Current Status and the Structural Leaderboard Block
AMI Labs (Advanced Machine Intelligence) was incorporated in late 2025 by Yann LeCun and Alexandre LeBrun, closing a massive $1.03B seed round in March 2026 2 sources. The lab is explicitly committed to non-LLM, JEPA-style world-model architectures over sensor data for planning, with LeCun characterizing current generative LLMs as a "dead end" 3 sources. Consequently, resolution via a top-5 flagship leaderboard placement is nearly structurally blocked. The Artificial Analysis Intelligence Index is an explicitly text-only, English-language LLM evaluation suite (weighting agents, coding, scientific reasoning, and general text tasks) currently dominated by Anthropic and OpenAI variants artificialanalysis.ai. Unless AMI drastically pivots its research or the flagship leaderboard fundamentally reorients around embodied world models, the lab will not appear on this index in the foreseeable future.
The Expert Recognition Path and Category Mismatch
This leaves expert recognition of an unreleased or unlisted system as the only realistic path to resolution. However, this criterion requires the model to be broadly regarded as comparable to contemporaneous top general models from OpenAI, Anthropic, or Google. This creates a severe category mismatch: an expert can easily label a JEPA system "state-of-the-art for physical-world prediction," but the resolution demands comparability to frontier general or agentic capabilities. Even on AMI's own optimistic schedule, industrial prototypes are hoped for in 2027 and "fairly universal intelligent systems" within three to five years (~2029-2031) 2 sources. Achieving physical-world understanding on the order of a "cat or a rat" by 2030 nebius.science will clearly fall short of the frontier comparability standard against 2030-era OpenAI or Google general-purpose flagships.
Capital, Compute, and Acquisition Risk
While $1.03B is an enormous seed round, it represents roughly the capital of a single contemporary frontier training run and is an order of magnitude short of the multi-billion-dollar, multi-hundred-megawatt scale of current leading competitors 3 sources. Even with extreme architectural efficiency — such as early JEPA models running on a fraction of frontier GPU counts financialexpress.com — matching 2029+ Anthropic or OpenAI systems would demand massive follow-on funding. Leaked materials suggest AMI is already seeking additional capital sifted.eu, meaning prolonged scaling is contingent on continually open capital markets. A funding contraction would hit a pre-revenue research lab heavily, driving up the likelihood of an acquisition by strategic backers (e.g., Nvidia or Samsung) or a pivot to a profitable but non-frontier robotics or medical niche.
Net Assessment and Tail Risks
Because of the steep definitional requirements and the hurdles to commercialization, there is roughly a 40-50% chance this event never resolves. The earliest realistic scenario (late 2029 to early 2030s) assumes an early, highly efficient architectural breakthrough in world models that experts validate as a frontier equivalent. A more plausible success window falls in the mid-to-late 2030s, assuming AMI successfully scales its approach over several funding rounds and the gap between "world models" and "general AGI" narrows. The upper percentiles are pushed deep into the future to appropriately represent the heavy "never" tail, capturing the substantial risk that AMI either winds down, is acquired, or remains permanently specialized and categorically mismatched with the established AI frontier.