HomeCompaniesDreamscale Labs

Running robot brains in the cloud

Dreamscale Labs is building the fastest cloud inference platform for robotics. As complex physical AI models become difficult to fit on expensive on-robot GPUs, we let users offload inference to the cloud and serve them with real-time latency. So far, we've hit 40-50% lower latency than existing neoclouds.
Active Founders
Eric Ren
Eric Ren
Founder/CEO
Co-founder @Dreamscale Labs, building fast cloud inference for robotics. Prev. AI safety.
Kevin Lin
Kevin Lin
Founder/Chief Scientist
Co-founder & Chief Scientist at Dreamscale Labs (YC F26). 2x IChO silvers. Former quant turned roboticist.
Antoine Nguyen
Antoine Nguyen
Founder/CTO
Co-founder & CTO at Dreamscale Labs (YC F26). Previously VLA quantization researcher at UTS.
Chris Yoo
Chris Yoo
Founder/CPO
Co-founder & CPO at Dreamscale Labs (YC F26). 15x hackathon winner. Youngest CCNA at 13yo.
Company Launches
Dreamscale Labs: Running robot brains in the cloud
See original launch post

TL;DR: We're 40-50% faster than existing cloud providers in serving frontier robot AI models, e.g. DreamZero, MolmoAct2.

https://www.youtube.com/watch?v=jVqvJFBRcd0

[1x speed, autonomous]

DSL is building the fastest cloud inference platform for robotics. We let you deploy physical AI without needing expensive local GPUs.

Robot foundation models are growing more and more computationally expensive. The latest frontier of robotics no longer fits feasibly on local GPUs. Compute costs keep soaring amidst a global LPDDR shortage; NVIDIA just doubled the prices of Jetsons in July.

We're taking a contrarian bet that the future is moving inference off the robot and into the cloud.

uploaded image

How do we do this? We work across the full stack of robot inference to push latency down:

  • Model-level architectural adjustments for faster inference
  • Kernel-level optimizations to maximally leverage server-grade GPUs
  • Network-level optimizations for fast and reliable transport over the cloud
  • Region-specific compute allocation to ensure the shortest physical distance between the servers and our customers' robots

These optimizations enable us to be 8% faster than NVIDIA's implementation at half the GPU count (320ms vs 350ms, for DreamZero 14B).

uploaded image

Furthermore, something we noticed for robotics companies with fast R&D cycles was that their GPUs had long idle periods with concentrated bursts of inference traffic. We're one of the first companies in the world to offer on-demand pricing for physical AI inference, giving users more flexibility with the cost.

We're already working closely with a range of early partners across deployment, research and evaluation. In full FDE spirit, we collaborate with our customers to custom-fit inference stacks that address their unique deployment challenges. We help you get your robot AI to deployment faster.

Our mission is to democratize access to frontier physical intelligence, and we believe cloud is the way forward. You can read our thesis here:

•⁠ ⁠http://dreamscalelabs.com/blog/why


Our ask: Reach out at founders@dreamscalelabs.com !

Dreamscale Labs
Founded:2026
Batch:Fall 2026
Team Size:4
Status:
Active
Location:San Francisco
Primary Partner:David Lieb