Key takeaways
- World model startups including AMI Labs and World Labs are deliberately avoiding disclosure of their commercialization plans, citing research and building phases with no public timelines.
- The versatility of world models — applicable to robotics, video generation, autonomous driving, and visual effects — makes silence strategically valuable because announcing any single focus area would trigger immediate competitor mobilization.
- Even data suppliers serving the world model industry lack visibility into what these companies are building, limiting their ability to optimize products for known commercial objectives.
- The easy fundraising environment removes pressure for these companies to focus on any single market, allowing them to maintain ambiguity while rivals face a 'dark forest' competitive scenario with unclear targets.
The world model space is rapidly becoming one of AI’s most opaque territories. Last week at the All In conference, I moderated a panel on the technology alongside Michael Rabbat, co-founder and VP of World Models at Yann LeCun’s AMI Labs. When I asked Rabbat directly about what his company was actually building, the answer was carefully non-committal: “We’ll talk about it when we’re ready to talk about it.”
By email, he elaborated slightly: “We’re still in a research and building phase, so we’re not talking publicly about any product plans or timeline.” AMI Labs itself is less than a year old.
This evasiveness extends far beyond a single company. Both AMI Labs and Fei-Fei Li’s World Labs have accumulated substantial funding and industry buzz, yet neither has articulated a clear path to commercialization. Even more striking, it affects the entire supply chain. Alex de Vigan, CEO of Physicl — a data company serving the world modeling sector — told me he knows his company’s data has been useful to the major players, but he remains in the dark about what they’re actually building. “I wish they would tell us more,” he said. “We could build more useful data if we knew what they were working on.”
What Are World Models, and Why The Mystery?
At their foundation, world models are systems designed to automate spatial intelligence. The core idea is relatively straightforward: train a model to understand and predict how physical space behaves, enabling machines to reason about movement, interaction, and change in the world around them.
The versatility of this approach is precisely what makes the secrecy strategy so effective. World models can power several completely different products and markets. The simplest application is a navigable map of the world — similar to the spatial reasoning systems that autonomous vehicles use to navigate traffic. Yet the same underlying modeling approach could help humanoid robots understand how to manipulate objects, transform a few minutes of video footage into an interactive 3D environment for video games, or generate photorealistic CGI effects for film and television.
This technological flexibility means a company could plausibly pivot toward any of several markets, each with its own timeline, maturity level, and competitive dynamics. The choice of where to focus could be worth billions in value — or could determine whether the company ends up competing directly with OpenAI, Anthropic, or hardware startups building robotics systems.
The Known Products and Capabilities
World Labs’ Marble stands as the most mature product in the space so far. In public demonstrations, Marble has shown capability across several distinct applications: media creation, generation of explorable environments suitable for video games, and production of CGI-quality visual effects. The platform appears deliberately positioned to showcase technical capabilities rather than to serve as a finished commercial product ready for specific enterprise customers.
AMI Labs has taken a notably different approach by exploring multiple vertical markets simultaneously. The company has already dipped into applications spanning manufacturing, biomedicine, robotics, and AI software for physicians — the last through a partnership with Nabia. This diversified exploration could indicate genuine multi-market potential for the underlying technology, or it could simply be a way to keep investors, potential partners, and stakeholders engaged while the company determines which direction offers the strongest business opportunity and competitive advantages.
The Strategic Calculation Behind Silence
Avoiding Premature Competition
The moment AMI Labs announced it had built an advanced humanoid robot, a commercial autonomous driving system, or a next-generation rendering platform for Hollywood studios, the entire competitive landscape would shift instantly. Other well-funded world model companies would immediately adjust their research priorities toward that same market. Investors would double down on funding similar efforts. The neolabs and established AI giants like OpenAI and Anthropic would recognize a clear market opportunity and mobilize their own resources.
Timing as Competitive Advantage
In that scenario, AMI would face competition from World Labs, from newly funded competitors, and potentially from the largest AI companies on the planet — all pursuing similar applications simultaneously. Speed to market would become critical, and the company’s current head start would matter far less. Differentiation would hinge on execution rather than discovery.
Fundraising Without Accountability
By staying quiet, AMI and World Labs can continue building in relative calm, without prematurely triggering a competitive surge. The current fundraising environment removes pressure for these companies to lock into a single commercial direction. Capital flows readily to frontier AI companies regardless of whether they have demonstrated revenue or a clear path to profitability. As long as the technology’s commercial applications remain ambiguous, competitors lack a clear target to chase. The fundraising tap stays open without explicit pressure to monetize. The window for establishing dominance in whichever market the company eventually chooses remains wider.
The Data Supplier Dilemma
The secrecy extends to the entire ecosystem supporting world model development. Companies like Physicl are building specialized datasets to support world model training and evaluation. These suppliers have reason to believe their work is valuable — they’re being actively used by the major players in the space — yet they operate almost entirely without visibility into what the end products will actually be or which applications the major labs are prioritizing.
This creates an obvious problem: data companies could potentially build far more targeted and useful datasets if they understood the actual commercial objectives their customers were pursuing. Instead, they’re forced to operate with incomplete information, building general-purpose world modeling datasets and hoping they align with whatever the major labs eventually decide to build. The major companies’ reluctance to share product roadmaps with even their critical suppliers suggests how seriously they’re taking the competitive calculus of staying hidden.
The Dark Forest Scenario
The secrecy strategy maps closely onto what science fiction author Cixin Liu described as a “dark forest” scenario. In such an environment, where visibility is limited and intentions are unclear, the safest strategy is to avoid revealing yourself. Every announcement of capability risks triggering a competitive response from rivals who don’t know your exact position but recognize the strategic value of the territory you’re staking out.
This logic only works as long as funding remains plentiful and there’s no external pressure forcing companies to commercialize quickly. The moment any of the major labs faces capital constraints or shareholder pressure to demonstrate revenue growth, the calculus changes entirely. Companies will need to pick a market and move fast, and all the other players will scramble to follow.
For now, the incentives all point toward silence. The world model companies can raise money without explaining what they’re building. Their data suppliers must work blind. And the competitive landscape remains frozen in a state of mutual strategic uncertainty — exactly the situation all the major players prefer.
Frequently Asked Questions
What are world models?
World models are AI systems designed to automate spatial intelligence, enabling machines to understand and predict how physical space behaves. The same underlying technology could power applications ranging from autonomous vehicles to humanoid robots to video generation and CGI effects.
Why are world model companies being secretive about their plans?
By not disclosing their specific commercial direction, companies delay triggering a competitive response from rivals. Once they announce a product focus, competitors would swiftly mobilize. Easy fundraising removes pressure to focus on any single market, making secrecy the optimal strategy.
How does this secrecy affect data suppliers like Physicl?
Data companies serving the world model industry must build datasets without knowing what applications their customers are ultimately targeting. This limits their ability to create optimally useful datasets tailored to specific commercial objectives.
