San Francisco startup Dream Labs raised $90 million in seed funding led by B Capital and Kindred Ventures to scale Dream-1, a universal robot foundation model. Founded by former Nvidia GR00T researchers, Dream Labs also released the Dream API, charging $3.00 per hour for its flagship model and $0.75 for a lightweight version. Dream-1 uses human video data to control two-armed robots out of the box, offering rapid cross-embodiment skills transfer.
Seed funding and resource allocation
- ▪Investors participating in the Dream Labs $90 million seed round announced on October 19, 2026 included B Capital, Kindred Ventures, A.Capital, Unusual Ventures, SV Angel, Wing VC, Scribble VC, and Adrian Aoun
- ▪Dream Labs plans to allocate most of its $90 million seed funding toward compute resources to expand the capabilities of its Dream-1 robot foundation model
- ▪San Francisco startup Dream Labs, founded in 2026 by former Nvidia GR00T robotics leaders, raised $90 million in seed funding on October 19, 2026
Dream-1 foundation model capabilities
- ▪Dream Labs' Dream-1 model achieves an 80% success rate on zero-shot pick-and-place tasks and 50% to 60% on other tabletop tasks
- ▪Dream-1 treats robotic control as a video prediction problem, learning physical tasks from human egocentric video footage shot from a first-person perspective
- ▪Dream Labs released Dream-1, a universal robot foundation model designed to act as an adaptable software brain for physical hardware
Cross-embodiment control capabilities
- ▪Dream-1 provides zero-shot cross-embodiment control out of the box for any robot equipped with two arms and parallel grippers
- ▪CEO Joel Jang said Dream Labs plans to expand its Dream-1 model's cross-embodiment capabilities to support wheeled mobile robots and fully dexterous humanlike hands
Dream API launch and pricing
- ▪Dream Labs launched the Dream API on October 19, 2026, as the first publicly available API for a universal robot foundation model
- ▪Dream Labs' Dream API charges $3.00 per hour (approximately $26,000 annually running continuously) for the flagship Dream-1 model, and $0.75 per hour ($6,600 annually) for the lighter Dream-1 flash model
Rapid fine-tuning service
- ▪Dream Labs offers Fine-Tuning-as-a-Service, allowing developers to submit video demonstrations and receive a customized version of its Dream-1 model within hours
- ▪Traditional custom robot programming required three to six months, whereas Dream Labs enables hardware fine-tuning in hours for a few dollars
Robot Playground testing platform
- ▪Dream Labs chief executive Joel Jang stated that Robot Playground provides unedited live responses, including mistakes, to let users test the Dream-1 model directly
- ▪Dream Labs launched Robot Playground, enabling web browser users to send natural language commands to a live robot in San Francisco
Robotic hardware partnerships
- ▪Dream Labs refines its Dream-1 model specifically for its more than 20 hardware partners to optimize performance and enable single-model control over mixed robot fleets
- ▪Dream Labs collaborates with more than 20 robotic hardware partners whose machines cost between $2,000 and $10,000
Investor perspective on Dream Labs
- ▪B Capital general partner Daisy Cai stated that capable robotic arm hardware costs have fallen from luxury car prices to high-end laptop prices
- ▪B Capital general partner Daisy Cai stated that the Dream Labs team conducted defining work in robot learning during their previous roles at Nvidia
Industrial robotics limitations
- ▪Dream Labs estimates that nearly all of the 5 million industrial robots working globally in factories and warehouses are limited to single specialized tasks
- ▪Dream Labs chief executive Joel Jang stated that existing industrial robots cannot pass on acquired knowledge to train other machines or adapt to new tasks
Story comments
Loading comments…