From 6115b72a243cffa6a4b362d4aa7ce0033074c0a4 Mon Sep 17 00:00:00 2001 From: Mantra Date: Wed, 11 Feb 2026 01:55:32 -0500 Subject: [PATCH] Add Antioch deal memo --- antioch.md | 83 ++++++++++++++++++++++++++++++++++++++++++++++++++++++ 1 file changed, 83 insertions(+) create mode 100644 antioch.md diff --git a/antioch.md b/antioch.md new file mode 100644 index 0000000..f8c36c9 --- /dev/null +++ b/antioch.md @@ -0,0 +1,83 @@ +# Investment Memo for Antioch + +**Recommendation:** Root should pursue leading or co-leading Antioch's Seed round. + +**Company:** Antioch (antioch.com) + +**Stage:** Seed (Pre-seed of $4.25M raised December 2025, led by A*) + +## Overview + +Antioch is building a cloud-based simulation platform where teams can build, test, simulate, and validate robots entirely in software — the "CI/CD for physical robots". The robotics market is expanding rapidly, but the sim-to-real gap remains a barrier for much of robotics and autonomous system development and deployment. Each change requires real world testing, with time consuming and costly debugging. With backgrounds from Tesla Autopilot, Deepmind, Reality Labs, the team at Antioch is building a product solution from experience, and is essentially taking an internal tool from a massive company (Tesla's simulation stack) rebuilt as a product for the teams that can't afford to build it themselves. + +This company hits on two of Root's investing themes: frontier technology and engineering tools with outstanding founders, has a huge and expanding market, and Root has existing portfolio companies able to verify the value of their offering (Dusty Robotics, Tortuga, Veo Robotics). + +## Product + +Instead of teams bottlenecking on scarce physical hardware, each engineer can spin up their own fully configured robot environment in simulation, tied directly to their local dev workflow. When a pull request is opened, Antioch can provision an isolated digital twin of the robot and its environment for code review, QA, safety validation, and PM signoff, with shareable links across the org. + +This makes robot development look like modern software development: deterministic environments, parallel testing, CI/CD gating, and fast iteration. Antioch sits above physics engines and observability tools, acting as the workflow layer that orchestrates simulation, testing, and deployment across teams. + +- **Antioch Ark** — Container for the robot hardware and software stack into a single, version-controlled artifact. The company claims close to deterministic performance: helping close the sim-to-real gap. +- **Antioch Cloud** — Digital twins running in parallel. Engineers can run tests against simulated environments and integrate with standard CI/CD workflows, so robotics teams can mimic development speed of software teams. +- **Ecosystem integrations** — NVIDIA Omniverse/Cosmos for physics simulation and Foxglove for observability. (Foxglove CEO is an angel investor) + +## Why Now + +The robotics market is exploding. Startups like Figure, 1x, Agility, are seeing huge investment, and we are seeing pushes of robotics and autonomous systems into construction, ag, even more vehicles. But for the most part, the bottleneck is physical testing and development. Figure recently partnered with Brookfield to access their properties for training data creation. + +At the same time, embodied AI is actually making robots harder to test efficiently. Moving from rules-based control to non-deterministic or onboard foundational models (a trend I expect to continue) has increased risks and requires more extensive testing as robots are put further into novel environments. + +Finally, we are seeing the first major re-shoring of US manufacturing in half a century. This reshoring needs to include autonomous systems to be economically viable, which should add upward pressure to the need for efficient development in robotics/autonomous systems workflows. + +## Market + +Robotics companies currently rent Airbnbs to test household robots overnight, or spend millions building physical testing environments. Major players like Tesla, Waymo, and Anduril spend tens of millions per year on system evaluation using internal tooling. + +The robotic simulator market is projected to be a $3 - 4.8B+ by 2033, 15-18% CAGR, but Antioch's bet is that it can essentially create a broader devOps/CI market applied to robotics, which barely exists as a category today. I believe the number of companies building robots is going to grow dramatically in the next decade and the overall adoption of autonomous systems across the board will be strong (manufacturing, retail, household). There will be a real need for more rapid robotics development pipelines. Antioch is building that future infrastructure. + +## Team + +**Founders:** + +- **Harry Mellsop** — Co-founder. Former Tesla Autopilot vision team. Previously co-founded Transpose (blockchain data platform, acquired by Chainalysis 2023). Stanford. +- **Alex Langshur** — CEO. Former Head of Product at Deep Grey Research. Co-founded Transpose with Mellsop. Stanford. +- **Colton Swingle** — Co-founder. Former Google DeepMind, led large-scale model validation infrastructure projects. Stanford. +- **Collin Schlager** — Co-founder. Former Meta Reality Labs, foundational research for Neural Band wearable. Stanford. + +## Competition + +**Will NVIDIA build this?** + +NVIDIA already has Isaac Sim and Omniverse, which is the best-in-class physics/rendering/synthetic data platform layer. Antioch is integrating the developer workflow with these platforms, so not directly competing, but there is risk that NVDIA could expand offering down the development chain. + +**Gazebo** + +This is the academic default. Gazebo is powerful and widely adopted in research, but it's desktop-first, not cloud-native, and weak on orchestration, reproducible environments, and large-scale parallel testing. It is mostly for labs. + +**Internal tools** + +Tesla, Waymo, Anduril all have world-class internal simulation and development processes. These are proprietary, built over years with massive engineering investment, and unlikely to be commercialized by the companies themselves. It does provide a risk that as development costs go down with AI, more companies will build internally. + +## Risks + +- **Sim-to-real is genuinely hard.** The fundamental challenge of robotics simulation is that simulated physics never perfectly matches the real world. Antioch's "Ark" containerization claims deterministic sim-to-real behavior, which would be a genuine differentiator, but is a key technical bet. +- **NVIDIA could move down the stack.** If NVIDIA decides to build the CI/CD and workflow layer on top of Omniverse, Antioch's integration advantage becomes a dependency risk. +- **Market timing.** The robotics boom could cool, or consolidate around a few large players who build their own testing infrastructure. If the market bifurcates into 5 large robotics companies (who build internally) and a long tail that never reaches scale, Antioch's addressable market shrinks. +- **Pre-revenue-ish.** Antioch has Fortune 500 pilot customers but no disclosed revenue in their pre-seed round. I would expect revenue targets and pilot outcomes as they prepare for a seed. + +## Root Fit + +Antioch is a Root deal for several reasons: + +- **Engineering risk is the primary risk.** Sim-to-real transfer is a hard technical problem, not a go-to-market or business model puzzle. Root underwrites engineering risk — this is exactly the kind of company where Root's technical diligence adds the most value. +- **The founding team is engineers.** Four Stanford-trained engineers from Tesla, DeepMind, and Meta. Root invests in technical teams at the earliest stages. +- **Portfolio synergy.** Root's existing portfolio companies (Dusty Robotics, Tortuga, any company touching hardware + software) are potential Antioch customers. Root could actively drive adoption. +- **It's a developer tool for the physical world.** Root's two themes — frontier tech and engineering tools — converge here. +- **Check size alignment.** A $2-3M check into a Seed round would be Root's standard play and could give Root a lead or co-lead position. + +## Why I Wrote This + +I'm applying for the Associate role at Root. + +I found Antioch through the kind of sourcing I'd do as a Root Associate: scanning recent pre-seed announcements in robotics infrastructure, filtering for technical founding teams, and matching against Root's existing portfolio.