The artificial intelligence industry is currently locked in a high-stakes rhetorical race, with leading tech companies routinely vying to market their latest systems as milestones of "artificial general intelligence" (AGI) or "superintelligence." Yet Alexandre LeBrun, CEO of AMI Labs—the high-profile world model startup co-founded by Turing Award laureate Yann LeCun—is noticeably absent from this linguistic arms race. In an industry defined by hyperbole, LeBrun and his team have chosen a radically different path, dismissing these ubiquitous industry buzzwords as vague, unscientific, and ultimately unhelpful markers of progress.
This calculated rejection of tech industry dogma frames the early strategy of AMI Labs, a pre-product venture that recently secured a staggering $1.03 billion in funding at a $3.5 billion pre-money valuation. While the company has no commercial product to sell yet and refuses to commit to a formal public launch timeline, its leadership is already laying the operational groundwork for the next evolution of machine intelligence. That groundwork recently brought LeBrun to Seoul, South Korea, for the International Conference on Machine Learning, where he scouted potential industrial partners, manufacturing giants, and academic researchers to help bridge the gap between abstract algorithmic theory and the physical world.
The Limitations of Language and the Rise of World Models
To understand LeBrun’s skepticism toward terms like AGI, one must examine the foundational technical divide between traditional Large Language Models (LLMs) and the physics-aware systems AMI Labs is attempting to build. LLMs, which power modern generative AI breakthroughs, operate fundamentally on statistics and token prediction. They predict the next word in a sequence based on vast textual corpora. While highly efficient at processing human language, code, and conceptual synthesis, LeBrun argues that LLMs are profoundly limited when forced to interact with physical reality.
Drawing an analogy to human neurology, LeBrun compares LLMs to the language and symbolic processing centers of the brain, while casting world models as the brain’s intuitive mechanism for understanding physical cause and effect. If a person nudges a glass off the edge of a table, human intuition instantly calculates that the glass will fall, impact the floor, and potentially shatter. This intuitive physics engine is precisely what a world model aims to replicate in artificial intelligence.
"We never used the word AGI," LeBrun said during an interview in Seoul. "And I just noticed that nobody is using it anymore; they switched to superintelligence. Next time we’ll switch to something else. There’s no good definition. What is superintelligence? I don’t know. It’s not a very useful word."
Despite moving away from these labels, LeBrun does not position world models as replacements for LLMs. Instead, he views the two architectures as entirely complementary. Where LLMs excel at linguistics and symbolic reasoning, world models provide the crucial context, situational awareness, and environmental grounding that current AI systems desperately lack.
The Physical AI Deficit: Why Today’s Robots Lack a Brain
The most immediate application for world models lies within the field of robotics. Despite rapid, highly publicized advancements in robotic hardware—ranging from bipedal humanoid frames to agile industrial arms—LeBrun contends that the underlying software remains fundamentally stagnant. Current robots execute rigid, pre-programmed routines in controlled environments, exhibiting what he characterizes as being "really dumb in the physical world."
To illustrate the danger of deploying context-blind AI into unpredictable human environments, LeBrun points to recent high-profile incidents, such as a public demonstration where a dancing, kung-fu-performing robot approached and unexpectedly kicked a child. Incidents like these underscore a stark reality: while hardware innovation has accelerated dramatically over the past several months, the cognitive software required to ensure safe, context-aware interaction simply does not exist yet.
"The hardware is very advanced; progress in hardware in the last few months is incredible, but there’s no brain," LeBrun noted.
This deficit becomes even more glaring when comparing artificial intelligence to human expertise in complex domains. LeBrun, who previously founded the AI healthcare startup Nabla, likens current medical AI models to a physician who has memorized every medical textbook in existence but has never completed a clinical residency. While LLMs can assist with certain administrative or diagnostic tasks, they address only a fraction of real-world healthcare challenges, the vast majority of which require experiential, physical context.
In industrial settings, fixed manufacturing robots operate reliably because their environments are meticulously controlled and predictable. However, the moment a robot is introduced into an open, unstructured environment—such as a household kitchen, a busy urban street, or a dynamic hospital corridor—its limitations become hazardous. "Robots are not safe right now," LeBrun stated plainly. "There’s no solution for that today."
The Chronology of AMI Labs and the Billion-Dollar Bet
The genesis of AMI Labs represents one of the most significant realignments in modern artificial intelligence research. The company emerged from the shifting landscape of foundational AI research, catalyzed by Yann LeCun’s departure from Meta, where he served as Chief AI Scientist and long championed world models as the most viable path toward human-level machine intelligence.
The timeline of AMI Labs underscores the unprecedented velocity of venture capital deployment in the current AI economy:
- Late 2024 to Early 2025: Yann LeCun pivots his academic focus and industry efforts toward establishing world-model-centric architectures, separating from Meta’s immediate product cycle to form independent research initiatives.
- January 2026: AMI Labs officially steps into the public spotlight, drawing intense industry scrutiny as a premier contender in the race to build physical AI foundations.
- March 2026: AMI Labs closes a massive $1.03 billion funding round at a $3.5 billion pre-money valuation, cementing its status as one of the most well-capitalized early-stage startups in tech history.
- June 2026: The South Korean government announces a sweeping national initiative to mobilize approximately $880 billion toward semiconductor manufacturing, AI data infrastructure, and physical AI integration.
- July 2026: CEO Alexandre LeBrun travels to Seoul for the International Conference on Machine Learning, signaling the company’s intentional pivot toward Asian hardware and manufacturing ecosystems.
Despite holding a war chest exceeding one billion dollars, AMI Labs maintains a deliberately opaque product roadmap. When questioned about timelines for a commercial release or developer preview, LeBrun remains tight-lipped. "We’ll make a surprise when we’re ready," he said.
Looking East: Why AMI Labs is Eyeing South Korea
A central objective of LeBrun’s recent international outreach is addressing a fundamental limitation of world model development: training data cannot be synthesized exclusively inside a cleanroom laboratory. Because world models must learn the physics of reality, they require deep integration with real-world physical environments, sensors, and hardware systems.
This necessity is what draws AMI Labs toward Asia, and specifically South Korea. According to LeBrun, the strategic pull toward the region rests on two distinct pillars: an unmatched industrial foundation in robotics, semiconductors, and manufacturing, and a national culture characterized by technological agility.
"Korea was the fastest adopter of the internet 25 years ago," LeBrun observed, highlighting the country’s historical propensity for rapid digital transformation. This combination of heavy industrial manufacturing—sectors largely untouched by the initial text-and-image wave of generative AI—and a societal readiness to embrace emerging technologies creates a uniquely fertile ground for physical AI deployment.
JP Lee, CEO of venture capital firm SBVA and a prominent Asian backer of AMI Labs, has been actively advising LeBrun and his leadership team to embed themselves within the South Korean ecosystem. Lee emphasizes that while the South Korean government has successfully funded sovereign Large Language Models that function adequately for general-purpose text tasks, national technology leaders recognize that physical AI must develop in parallel.
This philosophy aligns directly with Seoul’s sweeping national infrastructure commitments announced mid-year, which direct massive capital expenditures toward bridging the gap between digital computing and physical robotics. For foreign startups like AMI Labs, South Korea offers not just advanced silicon and hardware components, but an agile developer community historically quick to adapt foreign tools to local industrial applications—a pattern that previously birthed powerful domestic tech giants like Naver and Kakao.
Industry Implications and the Road Ahead
The trajectory of AMI Labs highlights a broader, fundamental shift in the artificial intelligence sector. As the marginal utility of scaling text-based LLMs begins to plateau, venture capital and top-tier research talent are increasingly pivoting toward embodied AI—systems that can perceive, reason about, and physically manipulate the real world.
By bypassing the marketing frenzy of "AGI" and "superintelligence," AMI Labs is attempting to ground the conversation in rigorous engineering principles. Whether world models will successfully bridge the cognitive chasm between digital algorithms and physical safety remains to be seen. However, with over a billion dollars in backing, the endorsement of premier AI pioneers, and active strategic positioning within the world’s most advanced hardware ecosystems, AMI Labs is well-positioned to dictate the terms of the next major technological revolution—on its own terms, and without the need for catchy buzzwords.
