Dreamscale Labs GTM brain

Dreamscale sells cloud training and low-latency inference for robotics teams running physical AI models.

category

Robotics cloud inference

market profile

  • text: Large robot foundation models put pressure on on-board GPU budgets and deployment latency.
  • text: Robotics teams need both model fine-tuning and inference close enough to real time for their application.

personas

  • title: Robotics ML lead
    pains:
    • Experimenting with large vision-language-action models is constrained by robot GPU capacity
    • Fine-tuning and deployment require separate infrastructure work
  • title: Robotics infrastructure engineer
    pains:
    • Network and model-serving overhead can break control-loop latency targets
    • Provisioned GPUs sit idle during intermittent experiments
  • title: Robotics startup CTO
    pains:
    • Expensive onboard compute limits hardware design choices
    • Needs a clear cost and performance comparison before moving inference off-device

icp

  • text: Robotics startups and physical AI labs running open-weight or custom robot foundation models; team-size threshold is unconfirmed.
  • text: Teams with connected robots, bursty inference demand and a network path compatible with their control latency requirements.

message pillars

  • text: Use infrastructure designed for real-time physical AI.
  • text: Offload serving and bring a model suited to the task.
  • text: Match spend to actual inference use.

objections

  • objection: Can a cloud connection meet our control loop requirements?
    response: Benchmark end-to-end latency and jitter with the robot, network and model in the intended environment before deployment.
  • objection: What happens when connectivity drops?
    response: Evaluate offline fallback and safety controls with the team; the website does not establish a specific offline guarantee.
  • objection: Will we need to switch models?
    response: The site offers open-weight options and deployment of a customer's fine-tuned model.

competitors

  • name: Reflex
    why: Offers robotics-specific low-latency cloud inference and model fine-tuning.
  • name: Sidekick Robotics
    website: https://www.sidekickrobotics.ai/api
    why: Offers an API for serving physical AI models.
  • name: AnyDevice
    website: https://anydevice.ai/
    why: Optimizes physical AI models on customer hardware, an alternative to cloud offload.

keywords

  • robotics cloud inference
  • VLA model serving
  • robot foundation model inference
  • low latency robot AI
  • robot model fine tuning
  • robot GPU cloud

voice

  • direct and technical
  • confident
  • developer friendly

signals

  • text: A robotics team is testing VLA models such as pi0.5 or MolmoAct2.
  • text: An engineering team is evaluating GPU offload or struggling with onboard compute and intermittent workloads.

stack

  • pi0.5
  • MolmoAct2
  • GPUs

pricing

model: usage
starting price: $5 per robot-hour for inference; $6.84 per GPU-hour for training

motion

hybrid

channels

  • developer website
  • YC launch
  • Product Hunt
  • blog

confidence

medium

Founders

  • name: Eric Ren
    title: Founder/CEO
    role: CEO
    bio: Co-founder @ Dreamscale Labs : realtime cloud inference for robotics.
  • name: Kevin Lin
    title: Founder/Chief Scientist
    role: Founder
    bio: Co-founder & Chief Scientist at Dreamscale Labs (YC F26). 2x IChO silvers. Former quant turned roboticist.
  • name: Antoine Nguyen
    title: Founder/CTO
    role: CTO
    bio: Co-founder & CTO at Dreamscale Labs (YC F26). Previously VLA quantization researcher at UTS.
  • name: Chris Yoo
    title: Founder/CPO
    role: CPO
    bio: Co-founder & CPO at Dreamscale Labs (YC F26). 15x hackathon winner. Youngest CCNA at 13yo.

Back to YC companies