Dream Labs, an AI company founded by the former leaders of Nvidia’s GR00T robotics research program, announced on October 19 that it has raised $90 million in seed funding and launched Dream-1, a robot foundation model it calls the first you can simply plug into any robot, the company said in its official announcement.
What Was Announced?
The seed round was led by B Capital along with Kindred Ventures, with participation from A.Capital, Unusual Ventures, SV Angel, Wing VC, Scribble VC and angel investor Adrian Aoun, among others. The company was founded in 2026 by former leads of Nvidia’s GR00T effort, the team behind DreamZero and World Action Models, and is led by co-founder and CEO Joel Jang.
Alongside the financing, Dream Labs launched three products: Dream API, described as the first publicly available API that turns a robot’s camera feed and a plain-English instruction into motor commands, even for robots the model has never seen; Fine-tuning as a Service, which returns a tailored model within hours from a small batch of demonstration data; and Robot Playground, a browser-based demo where anyone can send commands to a live robot in the company’s San Francisco office and watch it respond unedited.
Graphic: GPTOnline, using Dream Labs assets
One Model, Many Robots
Dream-1 is trained on video of people performing everyday tasks rather than on any single machine. It predicts what will happen next in a scene and translates that into the action a robot should take, which is why its skills transfer across hardware from different manufacturers. Out of the box, the company says, Dream-1 can control any robot with grippers with zero-shot performance, meaning no additional training data.
The framing is a direct attack on the biggest cost problem in robotics. Factories worldwide run a record 5 million industrial robots, the company notes, yet nearly every one is programmed for a single task, and what it learns cannot transfer to another machine. Low-cost robots now sell for a few thousand dollars but ship with no built-in intelligence, and teaching one robot one task can take specialists months.
“The hardware is ready. What’s been missing is a brain any robot can use,” Jang said. “When you get into a rental car, you don’t relearn how to drive. You find the controls and your past driving experience does the rest. Dream-1 approaches a new robot the same way, so a new robot can do useful work in minutes, not months. And instead of edited demos, we’re giving people a working model they can test themselves, mistakes included.”
“A capable robot arm has gone from the price of a luxury car to the price of a high-end laptop. What it needs now is intelligence,” said Daisy Cai, general partner at B Capital. “This team did some of the defining work in robot learning at NVIDIA. Now anyone with an API key can build on it.”
Availability and Roadmap
Dream Labs says it works with more than 20 hardware partners, and the robots it supports today cost between about $2,000 and $10,000. Dream-1 currently handles tabletop tasks such as picking and placing objects and folding laundry, with support for wheeled mobile robots and more dexterous hands to follow. The company says it will use the new funding primarily for compute as it expands the tasks and robots Dream-1 can handle.
Graphic: GPTOnline, using Dream Labs assets
Early partners are already testing the claim. “With Dream-1, our robot was up and running in a few minutes and attempting tasks we never programmed,” said Ryan Chan, co-founder and CEO of MakerMods, which sells $2,000 bimanual robot arms. “We can focus on building better hardware, and our customers get intelligence that keeps improving.”
Why It Matters
Robotics has a software problem more than a hardware problem: cheap, capable arms exist, but every deployment starts from zero. If a single foundation model can genuinely transfer skills across different robots without retraining, it would do for robotics what shared pretrained models did for language and vision. Dream Labs is betting that learning from video of humans, rather than from scarce robot teleoperation data, is the shortcut. The public API and the live, unedited Robot Playground are a confident way to make that case: anyone can now test the claim themselves.