Tensr launches a 12,000-square-foot robot factory and aims to automate the rest

Why it matters

Tensr is turning customer production runs into training data, betting that operating a factory is the fastest route to automating one end to end.

Eric Berndt, Adith Sundram and C.K. Wolfe have launched Tensr, a Y Combinator-backed manufacturer operating a 12,000-square-foot Bay Area factory for building robots designed by other companies.

Tensr said in a launch post that the facility is manufacturing production runs ranging from complete humanoids to robotics intended for the International Space Station. Tensr describes its customers as large U.S. robotics companies and research labs, and says it has a partnership covering manufacturing for the space station.

The founders came out of UC Berkeley's autonomous racing program. Berkeley's Robot Open Autonomous Racing program lists Wolfe as team manager and simulation lead, Sundram as controls lead and Berndt as machine-learning lead for AI Racing Tech. At the Indy Autonomous Challenge's CES 2025 competition, the team completed a 162 mph autonomous overtake to win the two-car passing event.

Berndt also wrote a 2025 UC Berkeley technical report on using a transformer trained through behavior cloning to control a full-scale autonomous racecar. Tensr transfers that experience in simulation, control systems and real-world machine learning from a closed racetrack to a factory with changing designs, suppliers and production equipment.

The customer order becomes training data

Tensr's immediate business is manufacturing. Its larger bet is that operating the factory will produce the data needed to automate more of it.

The founders want Tensr to eventually process an order from prototype through a production run within 24 hours. That proposed system covers supply-chain management, design-for-manufacturing work, machine retooling, robotic adaptation, inspection, packaging and logistics.

Tensr says it has already established supplier relationships spanning electronics, actuators, surface-mount assembly, computing hardware, pneumatics and raw materials. Those relationships are essential to the 24-hour target: faster robotic manipulation does little good when a missing actuator or circuit board keeps the order off the line.

The factory is designed as a feedback system. Tensr plans to collect errors from customer designs, procurement forecasts, retooling jobs and robot movements, then use those failures to improve subsequent production runs. The founders argue that factory robots should learn from their own operations rather than depend primarily on demonstrations or other human-generated training data.

That approach gives Tensr a reason to start as a contract manufacturer before its autonomy software is complete. Every customer order supplies another combination of parts, tolerances, machinery and assembly steps. The production work can build the data set required for the autonomous factory Tensr ultimately wants to sell as an operating model.

Tensr's public materials draw a line between its current operation and that end state. Its homepage says Tensr operates "fully autonomous factories," while the launch post says Tensr is building a factory capable of handling orders autonomously from end to end. The 24-hour order-to-production cycle remains the stated goal.

From racing cars to production lines

The racing background matters because manufacturing automation has many of the same constraints the founders encountered on the track: systems must interpret noisy physical environments, act under time pressure and recover when simulations fail to match reality.

Factories add a broader coordination problem. Tensr must connect software with purchasing, inventory, machine setup, quality control and shipping. Success depends on whether the founders can make that entire chain respond to a new design without the manual engineering work that usually accompanies a hardware production run.

Tensr says its target is consistent unit economics whether a customer orders one robot or one million. That objective would require much of the setup work for small runs to be automated, since engineering time, tooling changes and supplier coordination usually make low-volume hardware disproportionately expensive.

The company was founded in 2025 and is listed in Y Combinator's Summer 2026 batch. Tensr plans to publish a whitepaper describing its manufacturing system on August 31st.

The factory is already the consequential part of the launch. Tensr has moved beyond proposing software for an eventual automated plant and taken responsibility for producing customer hardware. The founders now have to make the plant itself learn quickly enough to justify the autonomy claim.