Rerun bets physical AI needs its own data stack; $17M seed from Point Nine

Open-source multimodal data platform for 'physical AI': log, query, visualize and stream robotics data (video, point clouds, tensors, joint states) to training, with a commercial database and cloud hub on top.

That AI moving into robots needs its own data stack — multimodal, time-varying sensor data with visualization built in — and open source makes Rerun the default.

$17M seed announced March 2025, led by Point Nine with Costanoa Ventures, Sunflower Capital, Seedcamp and angels including Guillermo Rauch, Eric Jang, Oliver Cameron, Wes McKinney and Nicolas Dessaigne; total funding $20.2M.

Founded in 2022 in Stockholm by Nikolaus West, Emil Ernerfeldt and Moritz Schiebold, drawing on stints at Apple, AWS, Meta, Unity, Zenly and Zipline. Their thesis: vision and robotics were about to be transformed by AI, but the existing data stack — old robotics visualizers like RViz or generic data lakes — couldn't handle multimodal, time-varying sensor data.

For two years the team built the open-source logging and visualization framework, which became the most popular open-source tool for multimodal data; Meta, Google, Hugging Face's LeRobot and Unitree adopted it in their own open-source work. In March 2025 Rerun raised a $17M seed to build a database and cloud data platform around the same data model, with a commercial product under development with design partners and plans to double the team.

Still scaling: as of September 2026 the open-source project sits at 11.4k GitHub stars, the viewer remains free, and Rerun is building Rerun Hub — a scalable catalog and cloud platform for robotic data — while pushing deeper into robotics, drones and spatial computing.

Robot teams had no good option: old robotics visualizers ignore ML workflows, and generic data lakes can't represent spatial, time-varying data — so the gap was real and painful.

Open source made the viewer the default tool inside Meta, Google, Hugging Face and Unitree before any sales conversation, turning adopters into a distribution channel.

Ernerfeldt's egui gave Rerun a home-grown rendering stack, so the team controlled performance end to end instead of borrowing a slow off-the-shelf GUI.

The data flywheel argument — more robots, more data, smarter models — gave investors a reason the category would grow even before the commercial product shipped.

Stockholm's pool of ex-Meta/Zenly/Unity engineers let a small team ship world-class developer tooling from Europe.

Wedge first, monetize later: open-sourcing the viewer every robotics team needed built Rerun's community and demand proof, so the paid database arrived after open source made it default.

As of September 2026, Rerun's open-source project sits at about 11.4k GitHub stars and the viewer is free, while the commercial multimodal database and Rerun Hub remain in development with design partners (per the GitHub repo and the March 2025 seed announcement). The company is doubling down on robotics, drones and spatial computing, and its open-source work is embedded in projects from Meta, Google, Hugging Face and Unitree. The open question is whether a database business can monetize what the community expects for free.