Antioch Raises $32M for Physical AI Simulation

A warehouse does not reset because an engineer changed one line of code. Robotics teams still need the same machine, floor, lighting, sensor stack, people, and awkward edge case to appear again before they can learn whether a change actually fixed the problem.

Antioch is financing a faster answer. The New York physical AI company announced a $32M Series A led by Greylock, with A*, Category Ventures, BoxGroup, and Icehouse Ventures participating. The round gives Antioch more capacity to turn expensive field failures into repeatable simulation tests before hardware reaches production.

What Happened

Antioch announced the Series A on September 8, 2026. Greylock General Partner Saam Motamedi joined the board, and the company said the capital will support product development, engineering hiring, deeper simulation capabilities, and expansion to more physical AI teams.

The financing is also visible in a Form D filed by Valoros, Inc., the legal issuer associated with Antioch. The August 3 filing reported a $31,999,962 equity offering, $31,449,971 sold, eight investors, and a first sale on July 27.

Antioch says the Series A and an $8.5M seed announced earlier in 2026 bring its total funding to $40.5M. Separate December 2025 reporting described a $4.25M pre-seed, and public sources do not explain whether that capital was folded into the later seed figure or excluded from the company's cumulative total. The safest accounting is therefore Antioch's attributed $40.5M, not a reconstructed number that treats every public announcement as a separate pool.

The Product Is a Verifier, Not a Virtual Showroom

Antioch builds a simulation and evaluation environment for robotics, autonomous systems, drones, industrial automation, and perception-heavy products. The platform brings a customer's hardware, sensors, software, models, environments, and operating constraints into a calibrated digital system, then runs scenarios at cloud scale.

That lets teams test a perception update, planner change, sensor configuration, or mechanical redesign across thousands of combinations before sending it into the field. Engineers can reproduce known failures, search for adjacent edge cases, compare model versions, generate synthetic data, and connect evaluations to a continuous integration workflow. The commercial promise is a development loop that behaves more like software without pretending the physical world is simple.

Antioch combines explicit models of geometry, kinematics, hardware, sensors, and physical constraints with learned components trained on real-world data. Its platform integrates NVIDIA Omniverse libraries, Isaac Sim, and Isaac Lab, while Nebius provides cloud infrastructure. The company describes this as a real-to-sim-to-real loop in which field results keep improving the simulator rather than sitting in a postmortem document.

Why Physical AI Needs the Feedback Loop

Model progress has made robots and autonomous systems more capable, but every additional capability creates more combinations to validate. Field programs consume equipment, facilities, engineer hours, and calendar time, while rare failures may refuse to appear on command. A faster model can therefore push a team into the same validation constraint sooner.

Simulation has always promised leverage, but teams hesitate when fidelity is too low or the tooling sits outside normal engineering work. Antioch is betting that customer-specific calibration, cloud-scale evaluation, and repeatable scenario management can move simulation into the daily release process. That makes the test itself a durable engineering asset: each failure becomes a scenario that future builds must pass.

The early evidence is direct but still company-reported. Amazon's Ring says Antioch's simulations closely matched physical test results, including scenarios deliberately withheld from calibration. Launchpad Build AI says the platform helped it test production scenarios and otherwise impractical edge cases across teams on both sides of the Atlantic. Those references establish real use, while leaving broader customer scale and independent performance measurement undisclosed.

The Team Behind Antioch

Antioch's founding record combines autonomy engineering with a previous company exit. Harry Mellsop worked on computer vision for Tesla Autopilot, and Mellsop, Alex Langshur, and Michael Calvey previously built Transpose, which Chainalysis acquired. Collin Schlager brought simulation, machine learning, and hardware experience from Meta Reality Labs, while Colton Swingle worked at Google DeepMind.

An August 2026 SEC filing identifies Alexander Langshur as CEO. Current official and investor sources consistently describe Harry Mellsop as a co-founder, although some secondary profiles use a conflicting CEO label; no current CTO was verified. That distinction matters because the public story has moved quickly enough for stale profiles to outlive the role they were trying to describe.

What the Series A Changes

Antioch is hiring in New York across simulation, core platform, customer deployments, product marketing, and operations. The round can increase the engineering depth required to integrate unfamiliar hardware, improve fidelity, run more evaluations, and help customers turn field data into repeatable test suites.

The market consequence is bigger than a faster simulator. Physical AI teams need a trustworthy way to decide whether one proposed change improves the entire system before the cost of being wrong arrives in a warehouse, on a road, above a worksite, or inside a customer's home. Related investments in AI physics, autonomous infrastructure engineering, and robotics deployment show capital moving toward the systems that connect model intelligence to physical outcomes.

Antioch now has to make simulation accurate, usable, and repeatable across customers whose hardware and failure modes do not resemble one another. The value of the Series A will show up each time a bad field event stops being a story an engineer remembers and becomes a test every future release has to survive.