as VEON’s AI lab-as-a-service, beginning in Ukraine followed by Pakistan, Kazakhstan, Uzbekistan, and Bangladesh, with each site operated by MeetKai together with VEON’s local operator. A lab-as-a-service, here, means MeetKai runs the AI lab with the phone company, rather than only handing over software. Mobile customers and digital customers are two figures. The release does not add them. In every market, MeetKai and its operating partners train local engineers to run and extend the sites, so AI skills diffuse into each country’s workforce and stay there. Train, in that sentence, is the release’s plan for people, not a class this desk attended. The five markets in that paragraph are Ukraine and the four that follow. Brazil is the SERPRO track, not one of those five.

Calista Redmond, as a quote. Redmond, VP Global AI Initiatives at NVIDIA, said every nation wants AI that runs on its own infrastructure, in its own languages, under its own rules. She said MeetKai is building its AI platform and national language models on NVIDIA accelerated computing, giving governments and telecom operators a full-stack foundation for AI factories deployed and operated in-country. That is her sentence on the wire. Full-stack, and deployed and operated, are her words. They are not a count of factories already switched on.

What Brazil is, in the release’s tense. In Brazil, MeetKai Brasil won SERPRO’s public tender for the national sovereign AI platform with the highest overall score. A tender is a formal bid for government work. Won is the release’s past tense for the bid. Highest overall score is the company’s claim. It will deploy the platform and locally trained models on NVIDIA accelerated computing in-country, including a sovereign coding agent for secure workflows, now launching with SERPRO. A coding agent, here, is software that writes or edits code. Will deploy is future. Now launching is the release’s phrase for the agent.

What the Brazil paragraph says the market is for, still the company’s words. The Brazilian partnership opens one of the Western hemisphere’s largest regulated markets. Brazil’s banks, insurers, energy companies, and state-owned enterprises carry workloads their regulators expect to stay secure under Brazilian control, and SERPRO, called the federal government’s own IT company, operates many of the systems much of that economy already runs on. IT, here, is information technology, the computers and the software an office runs on. The air-gapped platform Ukraine’s ministries and banks can use serves the same need in Brazil: institutions train, fine-tune, and run AI on their own data, in-country. And this sovereignty is unlocked on NVIDIA accelerated computing. Opens, serves, and unlocked are the release’s words. They are not a regulator’s letter this desk read.

The Egypt lines on the same page, which are not a seventh country in the rollout. MeetKai, which the release also calls an NVIDIA Inception member, is expanding its work in Egypt with Smart Africa, building its AI stack on NVIDIA technology as part of its plans for the market. The company is among a growing group of AI innovators using NVIDIA technology to build and scale across Africa. NVIDIA Inception is NVIDIA’s program for startups. The page does not define Smart Africa in this release. Egypt is not in the six-country list. Plans, and expanding, are the release’s tense. Africa is this paragraph’s. It is not the Brazil-to-Bangladesh rollout.

What the release itself marks as not yet done. The forward-looking statements cover the planned deployment of the platform and locally trained models in Brazil, Ukraine, Pakistan, Kazakhstan, Uzbekistan, and Bangladesh; the construction, equipping, expansion, and operation of AI data centers, including the planned facility in Ukraine and its expected timeline; expected demand from public-sector and enterprise customers; and the expected benefits of the collaborations. The risks it names include negotiation and completion of definitive agreements, the timing and completion of infrastructure projects, the availability of hardware, power, and financing, regulatory approvals and export controls, political and security conditions in the markets described, and the performance of counterparties. Nothing in the press release is a guarantee of future results. Statements concerning VEON, Kyivstar, or NVIDIA are subject to those companies’ own public disclosures and risk factors. Those sentences are the release’s caution. A definitive agreement is the signed contract that follows an announcement.

Who the about box says MeetKai is. MeetKai is a sovereign AI company that builds and operates national AI platforms for governments and telecommunications operators. The about box does not print a founding year, a headcount, a price, or a headquarters street. The dateline names Los Angeles and Santa Clara. The page does not print a media-contact email. SOURCE MeetKai.

What the card shows. The card is MeetKai’s official six-country launch graphic, not a wordmark. A dark field carries the headline “MEETKAI TAKES SOVEREIGN AI BUILT ON NVIDIA TO SIX COUNTRIES.” A world map highlights Brazil, Ukraine, Kazakhstan, Uzbekistan, Pakistan, and Bangladesh. Logos on the graphic are MeetKai, NVIDIA, Serpro, VEON, and Kyivstar. Those names are the graphic’s. No date is printed on the card. It is not the PR Newswire MeetKai and VEON wordmark, media MS1996513.

Plain English for the rest of the card. MeetKai says it will put the same AI platform, on NVIDIA chips, inside six countries: Brazil, Ukraine, Pakistan, Kazakhstan, Uzbekistan, and Bangladesh. The idea of sovereign AI, on this page, is that the computers and the models stay in the country. In Brazil the company says it won the government IT firm SERPRO’s bid and is launching a coding agent with them. In Ukraine it is planning a data center with Kyivstar, VEON’s phone company, with first power aimed at 2027 and a design that could grow from 15 to 100 megawatts. The other VEON markets on the list are Pakistan, Kazakhstan, Uzbekistan, and Bangladesh. The factory, the power, and the six copies are plans. The tender win is the sentence the company puts in the past.