The world model cooled down very quickly in the second half of the year.
According to incomplete statistics, there were only 5 primary market financings related to physical AI and world models in October. Investors have raised the bar, and there are only a handful of projects that leading funds are willing to invest in. The valuation premium supported by concepts is shrinking rapidly.
The money hasn’t stopped, it’s just that the flow has changed. This year, there are still teams such as Inverse Matrix, Natural Will, Wuyuan Technology, LiberAI, and Ant Lingbo, which have received hundreds of millions to over 100 million US dollars in financing. These companies have two things in common: they have research backgrounds from top universities and have released model results.
The team that only makes PPT can't get any money, but those who have something in hand will be better off.
There is a detail that is easily overlooked: the official financing announcement is generally about three months later than the actual delivery. The cooling we are seeing now actually corresponds to the summer pace of investment; for the large financing just announced in the news, the money may have already arrived. Relying on the popularity of the news to judge whether the track is warm or not, it’s natural to slow down.
The old problems are still there. There is no unified standard in the industry for how to define and evaluate world models, and technical routes, product forms, and business models are all being explored.
What best illustrates the state of the industry is mergers and acquisitions. World Labs, founded by Li Feifei, plans to be acquired by AMD for US$8.2 billion, and Amazon is also in talks to acquire Decart for about US$7 billion. It is much faster for a large company to buy directly than to wait for a startup to grow up slowly. This is not good news for the teams queuing up for financing later.