What tech stack does 1X Technologies use?

1X's core robotics software is written in C++ and Python with ROS2 (Robot Operating System 2) as the middleware layer for robot communication, sensor fusion, and control. The company's public GitHub organization (github.com/1x-technologies) shows active repositories for ROS2 examples (eve-ros2-examples, with both C++ and Python implementations), rigid-body dynamics (a fork of the Pinocchio C++ library from stack-of-tasks for robot kinematics), and URDF robot model definitions used in simulation and hardware-in-the-loop testing.

Infrastructure tooling uses Rust, Go, Docker, Ansible, and Kubernetes — signals sourced from job postings including a Research Engineer, Data Infrastructure role in Palo Alto that specified vector databases, Ansible, and distributed systems experience. This is consistent with a company managing large-scale robot training data pipelines, OTA model update delivery to thousands of NEO units, and edge deployment at scale. A TypeScript codebase visible in the material-tracking repository suggests internal operational and logistics tooling is being built web-natively. Unity and C# packages in the public GitHub organization (halodi-controller-unity-packages) indicate Unity is used for robot simulation and policy validation.

On-device compute in NEO's head uses NVIDIA Jetson Thor, which runs the Redwood AI model and vision-action loop at 5Hz for real-time manipulation and locomotion decisions without cloud round-trips.

1X's production AI stack is proprietary and on-device. Redwood AI is a 160M-parameter vision-language transformer that runs directly on NEO hardware via NVIDIA Jetson Thor, enabling autonomous manipulation and locomotion tasks without cloud inference latency. The World Model is a physics and video prediction system trained on thousands of hours of video data collected from deployed robots — it predicts how the physical world will respond to robot actions, enabling NEO to reason about novel environments without explicit physics engine assumptions.

1X open-sourced 1xGPT (558+ GitHub stars), a world modeling challenge that invites external researchers to improve prediction of future video frames from humanoid robot data — simultaneously a research artifact and a talent signal. In March 2025, 1X collaborated with NVIDIA to demonstrate NEO Gamma performing autonomous domestic tidying tasks using a policy built on NVIDIA's GR00T N1 open foundation model, trained on data from 1X offices and employee homes via a dedicated dataset API. The 1X AI team built an inference SDK serving GR00T N1 predictions at a continuous 5Hz vision-action loop using the onboard NVIDIA GPU in NEO's head.

In June 2026, 1X established the World Model Lab under Samarth Sinha (ex-founding researcher at Luma AI, expert in scaling multimodal generative video models), signaling a major strategic bet on large-scale embodied world model pretraining as the path to full autonomy. CEO Børnich has explicitly stated that 'You can't fine-tune your way to AGI,' positioning the World Model Lab as the primary research differentiator against competitors that fine-tune general-purpose language models for robotics.

1X is a Python/C++/ROS2 robotics company that deliberately keeps critical inference on-device — meaning cloud inference vendors face a build-vs-integrate dynamic. The better infrastructure conversations are around GPU training compute (the World Model Lab and Redwood AI require significant cloud GPU capacity to train even if inference is on-device), data labeling and annotation platforms at humanoid-specific scale, and robotics simulation software that can interface with ROS2 and URDF pipelines.

The Kubernetes, Docker, and Ansible signals mean modern DevOps and platform vendors (observability, secrets management, infrastructure-as-code, CI/CD for robotics) have a natural technical audience in the infrastructure team. With 1,049 employees as of April 2026 and rapid growth toward likely 2,000+ in 2027, 1X is a live target for HRIS, payroll, finance and ERP systems, and enterprise collaboration tools at exactly the Series B-to-C scaling inflection point. The Hayward NEO Factory also opens procurement for manufacturing execution systems, quality inspection software, and supply chain tooling appropriate for a 10,000-unit annual run that must scale toward 100,000.

Vikram Kothari (VP of Operations) is the likely senior contact for manufacturing, supply chain, and factory operations software. Samarth Sinha (Head of World Models) and the World Model Lab team own AI/ML infrastructure purchases. CFO Mustally Hussain controls overall budget allocation and is the right contact for any contract with multi-million dollar annual commitment.