The Dark Forest of World Model Startups

At the recent All In conference, attempts to pin down clear product roadmaps for world models—systems designed to automate spatial reasoning—met with guarded responses. Michael Rabbat, co-founder of AMI Labs, maintains that the firm remains strictly in a research phase, declining to share timelines or specific commercial goals. This reticence is industry-wide: even data suppliers like Physicl CEO Alex de Vigan report being left in the dark about the end-use of their own products, complicating the development of specialized training sets.

The versatility of the technology creates a strategic paradox. Because world models could theoretically power anything from humanoid robotics and autonomous vehicles to high-end CGI rendering, companies are hesitant to signal their specific focus. Publicly committing to a niche like biomedicine or robotics would immediately invite scrutiny and potential market capture from well-funded rivals like OpenAI or Anthropic. By remaining opaque, these labs avoid attracting attention, effectively treating their development process as a dark forest scenario where silence is the only defense against premature competition.