Amazon and Nvidia are backing eight AI and robotics startups in a Physical AI Fellowship managed by MassRobotics, the Massachusetts-based robotics hub. The eight startups are getting science and engineering support, $200,000 in AWS credits, access to Nvidia’s hardware and software stack, and go-to-market support at major Amazon, Nvidia and MassRobotics events.
The focus: physical AI.
“Physical AI is really effectively saying how do you use spatial awareness and apply that with intelligence to be able to now have actuation done on physical objects and in physical settings,” Amazon’s head of generative AI innovation and delivery at AWS Taimur Rashid told me recently on the TechFirst podcast.
There’s a massive opportunity here, Rashid says: 2.5 billion people globally do physical labor that represents almost $50 trillion worth of annual output that automation and physical AI can help with.
The eight startups are applying AI to tasks as diverse as controlling self-piloting sea-going vessels, self-driving construction equipment, accelerating logistics, and growing food on farms. Amazon knows this works: Rashid says robotic automation in its fulfillment centers has contributed to boosting efficiency 25% already.
Here are the eight startups:
- Bedrock Robotics: autonomous construction equipment
Construction sites are chaotic, dusty, and uneven, all of which is tough on robots. Bedrock Robotics, based in San Francisco, tackles that by retrofitting bulldozers, excavators, and loaders with sensors, compute units, and AI software, turning them into self-driving heavy machines. - Blue Water Autonomy: robotic ships for the open ocean
The ocean is one of the hardest environments for autonomy: stormy, always-changing, and help is often far away. Blue Water Autonomy builds AI-powered surface vessels that can patrol, transport, or conduct research without human crews. Each ship uses real-time perception and planning to navigate changing currents and weather. - Diligent Robotics: humanoid helpers for healthcare
Diligent Robotics, out of Austin, Texas, develops socially intelligent wheeled robots that already operate in 25 hospitals across the United States. Their robot Moxi ferries supplies, handles deliveries, and supports nursing staff. - Generalist AI: teaching robots to learn like humans
Today’s robots are usually specialists without a lot of intelligence. Generalist AI wants them to become learners, sort of like humans. The company is building foundation-model architectures for robotics: systems that let machines master multiple tasks and adapt to new ones. - RobCo: modular automation for smaller factories
German startup RobCo provides modular robotic arms and low-code software for small and mid-sized manufacturers. Each arm can be reconfigured for new tasks, giving smaller firms affordable, flexible automation. Robots can be extremely expensive and out of reach for smaller companies: this could be the democratization of robotics, bringing physical AI to the long tail of industry. - Tutor Intelligence: warehouse automation as a service
Tutor Intelligence builds AI-driven warehouse robots that handle palletizing and logistics. The systems use computer vision and reinforcement learning and are sold as a subscription service. - Wandercraft: powered exoskeletons restoring mobility
Paris-based Wandercraft designs powered exoskeletons that let people with mobility impairments stand and walk more naturally. Its self-balancing design fuses robotics, control theory, and AI to restore movement. - Zordi: autonomous farm robots
Zordi builds agricultural robots that monitor crops, detect ripeness, and harvest safely. The company brings automation to greenhouses and fields while reducing waste and input costs.
The opportunity is huge right now, Rashid says, thanks to the confluence of three major forces: increasing labor shortages in wealthier countries as the population ages, falling hardware costs for sensors, cameras, processors and motors for robots, and increasingly capable generative and agentic AI that are helping us train robots and give them intelligence much easier than before.
But it’s still early days.
“While I do feel like we are still in the very early days of agentic AI,” Rashid says, “we are even in the earliest days of physical AI.”
That future will include many different types of robots, he adds.
Rashid says the future of robotics will include both general-purpose robots and highly specialized ones. In his view, physical AI will mirror the pattern already seen in generative AI, where some systems will be broad and adaptable, capable of handling a variety of tasks in many environments, and others will be trained and optimized for very specific, high-precision work.
Both types are necessary. Generalist robots provide flexibility and scale, while specialist robots deliver the accuracy and reliability required for sensitive or highly skilled operations.
These eight current companies should graduate from the Fellowship later this year. For next year, Amazon is looking at bringing more enterprise companies into the program.
