The Real Bottleneck Isn’t Hardware—It’s the Software That Makes Robots Actually Work
Fiveages, a Beijing-based embodied AI startup, has closed several hundred million chinese yuan (~$50M+) across six rounds in 12 months including Pre-A (led by Sequoia China). From an initial valuation of less than 100 million RMB (US$15M), it has now surged about 20 times to about 2 billion RMB (US$280 million).
But the headline isn’t the capital.
It’s the validation: Fiveages is now Unitree Robotics’ “brain supplier” for embodied intelligence—powering the brains behind China’s highest-volume humanoid shipments.
The Tech Edge: Few-Shot Learning That Actually Works
While competitors burn millions on simulation, Fiveages’ FAM (Few-shot Action Model) achieves 97% task success on international benchmarks using minimal human demonstration data.
It translates abstract actions into predicted visual outcomes—letting robots “see” the result before moving.
Commercial Proof: From Factories to Unitree’s Supply Chain
Fiveages isn’t selling demos. It’s shipping production-grade AI:
- Unitree partnership: Integrated FAM into G1 humanoids for industrial bin handling; deployed in power grid inspection with state-owned utilities
- Factory collaboration: Robots now operating in a microwave oven factories across 100+ SKUs and dozens of production lines
Critically, Fiveages builds its own wheel-based dual-arm robot—not to compete with Unitree, but to co-develop hardware-software co-design that accelerates real-world iteration.

Why Sequoia Jumped In Now
As one Sequoia partner noted off-record:
“Everyone’s building bodies. Few can build brains that work on Monday morning.”
Investment Takeaway: The Stack Is Inverting
The humanoid value chain is flipping:
- Tier 1 (highest margin): Embodied AI models that generalize across robots
- Tier 2: Actuators, sensors, batteries (commoditizing fast)
- Tier 3: Full-body integration (low-margin, capital-intensive)
Fiveages sits at Tier 1—with a 20x valuation jump in 12 months to prove it.
But risks remain:
- Customer concentration: Heavy reliance on Unitree for near-term revenue
- Competition: AgiBot, Physical Intelligence, and Skild AI all racing similar paths
For investors, the signal is clear:
The next wave of robotics winners won’t be the ones with the flashiest hardware.
They’ll be the ones whose software makes every robot—regardless of body.


