The next major AI race may be happening almost entirely behind closed doors. World models, which are AI systems designed to understand, predict and sometimes generate representations of physical environments, have become one of the industry’s most heavily funded areas.
Yet some of the biggest companies in the sector are revealing surprisingly little about what they ultimately plan to sell.
The clearest example is AMI Labs, the startup co-founded by former Meta chief AI scientist Yann LeCun. The company raised $1.03 billion in March at a reported $3.5 billion pre-money valuation to develop AI that learns from real-world data like video and sensor information.
But AMI is very vague about its commercial roadmap. Its vice president of world models, Michael Rabbat, recently said the company is still in a research and building phase and is not publicly discussing product plans or timelines.
Fei-Fei Li’s World Labs is further along publicly. The company raised another $1 billion in February and released Marble, a system that is capable of generating persistent 3D worlds from text, images and video. World Labs says the technology could eventually support areas ranging from creative production and robotics to scientific discovery.
Even so, the sector is unusually guarded. TechCrunch reported that Physicl, a company supplying data to world-model developers, does not always know exactly how its customers intend to use the datasets it provides. That secrecy matters because training world models can require very different data from the enormous text corpora used for large language models, including video, spatial information and sensor data.
Money is still pouring into the field regardless. In September, the Financial Times reported that newly formed DeepMind spinout Emulate was nearing a funding round of up to $700 million at a $3.7 billion valuation. Industrial world-model startup Noetive also emerged from stealth with a $41 million seed round.
Why Are World Model Companies Staying Quiet?
One explanation is competitive advantage. A world model that can reliably predict how environments change, reason spatially and plan actions could become valuable infrastructure for robotics, autonomous machines, manufacturing, games and simulation. Revealing a promising application too early could simply tell well-funded rivals where to concentrate.
There is another possibility: many of these systems may still be research projects searching for their strongest commercial use. World models cover an unusually broad technical category, which means that a company could begin with general spatial intelligence before deciding whether robotics, entertainment, industry or another sector offers the best path to market.
That makes world models one of AI’s strangest investment stories. Billions of dollars are flowing toward technology that investors clearly believe could become important, even while outsiders still know relatively little about the final products.
For now, the mystery may be part of the strategy. The bigger question is what happens when these companies finally show exactly what their models can do.
