Star Map secures another 2 billion yuan in financing, accelerating the segmentation of the embodied intelligence industry

How does AI and embodied intelligence industry segmentation affect the prospects of startups?

After announcing the completion of nearly 1 billion yuan in Series B funding in February 2026, on April 2, Xinghai Tu once again disclosed nearly 2 billion yuan in B+ round financing, with investment institutions including industrial capital, long-term leading funds, national team funds, and other institutions. After this round, Xinghai Tu’s valuation jumped to a 20 billion yuan gradient.

Founded in 2023, Xinghai Tu mainly builds open-world embodied intelligence solutions driven by “whole machine + intelligence,” with its technology roadmap focusing on an embodied brain built from real data and world model construction.

At present, Xinghai Tu’s valuation of 20 billion yuan is in the same first tier as Galaxy General. The latter chooses simulation data as its main line, while Xinghai Tu chooses a real-scene embodied data engine, facing challenges such as difficult and costly data collection and slow expansion.

Bai Zongyi, founding partner of Yao Tu Capital, believes that enough real machine data is effective, but there are still gaps. The entire industry is evolving rapidly, but in terms of algorithms, there is still a very long time before general intelligence can be validated.

As for the brain, the Xinghai Tu team adheres to the homology and symbiotic relationship between the Vision-Language-Action (VLA) model and the World Action Model (WAM). Based on massive data from real open-world scenes, it can deeply couple intuitive cognition of the physical world with precise task execution. Up to now, Xinghai Tu has released three VLA models open source.

This year, embodied companies have entered a large-scale mass-production stage. For example, Zhiren Robotics disclosed that its 10,000th general embodied robot, Expedition A3, has officially come off the production line; UBTECH Walker S2 has been delivered in batches; and Songyan Power’s consumer-grade humanoid robot Bumi has entered the batch delivery stage, among others.

In response, Xinghai Tu’s CFO Luo Tianqi said that mass production in the industry includes data training ground scenarios aimed at developer groups, productivity scenarios, and entertainment and performance markets, as well as data training ground scenarios targeting the production of high-quality data. In 2026, the industry is still in the POC (proof of concept) first year. The core is to first run through 0 to 1, rather than simply pursuing mass production. A reporter learned that, currently, Xinghai Tu has not disclosed data such as production volume.

Luo Tianqi judges that the embodied industry chain is extremely long, and survival requires becoming a “hexagonal warrior.” Although there are still many new companies entering the track in the near term, he believes that in the future, among China’s top ten embodied intelligence companies, more than half will come from the current startup cohort.

Within this year’s embodied intelligence track, the gap in financing amounts and valuation figures between leading enterprises and mid-to-lower-tier enterprises is gradually widening. Bai Zongyi believes that elements such as models and data have not converged significantly. The entire industry is evolving rapidly, but there is still a very long time before general intelligence can be validated. There is a huge gap between long-term endgame outcomes and currently deployable commercialization scenarios. At present, institutions mainly observe first, and then take concentrated positions in high-quality teams.

(This article comes from Yicai )