By Hamza Faisal

The people who helped build Nvidia's GR00T robot research program have walked out with $90 million to start a new one. Dream Labs, founded in 2026 by the former leaders of the GR00T effort, announced the seed round on October 19 alongside the launch of Dream-1 — a robot foundation model its makers call the first you can simply plug into any robot.

The round was led by B Capital and Kindred Ventures, with A.Capital, Unusual Ventures, SV Angel, Wing VC, Scribble VC and Adrian Aoun also taking part, according to the company's announcement.

One brain for every robot

Dream-1 takes a robot's camera feed and a plain-English instruction, and turns them into motor commands. Because it was trained on video of people performing everyday tasks — learning to predict what happens next rather than memorizing one machine's movements — its skills are meant to transfer across hardware. Out of the box, it can control any robot with grippers, with no extra training.

Right now that means tabletop work: picking things up, placing them, folding laundry. The company says more than 20 hardware partners are on board, and wheeled mobile robots and more dexterous hands are on the roadmap. The money, the company says, will mostly go toward computing capacity as it expands the list of supported robots and tasks.

A $90 million vote of confidence

A $90 million seed round is enormous by any standard — and it's a bet that robotics has the same shape of problem software had before the cloud. "The hardware is ready. What's been missing is a brain any robot can use," co-founder and CEO Joel Jang said in the announcement, comparing it to driving a rental car: "You find the controls and your past driving experience does the rest."

The context helps explain the excitement. A record five million industrial robots now work in factories around the world, yet nearly every one is programmed for a single task, and nothing it learns transfers to another machine. Meanwhile low-cost robots sell for a few thousand dollars and ship with almost no built-in intelligence. Teaching one robot one task can take specialists months; a new task or new hardware often means starting from zero.

The fine print

Alongside the financing, Dream Labs opened up Dream API — the first publicly available API for a robot foundation model — plus a fine-tuning service for unsupported hardware, and a browser-based Robot Playground where anyone can issue commands to a physical robot in the company's San Francisco office and watch it respond live.

The company is also honest about where the technology stands: Dream-1 is a research beta that "will make mistakes" and has not been validated for safety-critical use. The pitch isn't that robots are finished — it's that developers can now test a cross-machine robot brain for themselves, failures included, instead of relying on edited demos.

By Hamza Faisal