Dreamscale Labs

The real-time cloud for physical AI

Fast, reliable, and private infrastructure for robot training and inference.

Y Combinator NVIDIA Inception

Ridiculously easy fine-tuning.

  1. Choose a model
  2. Bring a dataset
  3. Start training

Connect your robot.
We handle the compute.

Dreamscale optimizes across the entire stack: network, model and GPUs to maximise throughput and minimise latency.

Bring your own fine-tuned model with Dreamscale Deployments, or choose a model off the shelf.

Deploy a model

Explore available open-weight models.

Run frontier robot foundation models on Dreamscale.

NVIDIAAi2Verne RoboticsRobbyantXiaomiPhysical IntelligencePerceptronOdyssey
Request access ↗

Pay for what you use.

Inference

$5 per robot-hour

  • Billed by the second
  • Only while your robot is connected
  • Any model, or your own deployment
Deploy a model

Training

$6.84 per GPU-hour

  • Exact price shown before you start
  • Set a budget for any run
  • Pre-configured recipe, or tune your hyperparameters
Start training

New accounts start with $5 of credit.

Get started ↗