The archive · Developer & Business Tools · Product decision · 2025–2026
Antioch's sim-to-real bet: $8.5M seed to make robot training happen in software
NYC team founded May 2025 bets cloud simulation can replace robot test warehouses; $8.5M seed at $60M valuation, TechCrunch-covered.
Antioch
What the business is
Antioch is a cloud simulation platform for robotics and autonomy: developers spin up digital instances of their hardware, connect them to simulated sensors, and train, test and generate data for autonomous systems without building physical test arenas.
Starting capital：$8.5M seed at $60M valuation, led by A* and Category Ventures with MaC Venture Capital, Abstract, Box Group and Icehouse Ventures
How it started
Harry Mellsop and four co-founders started Antioch in New York in May 2025. Langshur and Calvey had already co-founded Transpose with Mellsop and sold it to Chainalysis; Schlager came from Meta Reality Labs and Swingle from Google DeepMind. Mellsop's read: most of the industry doesn't use simulation at all, yet physical testing is what makes robot development slow.
What happened
In April 2026 Antioch announced an $8.5M seed led by A* and Category Ventures, with MaC Venture Capital, Abstract, Box Group and Icehouse Ventures, valuing the company at $60M. Some of its earliest engagements were large multinationals already investing heavily in robotics, while MIT CSAIL's David Mayo used the platform to have AI models design robots and test them against each other. Foxglove founder Adrian Macneil joined as an angel, arguing physical AI needs off-the-shelf tooling like GitHub, Stripe and Twilio.
No ending yet — it is still running.
Background
Antioch is a New York-based startup building cloud simulation for robotics and autonomy. Its platform lets robot developers spin up digital instances of their hardware, connect them to simulated sensors that mimic what the real world would deliver, and use those environments to test edge cases, run reinforcement learning or generate training data — no physical mock-up warehouses or test fleets required.
The company was founded in May 2025 by Harry Mellsop with four co-founders: Alex Langshur and Michael Calvey, who had previously sold their security startup Transpose to Chainalysis, plus Collin Schlager from Meta Reality Labs and Colton Swingle from Google DeepMind. Their bet is that simulation, built on models from Nvidia and World Labs with domain-specific libraries on top, can close the sim-to-real gap that keeps robotics development slow and expensive.
In April 2026 Antioch announced an $8.5M seed round at a $60M valuation, led by A* and Category Ventures with MaC Venture Capital, Abstract, Box Group and Icehouse Ventures participating. Early engagements included large multinationals already investing in robotics, and MIT CSAIL researcher David Mayo described using the platform to let LLMs design robots and benchmark them against each other in simulation.
What has to be true
- The case documents a clear contrarian bet: that software simulation, not more physical testing infrastructure, is how physical AI gets built — directly testable as the company ships.
- It has concrete, verifiable traction markers: a TechCrunch feature, a $60M seed valuation in under a year, and named early users including an MIT CSAIL researcher.
- The founders' background (Transpose sold to Chainalysis, Meta Reality Labs, Google DeepMind) makes the bet a deliberate repeat of the dev-tools playbook on a new category.
- The 'Cursor for physical AI' positioning shows a generalizable wedge: take an infrastructure problem and sell the layer that makes it unnecessary.
What can be applied
Don't build the physical world — sell the tool that removes it: an infrastructure problem recast as a dev tool rides model progress and makes small teams look like category leaders.
Aftermath
As of 2026-09-02, Antioch is live and selling a cloud simulation platform, with a stated focus on sensor and perception systems for automated vehicles, trucks, farm and construction machinery, and drones. No funding or revenue numbers beyond the April 2026 seed have been disclosed; the company continues to build on third-party world models and says most of the industry still doesn't use simulation at all.
Sources
- This simulation startup wants to be the Cursor for physical AI
- This simulation startup wants to be the Cursor for physical AI
spotted an error? The archive wants to know.
Your turn
You just read one. Describe what you are building, and see who is betting on the same thing.
Free account · 3 free questions · no card