December 9, 2020 | Investment Themes

Synthetic data will really scale AI: Announcing our Series A in Parallel Domain

John Cowgill

Written by

John Cowgill

Computer vision is core to so many of the AI companies we see today. It’s what enables computers to look at the world and automatically make decisions—be it autonomously navigating a drone, analyzing a warehouse to better control inventory, or even dramatically improve fish farming efficiency—computer vision will soon be embedded in many human tasks and processes that exist today.

But the bottleneck holding back the development of computer vision is data–specifically collecting large volumes of data, labeling it, QA’ing that data, training a model on it, and then determining what data is missing and repeating the laborious process. AI algorithms are only as good as the data they process.

That’s why we’re so excited about the potential of synthetic data, and are thrilled to be doubling down on our initial Seed investment in Parallel Domain and participating in their $11M Series A, which was led by our friends at Foundry Group, with participation from Calibrate Ventures, Toyota AI Ventures, and Ubiquity Ventures. I’m also excited to be joining the board as part of the round. 

Parallel Domain’s synthetic data generation platform breaks the data collection paradigm by procedurally generating photorealistic, perfectly annotated data, with no humans in the loop. The implications for the development of AI are profound. Developers can now call data instantly from an API rather than wait weeks or months for data to be collected in the field. Parallel Domain’s customers are seeing immediate results–in some cases as much as a 45% reduction in error rates.

At the heart of any great startup is a great team, and I’ll never forget my first meeting with Parallel Domain’s founder, Kevin McNamara. We met in October 2017—about six months before we led the Seed round. Kevin had recently left Apple and we scheduled a coffee in Menlo Park. I found him sitting at a large table with a tall stack of reports from the NHTSA on traffic collisions and automotive safety on one side of the table, and his laptop open with Houdini 3D animation software on the other. 

I’d met a lot of talented technologists working in autonomy by this point—but that meeting made an impression. Kevin was thinking deeply about what synthetic data could do for autonomy–how it could enable safety, scalability, and speed of development.

Fast forward three years to today, and it’s so exciting to see how far the company has come. At the time of our investment, Parallel Domain was just Kevin and a theory that synthetic data could improve the speed and safety of AV development. Since our investment, Kevin’s built out a complete executive team and attracted a world-class team of synthetic data experts distributed across the US and Canada. While AV remains core to our mission, we’ve found synthetic data is relevant to all applications of computer vision, including delivery robots, drones, defense, manufacturing, and more. 2020 has been a banner year with several customers doubling down on their commitment to synthetic data and expanding their usage of the Parallel Domain platform.

Looking ahead, we believe Parallel Domain has the opportunity to scale AI development in a way that real world data collection and labeling simply cannot. We’re thrilled to be partnering for the next phase of their journey. If synthetic data interests you, come join us—they’re hiring