What tech stack does Boston Dynamics use?

At the robotics core, Boston Dynamics uses C and C++ for real-time motor control, embedded firmware, inverse kinematics, and performance-critical modules where latency and determinism are non-negotiable. ROS and ROS2 (Robot Operating System) provide the messaging and middleware layer connecting C++ control nodes with Python-based perception and planning nodes across Spot, Stretch, and Atlas. The Spot SDK — the developer-facing interface for building custom behaviors and integrations — is Python-first, with gRPC and REST APIs for enterprise integration.

On cloud infrastructure, Boston Dynamics is a confirmed AWS Partner. The official AWS robotics blog documents the use of AWS IoT Greengrass 2.0 as the edge runtime deployed on Spot's Compute AI payload — a hardened Linux computer with an NVIDIA GPU that mounts on the robot's back. Amazon SageMaker handles cloud-based model training, and Amazon SageMaker Edge Manager optimizes and packages trained models for deployment back to robots at the edge. This edge-to-cloud architecture enables Spot to run ML inference locally while syncing training data and updated model weights through AWS. Google Cloud DNS is also detected on the public website, suggesting a dual-cloud posture, though AWS is the primary confirmed partner for robotics infrastructure.

With the CES 2026 Atlas launch and the Google DeepMind partnership, Boston Dynamics is also integrating Gemini foundation models into Atlas's behavioral stack — meaning Google's AI infrastructure (TPUs, Gemini APIs) is now part of the Atlas development pipeline. This represents a meaningful extension of the detected stack beyond what historical public signals show.

Boston Dynamics's public-facing website runs on Drupal CMS with a Java application layer, NGINX as the web server, and Varnish for HTTP caching and acceleration. New Relic is used for application performance monitoring. Google Analytics handles web analytics, and Google Cloud DNS manages DNS resolution — consistent with the AWS/Google dual-cloud footprint common in enterprise robotics companies. These signals are detected via BuiltWith/Wappalyzer and StackShare and represent the marketing/web stack rather than core robotics infrastructure.

For ML model development, Python is the primary language across training pipelines, data processing, and inference scripts. PyTorch, TensorFlow, and MXNet are all referenced in Boston Dynamics engineering materials and the AWS partnership blog as supported frameworks for computer vision, object detection, and reinforcement learning workloads. Orbit (fleet management) and Scout (teleoperation) are Boston Dynamics's own proprietary software products — running on cloud backends that integrate with the AWS IoT/SageMaker stack — providing REST and gRPC APIs for enterprise fleet integration.

On the GTM side, Boston Dynamics is scaling its commercial sales infrastructure ahead of a 2027 IPO. The company is likely evaluating or upgrading CRM, sales engagement, and revenue intelligence tooling to support the growing Spot and Stretch enterprise customer base (2,000+ units deployed, 40+ countries) and to meet public-company investor-relations and reporting requirements.

Boston Dynamics's AWS partnership makes AWS-native tooling the highest-fit category for cloud infrastructure, security, and MLOps vendors. Any platform that integrates with IoT Greengrass, SageMaker, or S3 has a direct architectural fit and a credible integration story. Security tooling for edge IoT device management is a growing need as Spot and Stretch deployments scale to 2,000+ units in enterprise oil-and-gas, logistics, and construction environments — particularly as the installed base expands internationally and compliance requirements vary by region.

The Drupal + Java web stack is aging by modern enterprise standards, creating potential displacement opportunity for headless CMS, React/Next.js, or composable web platforms as the company professionalizes its web presence ahead of IPO. The Python/PyTorch ML stack aligns Boston Dynamics with MLOps platforms (Weights & Biases, MLflow, SageMaker Pipelines) that help manage the model lifecycle from cloud training to edge deployment — a real operational need as Atlas's behavioral AI models grow more complex and must be versioned, validated, and deployed to production robots safely. The Google DeepMind Gemini integration also opens the door for vendors with LLM orchestration, multi-modal inference, or robotics foundation-model tooling.

On the commercial side, the company is scaling a sales organization (Spot and Stretch enterprise accounts, DHL/Maersk/H&M/Gap) while preparing for IPO — meaning CRM, sales engagement, contract lifecycle management, and revenue analytics tools are likely being evaluated or upgraded. The leadership transition (interim CEO, multiple C-suite vacancies) is a tailwind for vendors who can win champions in the remaining senior team and demonstrate fast time-to-value.