Launching today
Dreamscale Labs
Run the best robot AI models over the cloud
22 followers
Run the best robot AI models over the cloud
22 followers
Dreamscale Labs (YC F26) is building the real-time cloud inference platform for robotics. We serve frontier physical AI models (e.g. VLAs, WAMs) over the cloud with ultra-low latency. Instead of purchasing expensive GPUs, you can run large SOTA models using on-demand cloud inference!
Hey Product Hunt 👋
We started working on Dreamscale because of one observation: frontier physical AI models becoming more compute-intensive to run, while robots are bottlenecked by small on-board GPUs. It's a painful process to quantize and distill these large models to fit on these GPUs, and you lose quality in the process. Using existing cloud providers means you still have the load the weights and write up an inference server for each new model you want to try.
Dreamscale is an inference API that lets you easily run the best robot AI models over the cloud, with real-time latency. It's as simple as:
Choose the model
Choose the target robot
Write the instruction, and run it!
We achieve <100ms end-to-end latency for many models, with a flat hourly-based pricing.
As the state-of-the-art keeps moving rapidly in Physical AI, we felt there was a need for deployment, research and evaluation companies to have an easy way to use and experiment with the latest physical AI models.
What's one model that you'd like to run on your robot?
Try it out!!
Thumbs up
Excited for folks to try :)