At 22 rue Chapon in Paris’ ever-trendy third arrondissement, in a converted 18th-century mansion surrounding a vine-covered cobblestone courtyard, researchers from Meta and OpenAI are building the next generation of artificial intelligence.
These machine learning specialists cut their teeth in the perhaps less picturesque offices of Silicon Valley’s most powerful Big Tech companies, and they’ve now flown in to work on their own startups. And while they’re home to a tech scene that’s one-seventeenth the size of the United States, they say France is the perfect place to base an AI company today.
It comes as European nations are more and more trying to build strategic technology on home soil and reduce their dependence on American technology.
“I used to have two AI startups, and I moved to the US for those. If you weren’t in [San Francisco] you weren’t ‘someone’,” says Alex Lebrun, co-founder of health technology company IA Nabla and a former engineer in Meta’s IA division. “Now it’s actually really changed, and Paris is a big hotbed.”
Nabla — which uses large language models (LLMs) to help clinicians and reduce their administrative burden — shares its office building with Dust, an AI startup co-founded by former OpenAI research engineer Stanislas Polu. Not far from there, a generative AI startup Mistral – founded by former staff at Meta and Google’s AI labs – is looking to spend its €105 million funding round.
Solid foundations
Despite the backgrounds of some of these founders, Polu says the AI renaissance in France did not require a massive repatriation of prodigal sons and daughters from the United States.
“I don’t think it’s about people coming back [from Silicon Valley]; I think they have always been there. Paris is very successful these days because we have a very large pool of very good talent in AI research,” he says.
This includes Facebook and Google’s research labs in the capital, and “a lot of very good engineers and scientists”, according to Lebrun, who adds that it makes economic sense to expand the business there: “80% of our users are in the US but we are based in Paris because the AI engineers are good and cheap compared to the US.
Filtered newsletters
Subdued newsletter
3x per week
We tell you what’s happening in European startups and why it matters.
Nabla says the market salary for a junior ML engineer in Paris is around €60,000 today, and €80-90,000 for a senior ML engineer, while Silicon Valley rates are 2, 5-3 times that amount (data from recruitment platform Wellfound suggests that the average engineer salary at OpenAI is $268,000).
Polu says France’s talent pool is partly due to the country’s engineering degrees with a heavy concentration in mathematics, and partly to a government program called the Cifre system. The system encourages companies to hire doctoral students by contributing to their remuneration and is funded by the French Ministry of Higher Education, Research and Innovation.
“Basically, you get PhDs for free,” says Polu. “There has been a lot of this thesis applied research that has taken place in the labs of DeepMind [owned by Google] and Meta AI, so many of these PhD students have access to computation and can do real research during their thesis.
Aspiring Big Tech Talent
The fact that Paris is now home to three high-profile startups founded by alumni from Meta, OpenAI and Google represents the tip of the iceberg, Polu adds.
“DeepMind and Meta, as I see it from the outside, seem to have a harder time retaining talent these days,” he claims. “In the US, OpenAI is absorbing all that talent like a black hole, but they haven’t quite figured out how to do that in Europe yet. And so you end up with a pool of talent available in Paris and London – but in Paris in particular – that’s ready to explore and do things.
Sifted contacted DeepMind and Meta for comment, but did not receive a response before publication.
Polu’s Parisian optimism should be taken with a grain of salt. Recent Searches by Sequoia found that London has more than three times the share of AI talent in Europe than Paris (12.29% vs. 3.81%), and OpenAI recently chose the British capital as the location of its first office outside the United States.
But Mistral, Nabla and Dust have now raised money from some of the world’s biggest venture capitalists, like Sequoia and Lightspeed, as well as top business angels like Xavier Niel and former Google CEO Eric Schmidt – showing the impact of having a Big Tech name on founders’ resumes.
Different approaches
Although these startups share a hometown and experience with the world’s top AI labs, they take surprisingly different approaches to building businesses with generative AI.
At one end of the spectrum is Mistral – which raised his megaround four weeks after launch, days after staff was hired, and nothing close to a product to ship. The large sum will largely be used to train a proprietary AI LLM, something that costs tens of millions of euros due to the huge amount of computing power required.
Nabla, meanwhile, is building his product on OpenAI’s GPT-4 model, but hopes he’ll soon be able to transition to open-source LLMs as they improve in quality. Using third-party models means Nabla has been able to focus on its product – rather than basic research – and doctors are already using its AI clinician companion in hospitals across the US and Europe, including a pilot program with the British group Ramsay.
Dust falls somewhere between these two strategies. Like Nabla, the startup uses third-party LLMs – Polu saying models like GPT-4 still have a lot to offer.
“It’s the realization, working at OpenAI, that these models are already good enough to provide economic value, and yet they’re pretty under-deployed around the world,” he says. “The takeaway from this is really that it’s probably as much a research issue as it is a product issue at this point. And so the idea was to try to focus on the problem of the product.
Like Mistral, Dust is focused on building AI-powered products for enterprise customers, but — rather than building models — it tries to find new use cases for LLMs. The company is already working with six “design partners” to develop new generative AI applications in the workplace.
“When the smartphone came out, the app store was flooded with to-do list apps because that was the obvious thing. We didn’t know yet that we could build Uber and Airbnb and stuff,” he says “I think that’s where we are right now with LLMs. We are in this new paradigm, we have this new abstraction, as a new platform emerges.
Sovereignty
Dust and Nabla’s product-driven approach to building a generative AI business should yield faster results, with third-party LLMs ready to deploy today, but Mistral believes the biggest prize will go to those who own the underlying technology.
In a presentation note seen by Sifted, the company claims that “most of the value of the emerging generative AI market will be in the hard-to-manufacture technology, i.e. the generative models themselves”.
Lebrun takes a different view, arguing that free and open source LLMs will be able to compete with GPT-4 within a year: “I think the elements are new infrastructure, like cloud [computing]. Everyone will have access to it. Your LLM is not your fluke.
But Mistral is also betting on the ever-increasing attention paid to European technological sovereignty.
His note stresses that Europe has “not yet seen the appearance of a serious competitor” in the “great geopolitical challenge” of building founding models, adding that “specializing in the European market” will create “a defensible effort in itself”.
Lebrun agrees that the AI renaissance in Paris is music to the ears of the French political class, who lamented the exodus of AI talent from the country when companies like Hugging Face incorporated in the United States.
“Ten years ago, French politicians said: ‘All the brains are going to the United States!’ And I said to them: ‘Wait, they go to the USA, they earn money, they learn and then they come back to set up ambitious startups in France. This is exactly what happens with Mistral, Nabla and Dust.
It’s remarkable – and a little ironic – that these new European AI prodigies are backed by US dollars of venture capital. That said, more AI startups being created on the continent can only be positive if Europe tries to avoid a Silicon Valley hegemony of the latest wave of digital disruption.
