New York Startup Midcentury Raised $15 Million

The robotics firm emerged from stealth with a massive dataset and a cloud-based simulation platform.

Updated on Sept. 26, 2026 in Robotics

New York Startup Midcentury Raised $15 Million

Live Poll

Should companies be allowed to restrict access to the datasets used to train physical AI robots?

New York-based startup Midcentury has officially emerged from stealth after securing $15 million in seed funding. The company aims to overcome existing data bottlenecks in robotics training by providing large-scale human behavioral insights.

Why it matters

Robotics development has long been constrained by a lack of diverse, high-quality training data for machines to interact with the real world. Midcentury attempts to bridge this gap by offering a massive dataset that allows systems to learn from complex human scenarios.

The firm’s new egocentric dataset includes over two million hours of human behavior data, covering 50 environments and 20,000 tasks. This repository features 50,000 hours of gameplay data and 69,000 hours of conversational voice data across 25 languages.

The players

Midcentury

This New York-based robotics startup focuses on solving training data shortages for artificial intelligence systems.

Chetan Kulhari

He serves as the Chief Executive Officer of Midcentury and leads the company in its mission to scale robotics training.

The details

Midcentury also introduced Matrix, a cloud simulation platform that generates digital twins of real-world scenarios. Engineering teams utilize this tool to run parallel tests on GPU clusters, which effectively turns failed test iterations into new, actionable training examples for robots.

In January 2026, an SEC filing revealed the company had already sold $8.9 million in equity.
During the second quarter of 2026, the median seed round for big data companies reached $4.5 million.
The company officially launched its operations and platforms on September 23, 2026.

The Tech Race

Midcentury’s two million hours of data dwarfs the 3,600 hours available in the industry-standard Ego4D v2 research option. This massive expansion reflects an arms race to provide proprietary datasets that train next-generation robots for real-world autonomy.

For developers and researchers, this platform offers a new pathway to accelerate the deployment of intelligent robotics into domestic and professional environments. Users may see faster iterations in robotic capabilities as the industry gains access to more sophisticated training tools.

The takeaway

The company’s ability to turn failures into training examples through digital twins represents a significant shift toward automated robot learning. This approach could drastically reduce the time needed to prepare machines for complex human environments.