Today, 12.03.2026
Good morning, Human. Sometimes the most significant developments arrive not as disruptions but as declarations of intent. This week brought one of those moments: Yann LeCun, the Turing Award winner who spent twelve years building Meta's AI research operation, has raised $1.03 billion for a Paris-based startup that explicitly bets against the technology dominating the industry. The round is Europe's largest seed investment ever. The thesis is that large language models are a dead end.
The Lead: A Billion-Dollar Bet Against the Consensus
In November 2025, LeCun walked into Mark Zuckerberg's office and told his boss he was leaving. Four months later, Advanced Machine Intelligence Labs – AMI, pronounced like the French word for "friend" – announced $1.03 billion in seed funding, valuing the company at $3.5 billion on a pre-money basis.
The investor list reads like a who's who of global capital: Cathay Innovation, Greycroft, Hiro Capital, HV Capital, and Bezos Expeditions co-led the round. Nvidia, Toyota, Samsung, and Singapore's Temasek also participated, alongside French VC firm Daphni and a roster of prominent individuals including Tim and Rosemary Berners-Lee, Jim Breyer, Mark Cuban, and former Google CEO Eric Schmidt.
What exactly is AMI building? The short answer is world models – AI systems that learn abstract representations of physical reality rather than predicting the next word in a sequence. LeCun has been arguing for years that large language models, despite their remarkable fluency, are fundamentally limited. As he told WIRED: "The idea that you're going to extend the capabilities of LLMs to the point that they're going to have human-level intelligence is complete nonsense."
The technical approach centers on JEPA – the Joint Embedding Predictive Architecture – a framework LeCun first proposed in 2022. Rather than predicting the future in pixel-perfect or word-by-word detail, JEPA learns abstract representations of how the world works, ignoring unpredictable surface detail. The goal, according to The Next Web, is to build systems that understand physical reality the way humans and animals do: not through language, but through embodied experience.
The founding team is drawn almost entirely from Meta's AI research organization. Michael Rabbat, Meta's former director of research science, joins as vice president of world models. Laurent Solly, Meta's former vice president for Europe, becomes chief operating officer. Day-to-day operations will be led by Alexandre LeBrun, the French entrepreneur who previously founded medical AI startup Nabla.
Why does this matter for Europe? LeCun has been explicit about AMI's positioning as a European counter to American and Chinese AI giants. As Sifted reported, Pierre-Éric Leibovici, founder of Daphni, called AMI "the first European company to reach the scale of the GAFAM companies." The timing and geography are not coincidental. World models remain nascent, and the playing field is still open. Europe's expertise in industrial production and robotics – both cited as early use cases – could prove advantageous.
Within one to two years, LeCun told AFP, AMI plans to begin discussions with corporate partners. Within three to five years, the goal is to produce "fairly universal intelligent systems" capable of deployment across almost any domain requiring machine intelligence. Whether that timeline proves realistic or aspirational, the capital is now in place to find out.
The Funding Picture
AMI's round arrives in a week that also saw movement at the earlier end of the funding spectrum. Samaipata, the pan-European venture capital firm founded in Madrid in 2016, launched its third fund with a target of €110 million and an explicit focus on AI-native startups at the earliest stages of development. The fund has already held its first close at €70 million – 64% of target.
The investor base includes institutional names with significant reach: Germany's state development bank KfW, Spain's SETT (the state entity for industrial transformation and digitalisation), and a group of prominent Spanish family offices. The participation of state-backed capital signals growing conviction that Europe needs dedicated vehicles for AI investment at seed and early stage.
Samaipata's shift to AI-native marks a meaningful evolution in thesis. The firm made its name backing digital platforms with network effects. Fund III extends that logic into a new architectural layer: instead of backing companies that build platforms on top of existing software, Samaipata wants to back companies building AI systems from the ground up. The fund will invest in 25 to 30 companies, with capacity to allocate up to €10 million per startup over the relationship's course.
The broader European funding picture shows AI emerging as the region's leading sector for startup investment for the first time. According to Crunchbase data, around $17.5 billion flowed into European AI companies in 2025, compared to just over $10 billion in 2024. European venture funding overall reached $58 billion last year – up only 9% year over year, but with a notable shift toward deep tech.
The Infrastructure Play
While frontier labs capture headlines, the enterprise AI infrastructure layer continues its quieter evolution. Tech.eu reported this week on IBM's watsonx platform strategy, noting the company's collaboration with Datavault AI as evidence that IBM has become "a huge player in building AI systems and infrastructure."
The Datavault partnership, announced in January, deploys enterprise-grade AI at the edge in New York and Philadelphia through the SanQtum AI platform – a fleet of synchronized micro edge data centers running IBM's watsonx portfolio on a zero-trust network. The deployment enables real-time data tokenization, scoring, and ultra-low-latency processing without reliance on public cloud infrastructure.
IBM's positioning matters for European enterprises navigating AI adoption. The company has been named a Leader in seven AI-related Gartner Magic Quadrant reports in 2025 and 2026, spanning data science, AI application development, cloud database management, and data governance. The recognition reflects watsonx's ability to execute across the full AI lifecycle – not just in one area – with particular strength in hybrid and multi-cloud flexibility.
For European organizations concerned about data sovereignty and regulatory compliance, IBM's emphasis on deployment flexibility and governance integration addresses real operational constraints. The platform connects AI development, agents, governance, and data in a single ecosystem, reducing the fragmentation that comes from stitching together multiple vendors.
The Regulatory Calendar
August 2, 2026 looms as the next major enforcement date for the EU AI Act, when the regulation's core framework becomes broadly operational. As ISMS.online reported, this triggers comprehensive requirements for high-risk AI systems – spanning risk management, data governance, technical documentation, record-keeping, transparency, human oversight, accuracy, robustness, and cybersecurity.
The Commission, however, missed its February 2 deadline to provide guidance on how operators of high-risk systems can meet their obligations under Article 6. The guidance was supposed to include a comprehensive list of use cases to help businesses distinguish between high-risk and non-high-risk AI systems. According to IAPP, the Commission indicated it is integrating months of feedback and plans to publish a final draft by the end of March, with final adoption potentially coming in April.
The delay reflects ongoing tension between implementation ambition and operational reality. The Commission's Digital Omnibus proposal, introduced in late 2025, would simplify what counts as high-risk AI use and push back the high-risk entry into force by up to 16 months – to December 2027. But as Hyperight noted, the European Data Protection Board and EDPS warned in January that these "simplifications" risk diluting accountability and weakening fundamental rights.
For organizations building or deploying AI in Europe, the practical implication is uncertainty. The law exists, but the guidance needed to operationalize it remains incomplete. Spain's Agency for the Supervision of Artificial Intelligence (AESIA) has released 16 guidance documents to support compliance – a useful resource, though subject to revision once the Digital Omnibus amendments are finalized.
The Numbers That Matter
$1.03 billion – AMI Labs' seed round, Europe's largest ever, valuing the company at $3.5 billion pre-money (PitchBook)
€70 million – Samaipata Fund III first close, representing 64% of the €110 million target (BeBeez)
$17.5 billion – Total venture funding to European AI companies in 2025, up from $10 billion in 2024 (Crunchbase)
7 – Number of AI-related Gartner Magic Quadrant reports naming IBM a Leader in 2025-2026 (IBM)
18% – Share of companies that believe they are prepared to comply with EU AI Act rules, per Littler 2025 European Employer Survey (Anadea)
August 2, 2026 – Date when EU AI Act high-risk system requirements become enforceable (AI Act Implementation Timeline)
The Week Ahead
The Commission's delayed Article 6 guidance on high-risk AI classification is expected to reach final draft form by month's end. Watch for signals on whether the Digital Omnibus timeline adjustments will hold or face further negotiation in Parliament.
AMI Labs will likely begin building out its Paris headquarters and additional offices in New York, Montreal, and Singapore. The company has already agreed a partnership with Nabla to bring FDA-certifiable agentic AI systems to healthcare – an early indicator of where world model applications may land first.
Samaipata continues its fundraising toward the €110 million target. The fund's focus on B2B companies that can abstract AI deployment complexity for real-world use cases positions it to capture deal flow from the growing cohort of European AI-native startups.
The Thought That Lingers
There's something almost poetic about LeCun's timing. At the precise moment when large language models have achieved cultural ubiquity – when ChatGPT has become a verb and AI assistants are embedded in everything from email to enterprise software – one of the field's most decorated researchers raises a billion dollars to prove the whole approach is wrong.
The bet isn't that LLMs are useless. LeCun has been clear that they're becoming genuinely good at generating code and will prove useful across many applications. The bet is that they're not the path to intelligence. That real understanding starts not in language but in the world. That the systems capable of reasoning, planning, and navigating physical reality will look fundamentally different from the ones predicting the next token.
Whether he's right or wrong, the capital is now in place to find out. And for Europe, that capital is headquartered in Paris.
Human×AI Daily Brief is compiled from TechCrunch, WIRED, The Next Web, Sifted, PitchBook, Bloomberg, Tech.eu, BeBeez, IBM Newsroom, Crunchbase News, ISMS.online, IAPP, and regulatory sources. This is meant to be useful, not comprehensive.