VivaTech 2026, the event’s tenth edition, filled Hall 7 of the Paris Expo on the edge of the city. I spent three days inside, skipping most of the exhibition stands (the LVMH display was tempting, but the queues and the timing did not work out) and focusing instead on the talks staged across the three floors, where Publicis Groupe had built a world-class production.
Journalists ran most of the interviews at VivaTech, which added a good dose of perspective: Steven Levy, editor at large of Wired (one of my favourite publications, and a regular read since 1996), and Nicholas Thompson of The Atlantic, were able to tackle issues that an industry moderator or analyst cannot comfortably approach.
While the official slogan was “Artificial Intelligence: impact, not illusion”, in reality the show was, from start to finish, an argument about Europe: how far behind it has fallen in the AI race, and whether it can still catch up. The race runs on two tracks:
The first is the obvious commercial one, over AI models, chips and customers.
The second, and the one that mattered more here, is sovereignty: whether Europe controls its own AI, data and computing power, or depends on America and China. For Europe, that is a matter of survival, especially with the gap separating it with the other two giants widening fast.
That is the connecting thread of this special issue of The Compute. If you are deciding where to build, what to buy and whose technology to trust, anywhere from Abu Dhabi to London or Paris to Riyadh, VivaTech offered plenty of food for thought… read on!
LLM-pilled
Almost every leading AI today, American or Chinese, is built on the same principle: large language models (LLMs), the technology behind ChatGPT, which work by predicting the next word. Yann LeCun, one of the godfathers of modern AI, thinks that approach is close to its limit, and has bet his new company, AMI Labs, on a different one. His pitch: Europe should stop trying to beat America at the game everyone else is playing, and start a new game.
Mr LeCun told the audience that the whole Silicon Valley and tech industry has swallowed the LLM pill (taking an analogy from The Matrix’s red pill and blue pill scene), and that we should not expect a big leap while everyone races with nearly the same equipment. The way forward has to be built on a different technology foundation.
His alternative is world models: AI that learns how the physical world actually works, so it can predict what its own actions will cause, something an LLM cannot do. His example: an AI-enabled camera sweeping across the audience cannot exactly predict who is sitting in the part it has not yet reached; while a system that understood the real world could do that.
He likened the current times to the printing-press moment: great for spreading knowledge, but not the tool for creating it, and argued that whoever makes the pen should not get to decide what people write with it. He gave a nod of appreciation to the Chinese AI labs for keeping their models open for anyone to build on, after Meta pulled back from their open source direction.
Aidan Gomez, chief executive of Cohere and fresh from agreeing to buy Germany’s Aleph Alpha, dismissed Mr LeCun’s pitch on the grounds that AMI has nothing to show yet, while admitting that the breakthrough behind today’s AI, which he co-wrote in the 2017 Transformer paper, is nearly a decade old and that something new is soon overdue.
The takeaway: the experts cannot agree on whether the next big advance should merely improve today’s AI or actually replace it. Europe is quietly betting on “replace”, because a fresh start is the only chance of winning this race.
What AI cannot build
My favourite sessions were about the limitations: what today’s AI still cannot do, and where humans keep the edge. That gap is also where future opportunities stand in my opinion.
Peter DeSantis, who runs Amazon’s chip-building effort, said hardware and software now improve hand in hand, and that AI is still two to three orders of magnitude short of being broadly useful, with room to eventually run millions of times more efficiently than it does today.
He expects AI, and the chips beneath it, to move from generalists to specialists: Amazon’s own Trainium chips are built for training LLMs, while firms such as Groq and Cerebras build inference-optimised chips for running them, trading memory capacity for raw speed.
Mr DeSantis also expects the future to blend today’s LLM-based AI with Mr LeCun’s world models, and to add senses beyond text, such as ambient listening or reading a mood.
Tony Fadell, the engineer behind the iPod and Nest, doubts our gadgets will change much: screens are still the best way to deliver information, with voice doing more of the input work.
He had a blunt take that AI personal assistants would never work, since even good human personal assistants are rare to find and hard to train. According to him, the smarter use of AI is to let people handle the judgement at the start and end of a job, and hand the repetitive middle to the machine.
Peter Steinberger, whose tool OpenClaw helped push AI “agents” into the mainstream, said AI handles coding well because code can be checked, but taste is the thing that humans still do better than AI and they will keep this advantage for the long term.
Alexis Ohanian, who co-founded Reddit and now runs the $1.3bn fund Seven Seven Six, splits his bets in two: on one side on whatever fuels AI’s rise, the other on what AI cannot touch, such as live sport. That is why he started Athlos, a women’s track league, frustrated that the sport gets attention only once every four years, at the Olympics.
Mr Ohanian made the optimist case: no one will miss today’s cancer treatments once AI helps cure the disease. But he expects people to pay a premium for visibly human-made things, down to the labour of love in a good croissant (you will find a recipe worth the effort in After Hours at the end of this newsletter).
The takeaway: the opportunity for AI and beyond is in the work LLMs cannot currently perform well: judgement, taste and the human touch.
There is no you, digitally
Sovereignty has a personal side too: a country that does not control its citizens’ data controls very little, and neither do the citizens individually.
Sir Tim Berners-Lee, who invented the web, and John Bruce, who runs Inrupt, the company they founded together, argued that the web no longer works for the people who use it, and that individuals should own and control their own data.
They announced Charlie, a personal AI assistant that stores your data, picks the best AI for each task, and hides or disguises your personal details so a model can help you without learning who you really are. For now Inrupt is releasing it through trusted companies and governments, not to the public.
Ledger’s chief executive Pascal Gauthier highlighted the security gap bluntly: online, he said, there is “no digital you”, just data that anyone can copy. Proving who is who and who owns what is exactly what crypto was built to do.
He described a surge in cyber-attacks which we are only starting to see the damage of right now (like an underground nuclear explosion, whose tremors are barely felt at the beginning, but whose impending damage could be devastating), with AI making cyber attacks cheap and fast while cybersecurity remains slow and expensive.
According to Mr Gauthier, Europe should master three technologies: cryptography, AI and quantum computing, where he believes it can lead.
The takeaway: where your data sits, and who can read it, is not a technical detail, it is a national-security question, but also a very important individual issue. Solutions already exist to regain and maintain control, it is just a matter of adoption.
To infinity and beyond
Space was another theme throughout the week, and while I did not attendjust the Blue Origin headline act with Jeff Bezos and NASA astronaut Mike Massimino on day one, I did catch a couple of sessions on day three, masterfully hosted by space and science journalist Lise Pedersen, who managed to make rocket science appealing through a rare mix of know-how, warmth and stage command.
Maia Space, an offshoot of ArianeGroup, is building Europe’s first reusable mini-rocket, a launcher whose first stage flies back to land on a sea barge and be used again, the technique SpaceX has relied on for a decade.
Hélène Huby presented The Exploration Company, the Franco-German firm building Nyx, the first privately funded space capsule, designed to carry cargo to and from orbit and dock with the commercial stations that will replace the ageing International Space Station.
The Exploration Company has raised more than €200m (about $230m), including the largest space funding round Europe has seen, while this is dwarfed by the $87.5bn the SpaceX IPO raised, it loops us back to AI: orbital data centres are a way to feed our hunger for computing power by leaving Earth and tapping the sun’s energy directly.
Ms Huby’s stated aim is to become the Airbus of this new space industry, a European champion to rival the Americans. Europe has the building blocks, and in this realm, it is not far behind.
Where picks and shovels meet regulation
In a gold rush, the sure money is in selling picks and shovels, we saw this in the early days of the Internet too. In AI, that means the chips and the machines that the whole industry depends on. Here Europe holds a winning card, but is constantly trying to undercut it.
Christophe Fouquet runs ASML, arguably Europe’s most important technology company, and the biggest one many readers will never have heard of. He explained that the circuitry on an advanced chip is so fine that more than 10,000 transistor wires would fit across a single human hair. These transistors are etched through a very delicate process by extreme-ultraviolet (EUV) lithography machines that ASML is the only company in the world to build.
Nick Clegg, the former UK deputy prime minister and ex-Meta executive, now a venture investor, made an important comparison: the four biggest US tech firms are expected to spend more than $700bn on AI infrastructure in 2026, the equivalent of spending the entire UK defence budget every five to six weeks.
His point echoed Mr LeCun’s: Europe cannot win by copying America’s AI and must jump to something different, which is why he has personally invested in AMI Labs.
Regulation was on everyone’s mind. Renault chief executive François Provost warned of a “tsunami of regulation” by 2030. Tellingly, both he and Société Générale’s boss Slawomir Krupa played down how much AI they use themselves, cautious statements from two of France’s biggest employers, wary of alarming their troops.
Enrico Letta, the former Italian prime minister who wrote the EU’s 2024 report on the single market, named the fix as the fifth freedom, treating data as a basic freedom across the bloc. He also called for accelerating adoption of EU Inc: one Europe-wide company status, so a company can incorporate once and be able to operate seamlessly in all 27 member states.
The takeaway: Europe owns the one thing the entire AI industry depends on, the machines that make the chips, but risks burying that advantage under its own red tape.
For me, Mr Letta’s warning was the bluntest of the week: simplify the rules and pool the money, or spend VivaTech 2032 debating whether Europe should become a colony of America or of China.
Wirelessly yours,
Ziad Matar
Co-founder, Tokenando
Editor-in-chief, The Compute
After Hours
The Business France Vision Golfe 2026 programme ran on the sidelines of VivaTech, and although I could not attend the daytime conference and meetings, I made it to an elegant evening reception at the Palais Galliera, which houses the Paris fashion museum. The soundtrack that evening opened with the tenor Nabil Rashdi (Naestro on Instagram) performing a thundering Nessun Dorma. We wound down later at Cravan, a great bar on Boulevard Saint-Germain where they mix the classics and offer their own bottled cocktails. The Boulevardier was exceptional… with Night of Arara playing in the background.
I finally saw Paris properly on Saturday morning, with a leisurely run along the banks of the Seine and back past the Eiffel Tower, no tourists in sight except a few early birds, just like me.
While we do not have Alexis Ohanian’s croissant recipe as such, here is a vintage version to try next weekend, courtesy of the one and only Julia Child.
References
The less familiar names, for the curious:
AMI Labs, Yann LeCun’s world-models start-up: https://amilabs.xyz
Cohere, Aidan Gomez’s enterprise-AI firm: https://cohere.com
Aleph Alpha, the German lab merging with Cohere: https://aleph-alpha.com
OpenClaw, Peter Steinberger’s open-source agent: https://github.com/openclaw/openclaw
Inrupt and Charlie, Sir Tim Berners-Lee’s data-sovereignty venture: https://www.inrupt.com/meet-charlie
Ledger, Pascal Gauthier’s security firm: https://www.ledger.com