How InstaDeep became Africa’s biggest AI startup success

Karim Beguir starts his interview with Rest of World by bringing up Star Wars . This is something he does often, since he’s from the remote Tunisian city of Tataouine , which lent its name and otherworldly desert landscapes to the movie franchise.

For Beguir, the trivia highlights how distant his hometown is from the global centers of tech. It also conjures the image of an exciting, still-unexploited frontier.

Beguir, a mathematician, first made his career in finance, working for banks in Europe and the U.S. But in 2014, he left a cushy job in London and moved back to Tataouine. He wanted to do something more meaningful.

Starting an artificial intelligence company at home, where there were few precedents for tech entrepreneurship, was the most ambitious thing he could think of. He founded InstaDeep along with Zohra Slim, a Tunisian software engineer. Ten years later, the company has grown to more than 400 employees, with offices in London, Paris, Berlin, Tunis, Lagos, Cape Town, Boston, and San Francisco. InstaDeep is one of the few major AI startups based in Africa.

When the Covid-19 pandemic ground the world to a halt, InstaDeep trained a large language model to accurately predict new, dangerous variants before they spread. In 2023, InstaDeep was acquired for $682 million by BioNTech, a German pharmaceutical company that created the first Covid-19 vaccine approved by the U.S. Food and Drug Administration. The deal was the biggest tech acquisition ever in Africa.

For Beguir, moving into health care would be another sort of homecoming: He would follow in the footsteps of his father, a doctor.

In August, Beguir spoke with Rest of World over a video call from London, where he now lives. He spoke about InstaDeep’s early days, how the pandemic propelled the company, and why he thinks it’s vital to build AI companies in Africa.

This interview has been edited for length and clarity.

Tell me about when you started InstaDeep in 2014. How did the local ecosystem receive the idea of a Tunisian AI startup?

We went into AI after I read a paper on convolutional neural networks, which is essentially visual AI. My co-founder Zohra and I decided to take a stab at it and built a virtual AI system that would allow people to scan luxury objects like watches and bags to determine whether or not they were originals. A year later, we sold this to a client in London.

At that time, everyone was in complete disbelief that we were talking about doing AI in Tunisia. There was almost no startup ecosystem in the country, and building an AI startup seemed like a tall order.

In the early days, we were not set on expanding internationally. We tried to sell our first products in Tunisia, but we realized that the local market was not ready for AI. It was completely science fiction to people, and legacy companies had no appetite to pay for our technology. They didn’t understand the value; we only managed to make a few thousand dollars in sales locally. So we started looking outside.

When did you realize you had built a company that would last?

When I left London in 2014, I gave myself five years to try entrepreneurship. I began to see a blip of what we have become when I was invited to meet Mark Zuckerberg in Silicon Valley in 2017. Then, when we raised a series A led by AfricInvest , it became clear that this cool project actually had real potential.

At the time, we were building systems that optimized logistics in ride-hailing and trucking. DeepMind had built AI that could beat humans at complex games, and we found a way to apply its gaming concept to real-world problems. We were among the first in the world, if not the first, to apply DeepMind’s innovation to industrial optimization.

Our research papers attracted the attention of the DeepMind team, leading us to collaborations with DeepMind, Google, and the Massachusetts Institute of Technology.

Let’s talk about Covid-19. Everything shuts down, and InstaDeep becomes a global sensation after you build a system that predicts which Covid-19 variants might be a health risk. Can you walk me through that moment?

InstaDeep’s competitive advantage is that we are not fixated on one sector.

When Covid-19 hit, we were already working with BioNTech on personalized cancer vaccines. Very quickly, the BioNTech team deployed a Covid-19 vaccine. Then, we began to see new variants of the virus exploding by the day; at one time the number of variants sequenced every week exceeded 10,000. And I was like, “Who can follow this in real-time?”

It was clear to me that only an AI system could do it. So we built and trained an AI system that was essentially a generative model. We took transformers already pre-trained on all proteins that existed and then altered the language of SARS-CoV-2. We tried to see if we could predict to some level of accuracy whether or not something could be high-risk. It turned out that it worked. All the variants we identified as potentially dangerous were later confirmed as concerning by the World Health Organization.

Was that the beginning of the acquisition conversation with BioNTech?

It was definitely an important moment because when we started working with them, it was on a single project for a set amount of months. We were kind of being tested. The project was successful, so we expanded to do more projects. The more time passed, the more we proved the quality of what we were capable of doing and showed them our creativity, which is extremely important in AI. We built a great collaboration with our colleagues at BioNTech and the decision to join forces was a natural one.

Lightning Round

What’s the first thing you do in the morning?

Read the latest AI news on X

What’s your favorite Netflix show?

The Bureau

What was the last song you listened to?

Nightcall by Kavinsky

What’s the food you miss the most when you are away from home?

Mloukhia

If you had to start InstaDeep from scratch from another country, where would it be?

If it’s not Tunisia, I don’t think I’d ever do it

How do you unwind?

Reading history

The mainstream understanding of AI is generative AI, like ChatGPT, Midjourney, or Gemini. These are all programs that have demonstrated a bias toward developing countries. How is InstaDeep different?

The critical ingredient is to make sure those powerful technologies are actually adopted, used, and owned by different communities all across the world. However, those technologies are developed in the U.S. or Asia with their biases. Africans, for example, are far from the builders, and not the primary target audience. This is why the feedback loop from here doesn’t work.

At InstaDeep, we steer away from controversial applications of AI. We are focused on non-controversial ideas with net benefits. The only visual AI system we developed was in the early days and was entirely for objects, not people.

Nonetheless, our approach to fighting biases or imbalances in the AI world is to empower young people, who otherwise would not have the chance to build for their own countries. We have offices in Tunis, Lagos, Kigali, and Cape Town. Empowering young talent in Africa is really about creating a closer feedback loop. And it is also about ensuring that the incredible economic value that’s going to be unleashed in the coming years is also captured to some extent at home.

You are a big proponent of building in Africa, but you moved InstaDeep’s headquarters from Tunis to London in 2015. Why?

InstaDeep raised a $100 million series B in 2022. Due to how the Tunisian law operates for startups, I don’t think we would have been able to raise that amount locally. In this case, it was necessary to incorporate in the U.K., where I had lived and knew the ecosystem. The U.K. ecosystem is extremely dynamic, but more importantly, it’s easier to attract a high caliber of clients and investors as you scale.

What do you think are the advantages of building from Africa?

There are two big advantages: One is the community, and the second one is the opportunity to innovate. The African machine-learning community is a vibrant one. Take, for instance, the fact that Deep Learning Indaba, the yearly gathering of African AI researchers in different countries, was cited as a model for the world to adopt. Secondly, tech and digital is still nascent on the continent, and that makes room for innovation. Today, in the U.S., it’s challenging to come up with something new. ▰

Damilare Dosunmu is a reporter for Rest of World covering Africa's tech scene. He is based in Lagos, Nigeria.