Feather Robotics Starts Selling Developer Robots, Claims $1 Million in Revenue

Robotics developers can now buy a machine built to run someone else’s AI rather than build the hardware themselves. Feather Robotics has begun selling its customizable robots in small quantities and says revenue has passed $1 million, TechCrunch reported. That is an early commercial test of Feather’s approach, not evidence yet that its robots can perform reliably across the jobs customers want them to do.

Sell the machine, leave room for the brain

Feather’s bet starts with a division of labor. It provides a robot and software toolkit that developers can adapt to a job, including by changing the length of its arms. Feather says the hardware can run models from robotics AI providers such as Generalist, Skild and Physical Intelligence. A customer does not have to accept one intelligence system chosen by the robot maker.

That differs from the approach TechCrunch describes at Tesla and Figure, which are pursuing both the robot body and a general-purpose AI brain. Feather co-founder Hoa Mai’s argument is that developers need a machine they can work on today, even if no broadly adaptable robot intelligence is ready. His company is selling a platform for others to build around, rather than promising to supply every application itself.

The price of a platform, not a finished worker

TechCrunch puts Feather’s robot at about $30,000, roughly half the price of Unitree’s H2 Edu. That price makes the hardware comparison concrete. It does not settle the cost of putting a robot to work: Feather’s customers must still choose or build the AI and application for their task. The room to customize is valuable precisely because the end use is not fixed when the machine leaves Feather.

Gradient Ventures general partner Darian Shirazi, whose firm led Feather’s previously announced $7.6 million pre-seed round, offered a more ambitious comparison: a business might buy a Feather robot instead of hiring a worker earning $50,000 to $60,000 a year. That is an investor’s pitch, not a demonstrated labor-cost saving. The purchase price alone says nothing about how well a configured robot completes a particular job.

Sales are real; the operating picture is thinner

Feather says its robots are cooking in restaurants in Japan and cleaning science labs. It did not name those customers. Nor did Mai’s account of small-quantity sales put a number on the machines in either setting. The reported revenue milestone shows that the company has begun to sell its approach; it cannot, by itself, show how often the robots complete a shift or what support those customers need.

Mai told TechCrunch that Feather had spent the past year testing robots in the field and had resolved almost all the issues it found. He said the company was preparing for a bigger product launch. Those are meaningful steps if the machines must work outside a controlled demonstration, but the larger launch remains a plan. How the robots perform for identifiable customers at greater scale is still the sharper test.

An ecosystem is the next bet

Mai expects thousands of companies building physical-AI applications to exist within five years. Feather wants to supply the hardware beneath them. That forecast helps explain why its founders are prioritizing a configurable machine over a single finished use case. But the market Mai imagines would also need developers willing to take on the work that Feather leaves open: selecting models, adapting hardware and making an application useful on site.

The founders bring experience from both sides of that bet. Mai previously sold a humanoid robotics startup to 1X; co-founder Parsa Bakhtiari worked on Tesla’s Model 3. Feather has now moved beyond a platform idea to small sales and claimed workplace use. Its next challenge is less about proving that developers can choose their own AI than showing that enough of them can turn that choice into dependable work.