Why We’re Doubling Down on Rerun: Visualizing the Future of Physical AI

What struck me when I first met Rerun’s founders – Niko, Moritz, and Emil – back in 2022 was their unwavering conviction in a future that few others could see. Few were even using the terms “physical AI” and “multimodal AI,” yet their goal was to build the premier software for visualizing physical AI. 

They built it, and now the top teams in robotics, AR, and VR use Rerun to visualize and debug their applications, dramatically accelerating their iteration cycles. Next up: the team is expanding beyond visualization to build the complete data stack for multimodal AI.

If there’s one thing I’ve learned in the three years since meeting the Rerun team, it’s to trust their judgment. That’s why we at Costanoa are proud to have led their pre-seed round and are even more excited to double down in their seed round.

Why? Because the founders’ leadership experience at Volumental gave them a glimpse of the future before it was obvious to everyone else. They have been bucking conventions since day one. Here’s how:

1. They believed physical AI would transform the world well before the current robotics hype and before AR/VR was taken seriously. Meta Ray-Ban Glasses didn’t exist. Physical Intelligence, Sanctuary AI, and Skild hadn’t yet raised their massive rounds. Unitree wasn’t showcasing its robots’ capabilities daily. They chose to start with the visualization layer, even though visualization tools were considered difficult to commercialize. At the time, trendy data infrastructure companies were building compute engines and storage layers, all aspiring to become the next Databricks or Snowflake.

The founders recognized that visualization was the most acute pain point for physical AI developers. Interacting with multimodal data demands visualization, not charts or tables. Developers needed a responsive, intuitive tool that enabled rapid debugging and faster iteration. Teams struggled with disconnected, ad-hoc visualization tools built by individual developers and desperately needed a unified approach to data interaction.

1. They aimed for industry giants but took a developer-first approach. Rather than going top-down, they built software that developers and small teams loved. Tech giants quickly took notice.  By getting adoption from a single team, Rerun could then spread organization-wide, becoming the standard abstraction layer for physical world data. Meta adopted the open source project for Project Aria. Not long after, Hugging Face picked it as the default visualizer for LeRobot. Now, it’s been embraced by Unitree, Google, and many other leading Physical AI teams. 

The founders are bold and ambitious. They don’t adhere to conventions – and their big bets have paid off. They created open-source software that is critical to physical AI, and are now building the entire data stack for it. 

We are thrilled to be along for their unconventional ride.