Advanced Machine Intelligence (AMI), a newly formed artificial intelligence research and development firm based in Paris, officially announced on Monday that it has secured over $1 billion in initial funding. The startup, co-founded by Turing Award winner and former Meta Chief AI Scientist Yann LeCun, aims to pivot the trajectory of the artificial intelligence industry away from its current reliance on Large Language Models (LLMs) and toward what LeCun describes as "AI world models." This massive capital injection values the pre-product company at an estimated $3.5 billion, signaling intense investor confidence in a fundamental shift in AI architecture.
The funding round was co-led by a consortium of high-profile venture capital firms, including Cathay Innovation, Greycroft, Hiro Capital, HV Capital, and Bezos Expeditions, the personal investment vehicle of Amazon founder Jeff Bezos. The round also drew participation from influential individual backers such as billionaire entrepreneur Mark Cuban, former Google CEO Eric Schmidt, and Xavier Niel, the French telecommunications magnate behind Iliad and the Station F startup campus. The presence of such diverse and heavy-hitting investors underscores the perceived strategic importance of AMI’s mission to develop a more grounded, physically aware form of machine intelligence.
The Philosophical Shift: Moving Beyond Language
At the heart of AMI’s founding is Yann LeCun’s long-standing critique of the current AI status quo. For the past several years, the industry has been dominated by the success of Generative AI and LLMs, such as OpenAI’s GPT series and Anthropic’s Claude. However, LeCun has remained a vocal skeptic regarding the ability of these models to achieve Artificial General Intelligence (AGI) or even human-level reasoning.
In statements accompanying the launch, LeCun argued that human intelligence is not primarily linguistic but is instead grounded in an understanding of the physical world—a trait current AI lacks. "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," LeCun stated. He contends that language is a low-bandwidth medium that captures only a fraction of human knowledge, most of which is acquired through sensory observation and interaction with the 3D environment.
AMI’s objective is to build a "new breed" of AI systems. According to the company’s mission statement, these systems will possess persistent memory, the ability to reason and plan over long horizons, and a fundamental understanding of cause-and-effect in the physical world. Unlike LLMs, which predict the next token in a sequence based on statistical probabilities, world models are designed to internalize the underlying "physics" of their environment, allowing them to simulate outcomes and make informed decisions in real-time.
A Strategic Departure from Meta Platforms
The establishment of AMI marks the first major commercial venture for LeCun since his tenure at Meta, where he founded and led the Fundamental AI Research (FAIR) lab for over a decade. While LeCun will remain a professor at New York University, his transition to AMI follows a formal departure from Meta scheduled for November 2025.
The decision to spin this research into a private startup was born from a strategic divergence within Meta. As the "AI arms race" intensified following the release of ChatGPT, Meta—like its peers—reoriented its resources toward scaling LLMs to remain competitive in the consumer chatbot and advertising markets. LeCun noted that while Meta CEO Mark Zuckerberg remained supportive of world model research, the applications for such technology are primarily industrial and enterprise-focused, rather than consumer-centric.
"I told Mark I can do this faster, cheaper, and better outside of Meta," LeCun explained. He suggested that by operating independently, AMI could share the significant costs of development with industrial partners who have a direct stake in the technology’s success. Despite Meta not being an official investor in the $1 billion round, LeCun indicated that the two entities are in talks regarding future collaborations, particularly in the realm of augmented reality (AR) smart glasses, where world models could provide the situational awareness necessary for advanced digital assistants.
Chronology of Development and the "Meta Mafia"
The formation of AMI is the result of years of theoretical work conducted within academic and corporate laboratories. The timeline of this evolution highlights the growing friction between traditional generative AI and the "world model" school of thought:
- 2018: Yann LeCun, alongside Yoshua Bengio and Geoffrey Hinton, receives the Turing Award for their foundational work on deep learning.
- 2022: LeCun publishes a position paper titled "A Path Towards Autonomous Machine Intelligence," outlining the architecture for world models and the Joint-Embedding Predictive Architecture (JEPA).
- Late 2023: As the industry doubles down on "Scaling Laws" for LLMs, LeCun begins advocating for a shift toward non-generative AI that can learn from video and sensory data.
- November 2024: LeCun approaches Mark Zuckerberg to discuss the necessity of an independent entity to pursue industrial applications of world models.
- March 2025: AMI officially launches with a $1 billion seed round and a global footprint.
AMI is bolstered by a leadership team comprised of several high-ranking former Meta executives and researchers. Michael Rabbat, Meta’s former director of research science, joins the startup alongside Laurent Solly, the former vice president of Meta Europe, and Pascale Fung, a renowned AI researcher and former senior director at Meta. The company’s CEO, Alexandre LeBrun, previously led the AI healthcare startup Nabla, while Chief Science Officer Saining Xie brings expertise from Google DeepMind.
Industrial Applications and Supporting Data
While generative AI has found a home in creative writing, coding, and marketing, AMI is targeting high-stakes industrial sectors where precision and physical reliability are paramount. The startup has already announced plans to collaborate with global leaders in manufacturing, robotics, and biomedicine, including Toyota and Samsung.
One of the primary use cases cited by LeCun involves the creation of "digital twins" or realistic world models of complex machinery. For example, by building a world model of an aircraft engine, AMI could enable manufacturers to simulate thousands of hours of operation under varying conditions to optimize fuel efficiency, minimize carbon emissions, and predict mechanical failures before they occur.
The economic implications of this shift are significant. According to market analysis, the industrial AI sector is projected to grow at a compound annual growth rate (CAGR) of over 20% through 2030. While LLMs struggle with "hallucinations"—the tendency to generate factual inaccuracies—industrial world models must be grounded in physical reality to be useful. AMI’s approach seeks to address the $100 billion-plus market for autonomous systems, where the "black box" nature of current generative models is often a barrier to adoption due to safety and regulatory concerns.
The Ethics of Control and Open Source
A central tenet of AMI’s corporate philosophy is the commitment to open-source technology. LeCun has frequently argued that AI is too foundational to be locked behind the proprietary walls of a single corporation. This stance places AMI in direct contrast with the increasingly "closed" models of OpenAI and Google.
The debate over AI control has taken on geopolitical dimensions in recent months. The US Department of Defense recently moved to blacklist certain AI firms after they attempted to impose restrictive ethical "red lines" on military use. LeCun, while often critical of government overreach, maintains that it is not the role of tech CEOs to decide how society utilizes powerful tools.
"I don’t think any of us… has any legitimacy to decide for society what is a good or bad use of AI," LeCun remarked, specifically referencing peers like Sam Altman and Elon Musk. He argued that in liberal democracies, the democratic process—rather than corporate boards—should govern the technology. By releasing its models as open source, AMI intends to democratize access to high-level intelligence, ensuring that no single entity holds a monopoly on the "brains" of future autonomous systems.
Future Implications and the Universal World Model
The launch of AMI adds a powerful new player to the burgeoning AI ecosystem in Paris, which has already produced notable contenders like Mistral AI and Poolside. The city’s concentration of mathematical talent and favorable research tax credits have made it a global hub for AI development, often serving as a bridge between the American venture capital model and European regulatory frameworks.
In the short term, AMI plans to release a series of specialized models tailored to specific industrial tasks. However, the company’s long-term "moonshot" is the development of a "universal world model." This system would serve as a foundation for general intelligence, capable of being fine-tuned for any industry or physical environment.
If successful, AMI’s world models could solve the "robotics bottleneck"—the current difficulty in training robots to perform complex, non-repetitive tasks in unstructured environments. By providing robots with a "common sense" understanding of gravity, friction, and object permanence, AMI could catalyze a revolution in automated logistics and household robotics.
While the $1 billion investment is a staggering sum for a seed round, it reflects the high cost of the compute resources and elite talent required to challenge the LLM paradigm. As AMI establishes its offices in Paris, Montreal, Singapore, and New York, the tech industry will be watching closely to see if LeCun’s "world model" hypothesis can deliver on the promise of human-level intelligence where language-based models have, in his view, reached a plateau.
About the Author
