A keynote by Jean-Davy Nizigama, Founder of Mirai Technologies
TL;DR
- Today, almost every AI question asked in our region is answered abroad, under foreign laws, on foreign machines.
- That locks out the institutions holding the most important data: hospitals, banks, schools, ministries.
- Open AI models now exist that anyone can download and run, just as Linux once let any country run its own servers.
- The only thing missing is local infrastructure. Whoever owns the machines owns the answers.
- Kubana Cloud is building that infrastructure: AI where the weights and the prompts never leave the country.
Where your questions go
I want to begin with a simple picture. Imagine a doctor at a hospital in Kigali. She asks an AI system to summarise a patient's file. In that moment, the file leaves our country. So does her question.
It travels to a data centre in Virginia, or Dublin, or Singapore. It is processed under another country's laws, on machines owned by another country's companies. Then the answer travels all the way back.
None of us chose this. Not the doctor, not the hospital, not the government. It is simply the only arrangement that has ever been on offer. Every bank, ministry and clinic in our region that wants modern AI has faced one choice: send your data abroad, or go without.
We have been here before
Here is the good news. We have lived through this exact situation once already, and it ended well.
When I started out as an engineer, the older generation told us how things used to be. Running a computer system meant licensing someone else's software, on their terms, at their price, with no way to look inside. If you wanted to run a national payroll or a university network, you paid a foreign vendor for permission. And you kept paying.
Then Linux arrived, and everything changed. The software was open. Anyone could take it, run it on their own machines, and read every line of it. A ministry in Kigali or a bank in Bujumbura could build on its own servers and answer to nobody about how they ran.
My entire career has been built on that freedom. Ownership moved, not because someone negotiated a better deal with the vendor, but because the technology itself became free to hold.
The same moment, arriving again
I am standing here today because I believe that moment is happening again. This time, not for operating systems, but for intelligence itself.
An AI model's "weights" are simply what it has learned. For years, those weights were locked inside a handful of foreign companies. You could only reach them through their door, on their machines. That is no longer true. Powerful open models can now be downloaded, studied, adapted to our languages, and run on hardware we control.
So let me say this clearly: the obstacle to our countries running our own AI is no longer science. The science is published. The models are available. The only question left is where the machines sit, and who holds the keys.
That is an infrastructure question. And infrastructure is something we know how to build.
What sovereignty actually buys
Some of you may be thinking: this is a question of principle, of pride. It is not. It is a practical question about who gets to use AI at all.
A hospital cannot send patient records to a foreign processor. A bank cannot send its customers' transactions. A ministry cannot send its citizens' data. These are not preferences a clever contract can soften. They are obligations, and in most cases they are law.
Think about what that means. The institutions holding our most valuable data are exactly the ones shut out of this technology. They are not choosing between foreign AI and local AI. They are choosing between foreign AI and nothing.
Keep the machines inside our borders, and that impossible choice disappears. The data never moves, so the law is never broken. The hospital can use its own records. The bank can use its own transactions. What was forbidden becomes ordinary.
What this looks like in real life
Let me make this concrete, because this is not only a conversation for engineers. It is about people you know.
The district hospital. A doctor in small town in Rwanda wants an assistant that reads a patient's history and warns her about dangerous drug combinations before she prescribes. Today her hospital cannot allow it, because those records would leave Rwanda. With sovereign AI, the assistant runs in Kigali and the records never move. She gets her second opinion. Her patient keeps their privacy.
The farmer on the phone. A farmer calls a hotline and asks, in Kirundi, why his beans are turning yellow. An AI that knows our soils, our crops and our language answers him in seconds. Because it runs here, the service is cheaper and faster, and his question does not end up on a server on another continent.
The bank and the fraudster. A local bank wants AI to catch a suspicious mobile payment the moment it happens. Sending every transaction abroad to be checked is both illegal and slow. Run at home, the check takes a fraction of a second, and the customer's money is safe before the thief can withdraw it.
The ministry and the citizen. A government office wants to answer citizens' questions about permits, taxes and services, day or night. Those conversations carry ID numbers, addresses and family details. Kept in the country, the service can launch without asking citizens to trust a foreign company with their lives.
The school. A teacher wants a tutor that helps every child practise at their own pace, in their own language. Children's learning records are among the most sensitive data a nation holds. Local AI means a school can adopt the tool without exporting a generation's records.
The day the connection fails. An undersea cable is cut. Or a foreign provider changes its prices, or its rules, overnight. Every service built on distant AI stops, or suddenly costs twice as much. Every service built on local AI keeps running, because nothing it depends on sits outside our borders.
What we are building at Mirai Technologies
So what are we doing about it? At Mirai Technologies, through our cloud platform Kubana Cloud, our commitment fits in a single sentence: The weights and the prompts never leave the country.
The models will run on hardware in our region. The questions you ask them will be answered in our region. Nothing crosses a border, in either direction, at any point. For a developer, switching over takes one line of configuration. For a compliance officer, it means the conversation can finally begin.
This is not a leap into the unknown for us. We already run regional cloud infrastructure, we are an accredited registrar for the .bi/.rw domains, we already host and run servers in the region, we maintain and manage DNS infrastructure in the region. Sovereign AI is the same principle we have always worked by, carried one layer higher: technology built where it is used, under the laws of the people who use it.
This is early. The models are ready and the demand is forming. We would rather build the capacity before the region needs it than explain afterwards why it went elsewhere.
Who sets the terms
Let me close with this. Every generation of computing has divided the world into countries that own their infrastructure and countries that rent it. Those who own it set their own terms, keep the skills at home, and are not caught off guard when a foreign policy or a price changes. Those who rent it learn, again and again, how little of the arrangement was ever theirs.
This time, the infrastructure being divided is intelligence itself. And that division is being decided now, while the technology is still young enough for us to claim our share.
A country that owns its compute owns its answers. A country that does not will rent its intelligence, indefinitely, from people who owe it nothing.
I would like our region to be the first kind. Thank you.