Generalist AI, the robotics lab founded by former Google Brain researchers, closed a $400 million funding round on June 4, bringing its total raised to more than half a billion dollars. Radical Ventures led the round. NVIDIA participated as a continuing investor.
The round arrives six weeks after Generalist AI published results for GEN-1, its second-generation action model, which the company says achieved 99% reliability across dexterous tasks and ran at three times the speed of its predecessor. Those figures are company-reported and have not been independently validated. What matters to investors is the trajectory: GEN-0, released last November, was a pretraining proof of concept; GEN-1, released in April, is positioned as commercially viable.
Generalist AI’s core thesis is that physical intelligence requires a foundation model that treats action as a native output, not a downstream application of video generation or language modeling. The GEN series trains on real-world physical experience at scale, applying to robotics the same scaling-laws logic that produced capable language models. The company says the data flywheel is beginning to close: better models get deployed, deployment generates proprietary physical data, and that data trains the next model generation.
NVIDIA’s position in this round deserves a second look. Through its NVentures arm, NVIDIA has backed a portfolio of physical AI labs that includes Figure AI, Physical Intelligence, and now Generalist AI at a disclosed level. Last month, NVIDIA released Cosmos 3, an open foundation model that generates robot action outputs directly. Cosmos 3 is the model layer of NVIDIA’s physical AI strategy; Generalist AI is the company layer. Both sit inside the same thesis: that robots running on NVIDIA compute, trained on NVIDIA-adjacent foundation models, constitute the next major hardware demand cycle after the data-center GPU wave.
This round also completes a rough accounting of physical AI capital in 2026. Physical Intelligence raised $737 million at the end of 2024 and has continued deploying; Figure AI closed its most recent round at a $2.6 billion valuation. Generalist AI’s $400 million places it in the second tier of capital concentration. The combined investment in action-native robotics labs has crossed $1 billion in new capital this year alone. The bet, now repeated by multiple top-tier investors, is that robot intelligence scales the same way language intelligence did, and that the resulting systems are worth funding at frontier-lab prices before revenue materializes.
The practical implication for builders is worth stating plainly: the field is no longer running validation experiments. At $400 million per bet across four to six well-capitalized labs, the competition has shifted to time-to-market. The lab that reaches reliable, multi-environment deployment at commercial unit economics first controls the training-data flywheel; the others face a scaling gap that capital alone cannot close.
Generalist AI’s announcement does not disclose revenue, customer names, or deployment volume. The 99% reliability figure applies to a defined set of dexterous tasks; the company has not specified what those tasks are or how the benchmark was constructed. Independent replication has not been reported.
For teams evaluating robot intelligence vendors or making infrastructure bets on physical AI, the next twelve months are the window to form a view. The architecture question has been answered well enough to attract nine-figure capital. The market-viability question is still open, and the labs that close it first will set the terms for everyone else.
Generalist AI (generalistai.com), published June 4, 2026.
