Physical AI for Data Centers

Over the next five years, more than $7 trillion is projected to flow into data center buildout. But the supply chain is still heavily manual and fragmented, and can’t keep pace with current demand. Proprio Robotics is building autonomous robots for the entire data center lifecycle—assembly, maintenance, and decommissioning.

See our launch video!

The Problem

Building and running AI compute infrastructure requires dealing with delicate and fragile components that traditional automation simply cannot do. As a result, servers are manually assembled by hand overseas, and spend weeks in transit before they are integrated into racks and delivered to data centers. Once they get there, labor constraints exacerbate the issue of maintenance. Today, hundreds of thousands of technician positions go unfilled in the United States. Finally, when the servers are retired, there is no efficient way to extract valuable components inside like GPUs and memory.

What We're Building

We develop physical AI models and custom hardware, purpose-made to give robots the dexterity and intelligence to perform various server-related manipulation tasks.

• Assembly - Autonomous robots can assemble servers onshore without driving up labor costs, supporting the growing need to bring ODMs back to the United States.
• Maintenance - We use the same technology for live data center maintenance. We built a working AI data center in our office, and our robot can actively maintain it 24/7. We’re deploying this technology to commercial data centers in the coming months.
• Decommissioning - Our robots also break down previous-generation servers and pull out reusable parts, and we're working with enterprise circular programs and IT Asset Disposition companies to return scarce components to circulation.

Why This Matters

The immense demand for compute requires new innovation. 

To solve power and cooling constraints, operators are building data centers away from the grid, near nuclear facilities, on the water, and in space—environments where it’s not easy to staff humans. With autonomous data centers, we can take human ergonomics and labor constraints out of the picture and redesign entire facilities from scratch.                    

Who We Are

We're graduate robotics researchers out of Berkeley AI Research and Harvard AI and Robotics Lab. Anirudh’s spent years working at one of the country's leading robotics labs, training dexterous manipulation models alongside NVIDIA's GR00T team. Raghav’s led research centered on reinforcement learning and humanoid world models. Both of us are extremely excited to bring this expertise to the AI compute industry, where robotic automation will be indispensable for meeting demand.

Our Ask

As we advance our technology, we’d like to push for more commercial deployments. If you / anybody you know works in any stage of the AI compute lifecycle, we’d love to chat. Reach out to us at founders@propriorobotics.com :)