I meet Karim Beguir, the cofounder of AI systems developer InstaDeep, during one of his frequent visits to France. This time, he’s come over to help organise the French government’s AI Action Summit, which is due to take place next month — just over a year after the UK’s own AI Safety Summit.
Beguir, who is Franco-Tunisian, has just come from the Élysée presidential palace, where, he tells me, he’s been contributing ideas for the event. It’s shaping up to be a big show, starring big-hitters such as OpenAI’s Sam Altman, Anthropic’s Dario Amodei and Mistral AI’s Arthur Mensch. 100 heads of state have been invited, according to Beguir.
The event is another example of France’s hopes to re-assert its relevance on the AI stage — and Beguir, who launched InstaDeep over 10 years ago, can attest to how far French AI startups have come.
“In the past year, we’ve had fantastic stories of startups fundraising in France,” Beguir tells Sifted. “But you have to remember that five or six years ago, this was all science fiction.”
There’s still a long way to go for French — and European –– AI startups to compete against their US counterparts. And for Beguir, the make or break factor will be compute power.
“It’s literally almost only a question of scale,” says Beguir. “The winners will be… those who will be able to access 1m or even 10m GPUs in the next five years.”
Keeping up Europe’s AI momentum
Beguir launched InstaDeep in 2014 in Tunis, before moving HQ to London in 2016. The startup, which builds AI-powered systems for different industries, was acquired in 2023 by German biotech company BioNTech (of Covid vaccine fame) in a €664m cash and earnout deal. After the acquisition, InstaDeep remained operationally independent and Beguir stayed on as CEO.
The founder, who grew up between Tunisia and France, has kept close ties with French tech. Paris is InstaDeep’s largest office, with 150 employees — mostly AI engineers, researchers and high-performance computing specialists. As an alumni of prestigious engineering school École Polytechnique — which was also attended by two of Mistral’s three cofounders — Beguir says that he’s seen first-hand that French academia can produce some of the brightest minds in AI.
This is why France now has a number of serious AI players, says Beguir, including Mistral (which he is an investor in). But keeping up the momentum, he says, will require more than talented teams.
“The winners will be those who have talent, supported by a mountain of compute power,” says Beguir.
“In terms of technical feasibility, we have all the expertise to do it at scale… But what’s missing is the financial means to fund infrastructure for compute power.”
Investing in GPUs
Training AI models requires access to specialised chips called graphics processing units (GPUs). GPUs lie in data centres, many of which are currently owned by US cloud companies like Microsoft, AWS and Google.
The investments made by US companies towards increasing their compute power are “simply stupendous,” says Beguir. Earlier this year, Microsoft announced that it was on track to invest about $80bn in data centres for AI, about half of which will be in the US. “This gives you an idea of the scale we’re at,” says Beguir.
The US is also home to the world’s largest AI supercomputer, Colossus, which was built by billionaire entrepreneur Elon Musk’s AI startup xAI and was used to train the company’s large-language model (LLM) Grok.
Colossus has 100k GPUs and is currently doubling in size with the objective of reaching 200k GPUs.
In comparison, French cloud provider Scaleway, which was founded by French billionaire Xavier Niel, a strong advocate for the need to build homegrown AI infrastructure, recently announced that it is operating 5k GPUs.
Beguir says that it is not just France, but Europe as a whole, which faces a fundamental challenge. “The question of European sovereignty is pretty simple,” he says. “Do we have, in Europe, clusters that can develop and train the biggest models in the world, or will this all be done in the US via the cloud?”
European governments are starting to acknowledge the issue. Last week, the UK government unveiled a number of actions it plans to take to boost the country’s AI sector, which included a major focus on investments in compute infrastructure.
The plan mentioned a £2.5bn ($3bn) investment commitment into the country over the next three years from UK data centre startup Nscale; as well as a £12bn ($15bn) commitment from US company Vantage Data Centres (which did not come with a specific time frame).
Beguir says there needs to be political willingness to build more AI infrastructure, not at the level of a single country but across Europe.
“Politicians need to find ways to attract private investments,” he says. “If it’s purely a government initiative, it won’t be enough. We need to bring in large European funds.”
“It’s a delicate moment for Europe. We’ve invested a few hundreds of millions of euros, but it’s not enough — we should invest at least ten times more… So, are we ready to multiply the scale of our efforts?”
Read the orginal article: https://sifted.eu/articles/karim-beguir-instadeep-interview/
