Generalist AI Inc., a San Mateo-based startup that builds foundation models for robots, has raised approximately $200 million in a new funding round led by venture firm 8VC, Axios reported on Monday, citing a source familiar with the deal and a federal filing disclosing the raise.
The round comes just two months after Generalist closed a $400 million financing in June at a reported $2 billion valuation, a round that included participation from Nvidia and Bezos Expeditions, according to SiliconANGLE. With the new capital, the company has now raised more than $650 million total as investors continue to pile into so-called physical AI companies building general-purpose robotic systems. For more context on this story, see our ongoing AI news.
Software Brains, Not Metal Bodies
Generalist occupies a distinctive position in the robotics boom: rather than manufacturing robotic hardware, the company develops the AI "brains" — foundation models trained on vast amounts of data — that can power machines built by other manufacturers. The approach mirrors the strategy that made large language models the underlying infrastructure for the chatbot industry, applied instead to the physical world.
The startup's bet is that as robots proliferate across factories and warehouses, the value will concentrate in the intelligence layer rather than the actuators and chassis. Investors appear to share that thesis. The June round drew Nvidia — whose Jetson chip line targets robots specifically — along with more than half a dozen other backers, and the new $200 million injection was joined by several unnamed existing investors, Axios reported.
Gen-1.5: Teaching Robots by Demonstration
The funding news lands about a week after Generalist debuted its newest model, Gen-1.5, which is designed to power robotic arms in industrial settings. The model addresses one of the most persistent bottlenecks in factory automation: the cost and fragility of programming robots for each new task.
Historically, engineers had to write custom code for every task a robotic arm performs, and even minor operational changes — swapping in larger packaging boxes, for example — could require rewrites. While some modern robots ship with AI that reduces hand-coding, teaching them new tasks typically means fine-tuning neural networks, a process that can consume significant engineering time.
Gen-1.5 takes a different approach. Users can teach the robot a new task simply by demonstrating it with their own hands, with the motion captured by the robot's built-in cameras or by sensors worn on the demonstrator's hands. The model can also learn from recorded clips of simulated robots performing tasks.
The performance numbers the company published are notable. Across 10 sample tasks, Gen-1.5 achieved an average task completion rate of 59 percent when given a single demonstration of how to perform them. When users supplied a few additional examples, the completion rate climbed to 83 percent. Generalist says the model is the first capable of learning a wide range of robotics tasks from just one or a few examples.
The model also showed signs of autonomous judgment during testing: on some tasks, Gen-1.5 chose to complete an assignment using a different tool than the one it had been instructed to use.
The Compute Cost of Physical Intelligence
Training such models is not cheap. Generalist says Gen-1.5 took more than eight months to train, and while the company has not detailed how it will deploy the new $200 million, a sizable share is expected to fund AI infrastructure. Nvidia's participation in the June round hints that Generalist's training runs lean heavily on the chipmaker's data center GPUs.
The rapid back-to-back raises — $400 million in June, another $200 million by late August — reflect the intensifying capital race in embodied AI. Venture funding into robotics and physical AI startups has accelerated over the past two years as improvements in vision, manipulation, and reasoning models have made general-purpose robots commercially plausible.
Generalist faces competition from well-funded rivals pursuing similar foundation-model strategies, including Physical Intelligence, whose generalist robot policies have drawn significant attention from researchers. What the segment's winners will look like remains uncertain, but the scale of capital now flowing into robot-brain startups suggests investors believe the software layer of robotics could become as consequential as the model layer of the chatbot market.
The pace of the raises also reflects how quickly the robotics software market has matured. A company that two years ago would have been a research project is now commanding multi-billion-dollar valuations on the strength of demonstration videos and benchmark task-completion rates — metrics that would have seemed fantastical for general-purpose manipulation models only a few hardware generations ago. The June round alone drew Nvidia, Bezos Expeditions, and more than half a dozen additional backers in what SiliconANGLE described as a crowded syndicate, and the fact that 8VC returned to lead again barely eight weeks later signals that demand for allocation in the company exceeded what a single round could absorb.
For now, Generalist is signaling confidence with its wallet. A company that has raised nearly two-thirds of a billion dollars in a single summer is effectively telling the market that the era of robots learning tasks by watching humans is close enough to commercial reality to justify the largest of bets.
---
Stay Ahead of AIGet the latest AI news, analysis, and breakthroughs — all in one place.
Read more AI news →
