Midcentury
Data and simulation infrastructure for physical AI: a 2M+ hour egocentric human-action dataset plus Matrix, a cloud simulation platform for robot policies
Midcentury (Midcentury Labs Inc.) is a New York applied research lab building data and simulation infrastructure for robotics and embodied AI. Its bet is that first-person human action data will do for robot pretraining what web scrapes did for language models. The core product is a proprietary egocentric vision dataset of more than 2 million hours of unscripted footage across 50+ industrial and everyday environments and 20,000+ tasks, captured as video, IMU and audio and annotated with 3D hand pose, point tracks, depth and task labels; public research sets are far smaller (Ego4D v2 ~3,600 hours, EPIC-KITCHENS-100 ~100 hours). It also sells custom gameplay data (50K+ hours supported, engine-level signals, frame-aligned inputs and telemetry) and ~69K hours of multilingual full-duplex conversational voice across 25+ languages. The second product, Matrix, is an agentic cloud simulation platform: digital twins that mix classical simulation with physics learned from real data, massively parallel GPU evaluation of robot policies, and a loop that turns failures into new training scenarios. Matrix is early access, not self-serve, and the company says it already works with unnamed frontier labs. It came out of stealth on 2026-09-23 with a $15M seed; investors and valuation were not disclosed. CEO Chetan Kulhari previously worked at AI coding startup Magic; the team cites backgrounds at Stanford AI Lab, OpenAI, DeepMind, NVIDIA, Scale AI and Invisible. Research releases announced as coming soon: MC-EgoHands (egocentric 3D hand/motion reconstruction), MC-Shade (engine-agnostic real-time photoreal rendering) and MC-PhysBench (a long-horizon physics benchmark for world models).
AI
📍 New York, NY
Founded 2024
Current Valuation
Private
PRIVATE
IPO Outlookprivate
Is Midcentury going public?
No. Midcentury is private and has not filed to go public. There is no S-1 on file.
What is Midcentury's valuation in 2026?
No confirmed valuation is on record for Midcentury. The roster shows funding and investors where they are public; it does not estimate a mark.
Has Midcentury filed an S-1?
No public S-1 is on file for Midcentury.
How much has Midcentury raised?
Midcentury has raised $15M in total across 1 disclosed round, most recently a Seed of $15M in Sep 2026.
Is Midcentury publicly traded — can you buy Midcentury stock?
No. Midcentury is a private company — its shares are not listed on any exchange and there is no ticker. The routes to exposure are pre-IPO secondary marketplaces and funds (accredited investors), employee tender offers when the company runs one, and public companies with read-through exposure. QuantLogix maps every route it can verify for Midcentury on its Pre-IPO Access Desk.
Answers are derived from the roster record shown on this page — filing status, last round and confirmed marks. Where the record is silent, the answer says so rather than guessing.
Company Profile
Growth
Out of stealth 2026-09-23 with a $15M seed; says it already supplies frontier labs with its egocentric dataset; Matrix in early access
Last Round
Seed — $15M (Sep 2026)
Fundraising Status
Seed · $15M · this month
privateTotal raised $15M1 disclosed rounds
Funding History
| Round | Amount | Date | Lead Investor | Post-Money |
|---|
| Seed | $15M | Sep 2026 | — | — |
Vertical Intelligence — AI / ML
AI / ML registry data for Midcentury has not been collected yet — the weekly vertical sweep hasn't reached this company. Check back soon.
Talent Signal
Hiring data for Midcentury has not been collected yet — the monthly talent sweep hasn't reached this company. Check back soon.
Products
- Egocentric Vision dataset: 2M+ hours of unscripted first-person video, IMU and audio across 50+ environments and 20,000+ tasks, with 3D hand pose, point tracks, depth and task annotations (proprietary; sample viewer at samples.midcentury.xyz)
- Custom Gameplay Environments: on-demand gameplay data built to spec, 50K+ hours supported, 100+ environment types, frame-aligned inputs, telemetry, camera state and engine G-buffers
- Conversational Voice dataset: 69K+ hours of full-duplex, multi-channel multilingual conversation across 25+ languages
- Matrix: agentic cloud simulation platform to design digital twins, run massively parallel GPU policy evaluation and turn failures into training data (early access)
- Research (announced, releasing soon): MC-EgoHands, MC-Shade, MC-PhysBench
Competitors
Rerun · Vision Lab · Physical · Human Archive · Antioch
AIPhysical AIRoboticsTraining DataSimulationWorld ModelsRecently Funded
Private-company numbers are not real-time. Reflects publicly disclosed valuations from press releases, news reports, and tender offers as of recent date. Refreshed quarterly.