Grey Materia Explore

Infrastructure purpose-built for Physical AI

For your most demanding E2E robotics workloads.

Explore the platform
A blue architectural ink drawing of four connected levels: GPU infrastructure supports software, simulation and policy training, with an athletic, helmeted humanoid dribbling a basketball on the top court. The same narrow visor, ivory chest plate and athletic proportions appear in the development views below.

Use casesHumanoid robotics
Basketball task

WorkloadsSimulation · Data
Policy training
Evaluation

SoftwareScene & robot assets
Simulation environments
Orchestration

InfrastructureHGX + RTX
Shared storage
Networking

Infrastructure built around the workload.

Simulation / Data / Training / Evaluation / Deployment

Different jobs.
One continuous workflow.

Our fleet is balanced across HGX and RTX architectures. A shared foundation from scene construction to deployment.

Build scenes with workstation-class responsiveness, on demand

Build scenes in the cloud with the responsiveness of local hardware. RTX PRO 6000 Blackwell GPUs give you a fully ray-traced (RT) viewport you can manipulate in real time — not the degraded preview you get from traditional AI infra lacking RT cores. Your simulation stack, assets, and saved scenes persist between sessions, so you can spin down compute to save costs and pick up where you left off.

A navy architectural ink illustration of an indoor basketball simulation with the athletic helmeted humanoid, an orange basketball, padded walls, lights, bench, ball rack and hoop. These assets assemble digitally in the animation.
Scene assets · RTX viewport · Persistent workspace

02 / 05

Accelerate synthetic data generation with hyperscale infrastructure and economics

Replicate your scene across multiple RTX PRO 6000 nodes based on the scale of your data generation needs, operate your own randomization logic, and release the compute capacity when the run finishes. Every node writes to the same NVMe namespace, so what comes out of generation is already sitting where training reads it — no consolidation step, no copying terabytes between systems.

An ivory-and-navy humanoid performs basketball moves in an indoor gym. Each move is captured as a snapshot that joins a growing strip of synthetic data examples above the scene.
RTX PRO 6000 · Synthetic data · Shared NVMe

03 / 05

Scale physical AI training, across foundation-model pre-training and RL post-training

Pre-train world models and vision-language-action (VLA) models, then adapt robot policies through reinforcement learning (RL). Choose Blackwell and Rubin HGX configurations for your model and training scale. Bring together RTX PRO 6000 simulation, policy inference, and training capacity for RL post-training, with resources sized to each workload and released when the run finishes.

An ivory-and-navy humanoid misses a short basketball shot, absorbs a stream of model parameters, then uses improved form to make a flaming jump shot.
World and VLA pre-training · RL post-training · Blackwell / Rubin HGX

04 / 05

Put every checkpoint back into the world.

Evaluate trained policies across controlled simulation scenarios. Pair RTX sensor rendering with the policy compute each workload needs, and preserve the results with the checkpoint.

Checkpoint evaluation
Put every checkpoint back into the world. Checkpoint evaluation, illustrated in blue and ivory.
RTX sensor rendering · Workload-matched policy compute

05 / 05

Validate on the hardware you’ll deploy.

Run containerized tests on physical Jetson devices. Compare model quality after quantization, measure runtime behavior, and exercise the target pipeline with simulated sensor inputs.

Physical Jetson tests

Build and test in the cloud.
Run inference onboard.

Validate on the hardware you’ll deploy. Physical Jetson tests, illustrated in blue and ivory.
Target optimization · Container runtime · Hardware tests

Scroll to follow the workflow

Spend more time
improving the model.

Keep the workload moving, from the first simulation to a build ready for the robot.

A continuous infrastructure loop connecting a shared dataset, cloud compute and target-hardware testing
Faster iteration
Reuse data and checkpoints as you move through the loop.
Less infrastructure work
Start with compatible environments for specialized workloads.
Fewer data transfers
Work from a shared, versioned data foundation across compute pools.
Confidence on target
Measure model quality and runtime behavior on physical hardware.

Physical AI workloads.

Enable robotics foundation model and world model training. Compute, data, and validation matched to each application.

Humanoid robot hands positioning a connector in a laboratory fixture

Humanoids

Train whole-body and manipulation policies from demonstrations and simulated experience. Evaluate checkpoints across tasks and robot configurations.

  • Parallel simulation
  • Policy training
  • Hardware validation
Industrial gripper positioning a machined component in a fixture

Industrial robotics

Generate sensor data for assembly and machine tending. Train manipulation policies and test precision across fixtures, parts, and contact conditions.

  • RTX data generation
  • Shared training data
  • Checkpoint evaluation
Autonomous mobile robot carrying a tote through a warehouse junction

Warehouse robotics

Train navigation and picking policies across warehouse layouts. Evaluate perception and multi-robot coordination under changing traffic and inventory.

  • Scene variation
  • Fleet simulation
  • Policy evaluation
Cameras and lidar on a test vehicle facing a road intersection

Autonomous vehicles

Train perception and planning models on recorded and synthetic sensor data. Evaluate rare scenarios and measure inference performance on target hardware.

  • Sensor pipelines
  • World model training
  • Target validation
Agricultural robot sensors inspecting plants beside a crop row

Field robotics

Train outdoor navigation and manipulation policies. Test against variations in terrain, lighting, vegetation, and weather.

  • Synthetic data
  • Policy training
  • Scenario testing
Drone camera inspecting the surface of a concrete bridge support

Inspection & aerial robotics

Train perception models for defect detection and policies for autonomous inspection. Evaluate sensor coverage, navigation, and inference constraints.

  • Dataset preparation
  • Model training
  • Edge validation

Explore the infrastructure behind these workloads.

Explore the platform