World Model Funding Cools Sharply: Only 5 Related Deals in October as Capital Becomes More Selective

Beating AI News Flash: World models, which were eagerly sought after by capital in the first half of this year, have begun to cool down in the second half.

Quantum位 recently interviewed several frontline investors and startup financing heads and found that more and more world model projects are struggling even to secure term sheets (TS). The list of teams several investors are willing to continue betting on overlaps heavily, with only about four or five companies.

Incomplete statistics from Cailian Press Chuangtoutong show that in 2026 there have been more than 100 financing deals related to physical AI and world models, but only 5 in October.

Wang Sheng, a partner at Inno Fund, revealed that leading VCs began slowing their investment pace as early as May and June. The number of project closings fell successively in July and August. Because financing is usually officially announced with a delay of several months, many of the seemingly lively financing news in the second half of the year is actually a continuation of deals from the first half.

World models themselves also face technical implementation challenges. The concept still has no unified definition; generating realistic videos, simulating physical laws, and predicting robot actions are all called world models by some. World Labs, founded by Li Feifei, once divided it into three categories—renderers, simulators, and planners—in an attempt to distinguish different technical routes.

In May this year, a study tested 9 video generation models using 319 sets of physical condition comparisons, and the highest paired comprehensive score was only 51.7%. Even when the images generated by the models look realistic, they often fail to correctly reflect the effects of changes in conditions such as friction and object mass.

However, capital has not completely withdrawn. Quantum位 interviews found that a small number of projects with top research teams, technical achievements, or big-tech resources can still obtain large financing. Compared with the rush to chase concepts in the first half of the year, investors now place more emphasis on whether teams can deliver actual results.