PHYMI raises seed funding for physical AI agents

THE ESSENTIALS

PHYMI raised nearly $100 million in a seed round led by IDG Capital, according to investor Yunqi Partners.
Founder Nianqiu Liu previously held a technical leadership role at DeepRoute.ai.
The company describes a system combining models, a robot body, real-world data and human-agent coordination.

A large seed round for an early physical-agent company

PHYMI has raised close to $100 million in seed financing, according to an announcement carried by participating investor Yunqi Partners. IDG Capital led the round, with Yunqi, DiDi, Fosun RZ Capital, Glory Ventures and Hesai among the participants reported by TechNode.

Founded in 2026, the company is led by Nianqiu Liu, previously a technical partner and board member at DeepRoute.ai. The stated use of funds covers research, real-world data infrastructure, product engineering and team growth. The announcement does not name a shipping robot, a sale price or a commercial deployment schedule.

Four parts of the proposed system

The investor account describes PHYMI BRAIN as the model layer for perception, world modeling, planning and action generation. BODY covers the physical machine, including movement, sensing, computing, energy and reliability. DATA collects observations and task feedback. LINK connects goals, permissions, status and feedback across people and machines.

These are descriptions of an architecture and research program. They are not four independently benchmarked products. Their significance is that the company is taking responsibility for the full loop between an instruction and a physical result, rather than presenting a model as sufficient on its own.

A changing environment is the hard part

A robot executing a fixed movement can be evaluated against a known path. A robot asked to complete an open-ended goal has to notice when the surroundings change and decide whether its earlier plan still applies. Moving to the right place and manipulating an object are also coupled: an object reached from the wrong position may be impossible to grasp.

The investment case therefore rests on sustained task completion, not just a short demonstration of walking or picking something up. The public account supplies no independent success rate, intervention frequency or operating-cost comparison. The next useful evidence would be repeatable tasks in a disclosed environment, showing failures and human assistance as well as successful runs.