Introducing Midcentury. We’re building the data and simulation infra for physical AI. Today, we’re coming out of stealth with a $15M Series Seed to scale robotics beyond polished demos. We’re already supporting frontier labs with: → The world’s largest egocentric dataset: 2M+ hours, 50+ environments, 20,000+ tasks → Matrix:... a frontier simulation platform scaled with our real-world data to evaluate and post-train policies at scaleshow more

Human action data has finally unlocked a scaling law for Physical AI. It’s time to join LLM researchers in taking the bitter lesson pill. Scaling robotics now requires internet-scale pretraining data and reliable simulation for evaluation + RL. We build the core infra for both.

Robotics can’t scrape the internet for action data. So we built the world’s largest and densest labeled egocentric dataset. 2M+ hours across 50+ environments and 20,000+ tasks, with hands visible in 90% of frames. Paired with SOTA action supervision: 3D hand + body pose, depth, tactile, and motion. Subtask and cycle annotations are frame-aligned with millisecond timestamps.

Real-world evaluation and RL don’t scale in robotics. Matrix is our agentic simulation platform for evaluating and improving robot policies thousands of times before deployment. It combines classical simulation, learned physics, and real-world data to close the sim-to-real gap. Massively parallel cloud simulation scales evaluation and RL.

We’ve brought together a world class team of researchers and operators from the Stanford AI Lab, OpenAI, DeepMind, NVIDIA, Scale AI, Invisible, and more. Our team has built datasets, benchmarks, and models at the frontier of robotics and AI, including RoboNet, GDPVal, and NVIDIA Cosmos 3. We’re a small, deeply technical team hiring across research, engineering, and operations.

Thanks to our partners, collaborators, and investors on this journey, including: @buildpbc, @eddybuild, @cegapereira, @justswart, @jay_drainjr, @liz_harkavy, @Shaughnessy119.

The scaling era for robotics is here. If you’re training, evaluating, or deploying physical AI, we want to hear from you. Explore dataset samples and get early access to Matrix below.

🐐🐐🐐

coming out the gate hot 🔥

so much more to come

Go big or go home!

let's get it

That's just the beginning, we're about to start 🚀

huge things!!! 😋🔥

Congrats team! 🫡

out of many, one. 🚀

cracked team

game changer imo, congrats!

> The world’s largest egocentric dataset funny i heard this so many times

insane launch video, congrats on the raise team!

Midcentury is cooking!! 🔥

congrats!

Hell yeah!

How did you source the training data?where did it come from?

Congrats on the launch!

2milli+ hours of real-world data is seriously impressive!

Physical AI’s bottleneck is sim-to-real gap. $15M buys compute for better physics engines, crucial if they want robots that don’t quit when a table wobbles real-world style. Scaling demo bots needs robust data, not just polish. I’m intrigued by their stack!

fire launch!

"World's largest egocentric dataset" needs a number to mean anything - hours, scenes, embodiment diversity. Ego4D and Epic-Kitchens set the bar years ago and sim-to-real still faces on texture/contact physics, not dataset size alone.

congrats team!! this might be generational

🔥🔥

Data and simulation solve the demo problem; they don't solve the deployment one. The moment a policy leaves the lab it needs a certificate somebody will sign, an incident process, and an insurer willing to price the harm none of which live in the training set. Physical AI hits the same wall every autonomy programme hits: permission to be wrong in public.

lfg @dgmonsoon

the real bottleneck for physical AI may be data and simulation not just better robot models

Big things happening…

Finally live. We have been eagerly watching what you are building and creating here at @vercel for some time. We would love to chat and see how we can support this incredible scale.

building a new civilization, this is so amazing

lfgg

grats team, amazing job

Scaling physical AI will take more than better models. It will take the infrastructure to connect data, simulation, and intelligent systems. This is where orchestration becomes increasingly important.

This just gave me a headache looking at your video flashing all the images. Thanks

congrats guys, 2M hours of data is insane