AMI Labs’ Billion-Dollar Seed: How JEPA and Mega-Rounds Are Reshaping AI Funding

The Round That Rewrote the Playbook

In March 2026, a company with roughly 12 employees and no shipped product raised more money in a single seed round than most startups raise across their entire lifetime. AMI Labs, the Paris-headquartered AI research company co-founded by Meta’s former Chief AI Scientist Yann LeCun, closed a $1.03 billion seed round at a $3.5 billion pre-money valuation — making it the largest seed round in European history.

LeCun announced his departure from Meta in November 2025 after 12 years — five as founding director of Facebook AI Research (FAIR) and seven as Chief AI Scientist. By December 2025 the company name was confirmed. Four months later, the seed was done. The speed alone tells you something about how the venture market has recalibrated for AI.

The round was co-led by Cathay Innovation, Greycroft, Hiro Capital, HV Capital, and Bezos Expeditions. Strategic investors include Nvidia, Samsung, Sea, Temasek, and Toyota Ventures, alongside French backers Groupe Industriel Marcel Dassault and Publicis Groupe. Individual investors include Eric Schmidt, Mark Cuban, Xavier Niel, and Tim Berners-Lee.

Alexandre LeBrun, a French entrepreneur who previously founded medical AI startup Nabla, serves as CEO. LeCun holds the Executive Chairman role while remaining a professor at NYU — a structure that signals AMI Labs is building a research-first organization with commercial leadership from day one.

Why JEPA Attracted a Billion Dollars

LeCun is not a newcomer seeking validation. He is a Turing Award winner, the architect of convolutional neural networks powering modern computer vision, and the person who spent a decade at Meta arguing — publicly and provocatively — that large language models are a dead end for genuine machine intelligence.

His thesis is straightforward. LLMs are sophisticated pattern matchers that predict the next token in a sequence. They generate fluent text, pass bar exams, and write serviceable code. But they do not understand the physical world — how objects behave, how gravity works, why a glass shatters on tile.

Joint Embedding Predictive Architecture (JEPA), proposed by LeCun in 2022, is his alternative. Rather than predicting sequences of tokens, JEPA systems learn to predict abstract representations of the world. Instead of generating pixel-by-pixel predictions of what happens next in a video, a JEPA model learns high-level abstractions: that a falling object accelerates, that a door opening reveals a room, that a person walking toward an edge will stop or fall.

AMI Labs is training on video, audio, and sensor data — not just text. The target applications are industrial automation, robotics, and healthcare, with Nabla as the first disclosed partner. If JEPA works as theorized, it could produce AI systems that genuinely understand physical causality — something no amount of text-prediction scaling has achieved.

The investors are not pricing AMI Labs on revenue multiples. They are pricing three factors: founder scarcity (perhaps five people alive could credibly lead a from-scratch alternative to the LLM paradigm), paradigm optionality (if JEPA proves even partially superior for physical world understanding, the robotics and embodied AI markets exceed $500 billion by 2030), and competitive urgency (every major lab is now investing in world models).

The New Economics of AI Mega-Rounds

AMI Labs’ round did not happen in isolation. It is the most dramatic data point in a broader restructuring of how venture capital flows into AI companies.

The numbers from early 2026 are staggering. In February alone, AI-related startups raised $171 billion — 90% of all global venture funding that month. AI companies accounted for 41% of the $128 billion deployed through Carta’s platform in 2024-2025, a record annual share. The concentration is intensifying.

The mega-rounds tell the story:

These numbers would have been inconceivable in any previous technology cycle. Three companies — OpenAI, Anthropic, and xAI — raised a combined $160 billion in the first two months of 2026 alone.

Why Seed Rounds Now Look Like Growth Rounds

The traditional venture funding ladder — pre-seed, seed, Series A through C — was designed for software companies where primary costs were developer salaries and cloud hosting. A talented team of five engineers could build a viable SaaS product for under $5 million.

AI infrastructure companies operate on fundamentally different economics. Training a frontier model costs $100 million to $1 billion in GPU compute alone. Senior AI researchers command $1-5 million annual compensation packages. The competitive window moves in quarters, not years. A company spending 18 months on successive small rounds while competitors train next-generation models will arrive at market with yesterday’s architecture.

The billion-dollar seed is a time-compression strategy: raise everything upfront and focus exclusively on research.

The Barbell Effect

The AI funding landscape is developing a pronounced barbell shape. On one end, a small number of companies raise rounds measured in billions. On the other, a long tail of AI application companies raise modest rounds ($5-20 million) to build vertical products atop foundation model APIs.

The middle is hollowing out. The $50-200 million range that historically funded infrastructure-stage companies is increasingly inadequate for AI research but excessive for application companies. The word “seed” has been emptied of its traditional meaning — it no longer implies small or experimental. It simply means “first round.”

The European Angle

AMI Labs’ choice of Paris as headquarters is significant. European AI has long suffered from a capital gap — talented researchers trained at European universities who migrated to Silicon Valley because that was where the funding was.

This round suggests the dynamic may be shifting. Paris, London, and Zurich have world-class AI research communities anchored by INRIA, DeepMind London, and ETH Zurich. Senior AI researcher salaries in Paris run 30-40% below San Francisco rates. European governments and sovereign wealth funds are increasingly willing to back AI infrastructure to reduce dependence on US-dominated platforms.

AMI Labs joins Mistral AI (which raised €1.7 billion in September 2025 at a ~$14 billion valuation) as evidence that European AI can compete for capital at the highest levels. If these companies succeed commercially, they could break the long-standing pattern of European researchers building American companies.

What This Means for Emerging Market Founders

For technology ecosystems outside the Silicon Valley-London-Paris triangle, AMI Labs’ round contains both a warning and an opportunity.

The warning: competing at the AI infrastructure layer now requires capital reserves inaccessible to startups in most emerging markets. No North African, Middle Eastern, or Southeast Asian seed round has ever approached $100 million.

The opportunity lies in the application layer. As the AI infrastructure stack matures through APIs and open-source models, competitive advantage shifts from “who has the most GPUs” to “who understands the local problem best.” Healthcare, logistics, energy, agriculture — these are domains where local knowledge and market access matter more than training compute.

The strategic response for emerging market founders is not to compete with AMI Labs but to build on whatever paradigm wins. The first companies to apply world-model APIs to local industrial challenges will capture enormous value.

The Risk Factors

JEPA is a research agenda, not a proven commercial architecture. Meta’s own JEPA research produced impressive academic results but has not demonstrated clear commercial superiority over transformer-based approaches. The $1.03 billion gives AMI Labs perhaps 3-4 years of runway. If JEPA does not produce commercially viable results in that window, the next fundraise faces brutal valuation pressure.

The concentration of capital in AI raises sector-wide risk. When AI absorbs the majority of venture capital, climate tech, biotech, and fintech compete for a shrinking pool. If the AI thesis stumbles — scaling limits, regulatory constraints, slower enterprise adoption — the concentrated bets could produce a correction rivaling the dot-com crash.

And the talent market is overheated. AMI Labs, OpenAI, Anthropic, and Google DeepMind compete for the same roughly 2,000-3,000 researchers who can do frontier work, with compensation frequently exceeding $5 million annually. The result is a self-reinforcing cycle that concentrates AI capability in a shrinking number of organizations.

Sources & Further Reading

• Yann LeCun’s AMI Labs raises $1.03B to build world models — TechCrunch
• Yann LeCun’s New AI Startup Raises $1 Billion in Seed Funding — Bloomberg
• Turing Winner LeCun’s New ‘World Model’ AI Lab Raises $1B In Europe’s Largest Seed Round — Crunchbase News
• Massive AI Deals Drive $189B Startup Funding Record In February — Crunchbase News
• Yann LeCun’s new venture is a contrarian bet against large language models — MIT Technology Review
• Anthropic closes $30 billion funding round at $380 billion valuation — CNBC