Feather Robotics has raised $7.6 million in pre-seed funding to scale a modular humanoid robot platf...

Feather Robotics has raised $7.6 million in pre-seed funding to scale a modular humanoid robot platform aimed at third-party physical-AI developers, with Gradient leading the round alongside Builder Capital, Geometry, Seed Innovations and Virgo VC.

The California company, founded by former Google and Tesla engineers Hoa Mai and Parsa Bakhtiari, is positioning its wheeled bimanual robot as a developer substrate rather than a closed system. At $29,990 the unit undercuts listed prices for Unitree's H2 Edu (~$68,000), Dexmate's Vega (~$83,000) and Agibot's G2 (~$135,000). Feather swaps legs for a multi-directional wheeled base to cut mechanical complexity and failure points, claiming up to 10 hours of battery life, roughly one meter of reach, adult-human-comparable payload, replaceable modular components and an SDK in every unit. The company says it built its first prototype in two months, fulfilled its first order within nine months and has passed $1 million in revenue from manufacturing and food-service customers, including a manufacturer with $4 billion in annual revenue. Its explicit thesis: as embodied AI models proliferate, developers will need hardware they can run those models on without building robots themselves.

The pitch matters because the physical-AI stack is bifurcating. One cluster is vertically integrated humanoids (Tesla, Figure, the Chinese consumer-oriented units) where hardware and model ship together; another is a horizontally layered market where a foundation model for embodied control runs on commodity-ish platforms, much as CUDA-era developers once needed generic GPUs rather than bespoke silicon. Feather is betting on the second cluster, and its own founding story is the proof point — Mai says she searched for a hardware platform with viable unit economics, reliability and support logistics and found none. That is also the risk: a $7.6M pre-seed is a rounding error against the capital intensity of humanoid manufacturing, and wheels sidestep rather than solve the hard legged-locomotion problem that defines many enterprise deployments, from stairs to uneven factory floors.

The concrete implication for builders is that the cost of an embodied-AI dev kit is now under $30,000 with revenue already proven at small scale, which lowers the experimentation bar for labs and startups that previously had to choose between $68K-plus research units and rolling their own. For investors, the round is a test of whether a hardware-layer platform play can attract enough third-party software to compound, or whether value accrues to the model layer and the vertical integrators on either side. Watch whether Feather converts its manufacturing and food-service pilots into recurring fleet revenue before the next raise, and whether that SDK produces third-party applications that other robot vendors cannot match.