World Model Firms Guard Secrets

Last week, I facilitated a discussion on world models at the All In conference (unrelated to the podcast), offering a deep dive into one of AI’s most enigmatic sectors. Industry leaders like Yann LeCun’s AMI Labs and Fei-Fei Li’s World Labs have generated significant buzz and secured substantial funding, yet their progress toward monetization remains notably slow.

At their foundation, world models aim to automate spatial intelligence, opening doors to diverse and potentially lucrative applications ranging from robotics and interactive video to advanced autonomous driving systems.

However, when I probed for concrete commercialization timelines, the answers grew vague. The most authoritative voice on the panel was Michael Rabbatt, co-founder and VP of World Models at AMI Labs. When pressed on specific projects, he remained evasive, stating, “We’ll talk about it when we’re ready to talk about it.” In a follow-up email, he clarified, “We’re still in a research and building phase, so we’re not talking publicly about any product plans or timeline.”

Given that AMI is less than a year old, this discretion is understandable. Yet, this opacity permeates the broader world-modeling sector. World Labs’ Marble appears to be the most mature product available, with demos showcasing straightforward media creation, explorable video game environments, and CGI effects. While robotics applications exist, the platform seems primarily designed to demonstrate technical capabilities rather than deliver immediate commercial value.

This secrecy even extends to upstream suppliers. At the same conference, I spoke with Alex de Vigan, CEO of Physicl, a data provider for the emerging world model industry. He confirmed that Physicl’s data is being used effectively but admitted he remains unaware of the specific applications. “I wish they would tell us more. We could build more useful data if we knew what they were working on,” de Vigan explained.

Part of the ambiguity stems from the inherent versatility of world models. At their simplest, they function as navigable maps of the physical world, akin to the AI systems powering self-driving vehicles. However, the same modeling techniques that enable a Waymo vehicle to navigate traffic could also assist humanoid robots in handling cargo or transform short video clips into interactive environments. AMI has already explored sectors including manufacturing, biomedicine, robotics, and medical AI through its Nabia partnership. While it is unlikely they will pursue every avenue, one or two core applications may eventually emerge as priorities.

There is no doubt that viable businesses can be built on world model technology. As long as fundraising remains straightforward, there is little incentive to narrow focus. In fact, maintaining a broad approach makes strategic sense. If AMI were to announce the development of a humanoid robot or a next-generation Hollywood rendering system tomorrow, numerous other labs would immediately pivot to compete. Soon, the lab would face potential competition from rival world model companies, emerging startups, and tech giants like OpenAI and Anthropic.

This dynamic is essentially the flip side of easy fundraising. Competitors can raise capital as well—the same financial flexibility that allows you to operate under the radar also funds potential rivals once the market path becomes clear. Even if competition is inevitable, delaying it as long as possible is advantageous, which means keeping specific product details under wraps.

For fans of Cixin Liu, this mirrors the Dark Forest theory: if you cannot identify other players in the woods, it is safest to avoid drawing attention.