Build AI's $5M bet on human-view robotics

Build AI has emerged from stealth with a $5 million seed round and a mission that borders on sci-fi: solving embodied AI.

The company, led by a founder who reportedly dropped out of Columbia and turned down over $25 million in equity to pursue this vision, believes the key to creating general-purpose robots lies in building the "largest egocentric data collection effort in history."

The round was led by a trio of notable early-stage investors: Abstract Ventures, Pear VC, and HF0, with participation from ZFellows and angel investor Alex Botez. The goal isn't just to build another language model or software tool. Build AI is making a fundamental bet that amassing vast quantities of 3D world data, captured from a first-person human perspective, is the missing ingredient for robots that can navigate and interact with the messy, unpredictable real world.

According to the company's sparse but ambitious website, this is about building "AI on a human foundation." While details on the data collection methods remain under wraps, the approach stands in contrast to relying purely on simulation or third-person video. The thesis is that for a robot to understand how to move through a kitchen or a workshop, it needs to see the world as we do.

The website notes that other team members have left "academia, top labs, and their own companies to do the same."

Collecting, processing, and training models on egocentric data at an unprecedented scale is a monumental technical and logistical hurdle.