Midcentury Emerges From Stealth With $15M Seed for Physical AI Training Data
New York startup Midcentury has emerged from stealth with a $15 million seed round to tackle the data bottleneck in robotics. Launched on September 23, 2026, the company debuted two The post Midcentury Emerges From Stealth With $15M Seed for Physical AI Training Data appeared first on Ventureburn .
Midcentury, a New York-based startup, has emerged from stealth mode with a $15 million seed investment to address the data shortage in the robotics industry. The company made its official debut on September 23, 2026, unveiling two key products: an extensive egocentric dataset consisting of over two million hours of human behavior and a cloud simulation platform known as Matrix.
Midcentury's latest funding round positions it among the top 1% of seed deals in the big data sector, with the median seed round in Q2 2026 averaging $4.5 million. The startup refrained from disclosing its valuation or investors.
The company's primary focus lies in leveraging first-person human action data to train robots, mirroring how web scrapes were utilized to train large language models. Midcentury's proprietary dataset encompasses more than 50 environments and 20,000 tasks. Each hour of data is meticulously annotated with 3D hand pose tracking, depth maps, and point tracks, significantly surpassing existing public research options such as Ego4D v2, which contains approximately 3,600 hours of data.
In addition to human action recordings, Midcentury holds approximately 50,000 hours of gameplay data with engine-level signals, and around 69,000 hours of conversational voice data in 25 languages.
Due to the dataset's proprietary nature, researchers will not have access to it without payment. Midcentury also introduced Matrix, a cloud simulation platform designed to enable teams to create digital twins of real-world scenarios. Unlike traditional methods, Matrix generates physics directly from real data rather than relying on hand-coded rules.
By harnessing Rich Sutton's 'bitter lesson,' which asserts that brute-force compute and vast datasets ultimately surpass hand-engineered features, the company utilizes GPU clusters to run thousands of parallel tests, transforming test failures into immediate training examples. Chief executive officer Chetan Kulhari, who previously worked at AI coding startup Magic, spearheads the company.
A January 2026 SEC filing indicated that Midcentury had distributed roughly $8.9 million across five investors, signifying the successful completion of the funding round in multiple stages. As other startups, such as Rerun and Vision Lab, vie for the data layer in embodied AI, the entities controlling the training data are likely to hold significant market power.
Written by urgent.news from Ventureburn's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.
