Source: Physical Intelligence, a startup making an AI model for “any robot or physical device”, raised $70M from Thrive, Khosla, Lux, OpenAI, Sequoia, and more
Physical Intelligence (Pi, π, @physical_int). We're focused on bringing the amazing recent breakthroughs of AI and foundation models into the physical world. Ashlee Vance / @ashleevance : It's a two scoop kinda day. Welcome, world, to Physical Intelligence - a team of robotics and AI all-stars out to build a universal AI for machines. $70 million in !seed! funding from Thrive, Khosla, Lux and Sequoia https://www.bloomberg.com/... Chelsea Finn / @chelseabfinn : I'm really excited to be starting a new adventure with multiple amazing friends & colleagues. Our company is called Physical Intelligence (Pi or π, like the policy). A short thread 🧵 @physical_int : Hello world! We're Physical Intelligence or Pi (like π). Marvin von Hagen / @marvinvonhagen : @physical_int if you had just waited 2 days, you could have launched Pi on Pi Day Sergey Levine / @svlevine : Since cat is out of the bag, it's time I share: I'll be starting a new adventure with an incredible team of friends and long-time collaborators to take on the big challenge of robot learning at scale! It's called Physical Intelligence (Pi... or π, like the symbol). 🧵👇 Forums: Hacker News : Physical Intelligence Is Building a Brain for Robots
Context & Ripple Effects
This seed round established Physical Intelligence as a well-backed attempt to build a shared AI layer for robots rather than a model tied to one machine. Later coverage shows the company pursuing that thesis through robot-task data used to build physical-world understanding.
The financing was an early step in a capital-escalation arc: Physical Intelligence later raised $400M at a $2B valuation, while subsequent reporting focused on whether its models can transfer skills to tasks beyond their training.
First-order effects
- Physical Intelligence gains $70M and backing from major AI and venture investors to recruit, build models, and gather the robot interaction data its universal-model approach requires.
- The investor group gives the startup immediate validation in the emerging market for robot “brains,” placing its platform ambition alongside the hardware makers that could use it.
Second-order effects
- Robot manufacturers and automation customers gain another prospective software layer to evaluate, increasing pressure on robotics AI rivals to show that their models work across machines and tasks.
- Because the product depends on data from robots performing real tasks, access to deployed hardware and task-data collection becomes a more important competitive complement to model development.
Third-order effects
- If cross-robot models prove reusable, value in robotics could shift toward foundation-model platforms and proprietary interaction-data pipelines, while hardware vendors differentiate through integration and deployment access.
- The subsequent focus on a model handling tasks it was not trained on underscores the key industry test: whether claimed generalization can reduce the need for bespoke robot programming at useful reliability.
The trend: This is an early marker of physical AI moving from specialized robot control toward capital-intensive, general-purpose model platforms built on real-world interaction data.