Product1 publisher3 min readPublished
World model labs leave their own data supplier guessing at the spec
AMI Labs says it is still in a research and building phase and will not discuss product plans or a timeline. World Labs' Marble ships now as a way to build explorable scenes. A roadmap needs more than that.
The Product Desk · Product desk

What happened
- TechCrunch names Yann LeCun's AMI Labs and Fei-Fei Li's World Labs as the field's big players, with plenty of buzz and funding and a low ranking on what it calls the trying-to-make-money scale.
- Physicl CEO Alex de Vigan, who supplies data to world model labs, said he wishes they would tell him more because he could build more useful data if he knew what they were working on.
- AMI has already dipped a toe into manufacturing, biomedicine, robotics and AI software for doctors through its Nabia partnership; it declined to say which of them it will pursue.
Compiled by The Product DeskSomething wrong?How this is made
Why it matters
- constraint A robotics or autonomy plan cannot take a dated dependency on a supplier that has declined to name a product or a date, so the work has to be funded as research or built on something procurable.
- cost The secrecy pushes cost onto the supply chain: a data vendor with no spec builds broadly and eats the waste, and the labs pay for that indirectly in less useful training data.
- decision Anyone evaluating Marble has to decide whether they are buying an environment-building tool for media work now or an option on robotics later, and the demos only settle the first question.
- precedent While capital stays cheap for these labs, staying quiet costs them nothing and delays rivals, so buyers should expect the disclosure gap to persist rather than narrow on its own.
Alex de Vigan sells training data to world model labs and does not know what they use it for. "I wish they would tell us more. We could build more useful data if we knew what they were working on," the Physicl CEO told TechCrunch [10]. He says the data has been useful, and he does not know what for [9].
A supplier working without a spec builds for every case. Building for every case is expensive. The first practical consequence of the secrecy lands on a vendor.
The idea being sold is broad. World models automate spatial intelligence, and TechCrunch lists robotics, interactive video and harder self-driving problems as directions the field could take [12]. The thing a team can actually buy is Marble, which the column calls probably the most fully developed product in the space, with demos covering media creation, explorable environments for video games and CGI effects [7]. Robotics use cases are in there too, but the column's assessment is that the platform seems more designed to demonstrate capabilities [8].
AMI Labs has dipped a toe into manufacturing, biomedicine, robotics and AI software for doctors through its Nabia partnership [11]. Four application areas, and AMI has put no product plan or timeline on any of them [17]. Michael Rabbatt, a co-founder and the company's VP of World Models, was asked on the panel what AMI is building and said: "We'll talk about it when we're ready to talk about it" [3][4]. Over email he added: "We're still in a research and building phase, so we're not talking publicly about any product plans or timeline" [5]. AMI is less than a year old [6]. Funding figures for the two labs are not in the column.
TechCrunch also gives the reason the silence holds. If AMI announced a humanoid product or a next-generation Hollywood rendering system, other world model companies, the neolabs, OpenAI and Anthropic would all have reason to move in [14]. And while fundraising stays easy, nothing forces a lab to pick one business [16]. The column calls this a Dark Forest scenario, after Cixin Liu [15].
Two things sort this for anyone holding a 2027 line item: whether the vendor names something you can buy now, on terms, and whether the plan needs it on a date. Both yes is ordinary procurement. Named product, no hard date: the pilot is scoped to exactly what the demos do, which for Marble means environments and effects [7]. No product and no date: research funding, with nothing expected to ship. No named product but a fixed date is the cell that breaks, because the plan is resting on a company whose co-founder said it will talk when it is ready [4].
What is on the record is one shipping product whose demos are media work, one lab that will not name a product or a date, and a data supplier who would like to know what to build [7][5][10].
What to watch
- Whether AMI names a first application area or ships anything through the Nabia partnership with doctors.
- Whether World Labs starts selling Marble to robotics teams with support terms attached, instead of showing robotics as a demo.
- Whether suppliers like Physicl begin receiving specs. A spec would mean a path to market has been chosen internally.