Dream Labs raised a $90 million seed round as it launched a metered cloud API for controlling robots, tying the financing to a plan for hardware makers and developers to rent a shared software brain instead of training a separate model for every machine.
The San Francisco company was founded in 2026 by former leaders of Nvidia’s GR00T robotics research, including CEO Joel Jang. B Capital led the round, with Kindred Ventures among the investors. Dream Labs says it already works with more than 20 hardware partners.
Developers send the model a robot’s camera observations and a plain-English task; the hosted service returns targets for the grippers, and the robot’s own software converts them into joint movements. The launch version works with two-armed robots using parallel grippers and concentrates on tabletop tasks such as moving objects and folding laundry. Other hardware can be adapted through a separate fine-tuning service.
That portability remains a company claim with visible limits. Forbes reported that Dream Labs put zero-shot success at about 80% for pick-and-place and 50% to 60% on other tasks—well below the reliability needed on a factory line. Dream Labs labels the product a research beta and says it has not been validated for safety-critical use.
Dream Labs charges $3 per hour for its flagship model and $2 per hour for a lighter version. At the flagship list price, continuous use works out to about $26,300 a year before the robot, integration or safety work—more than the roughly $2,000 to $10,000 price of the supported hardware cited in the launch materials.
Hosted access across different manufacturers is Dream Labs’ intended distinction in a field that includes Skild AI, Physical Intelligence, Google and Nvidia. The launch does not yet cover wheeled mobile robots or dexterous hands; Dream Labs says those are next, while most of the new capital goes to the computing capacity needed to support more tasks and machines.