AI Engineer
← All organizations

GPU cloud infrastructure for AI development and inference

RunPod

RunPod provides cloud GPU infrastructure for developers to build, train, fine-tune, and deploy AI models. Its three core products cover different workloads: Pods provide persistent compute for development, Serverless supplies autoscaling inference endpoints that scale to zero when idle, and Clusters support distributed training and large-batch inference. Independent researchers and teams building frontier models use the platform, bringing their own containers, frameworks, and code.

Launched in 2022, RunPod grew from cryptocurrency-mining hardware repurposed by co-founders Zhen Lu, now CEO, and Pardeep Singh, CTO. Its engineering approach reduces deployment work and repeated initialization: Flash packages Python code and dependencies separately from cached base images, avoiding full image rebuilds. FlashBoot pauses containers while retaining model weights in GPU memory, allowing them to resume without reloading the model when warm capacity remains available.

By June 2026, the company reported more than one million developers and over 20 billion cumulative Serverless inference requests. Its reported annualized revenue run rate reached approximately $240 million. That June, RunPod announced $100 million in growth financing led by Summit Partners at a $1 billion valuation, with capital intended to expand its platform, engineering team, and global developer access.

www.runpod.io ↗

2 talks

Newest first

1 speaker at AIE

Affiliations reflect their AIE appearances, not necessarily current employment.

Audry HsuSoftware engineer

Start here

  1. Under 5 minutes to a deployed LLM endpoint — Audry Hsu, RunPod

    Start here for Hsu’s comparison of Pods, autoscaling Serverless workers, multi-node Clusters, and the repository-backed Hub.

    Audry HsuAI Engineer Europe 2026

  2. GPU Cloud Deployment Without Leaving Your IDE — Audry Hsu, RunPod

    Use this demonstration to understand how Flash divides execution between local development code and remote GPU functions.

    Audry HsuAI Engineer Europe 2026

Messages from the stage

Endpoint configuration beyond model selection

The console demonstration covers selecting a Hub listing, configuring H100 GPUs with A100 fallback, and setting worker scaling before deploying an LLM endpoint.

Shortening the GPU development cycle

Hsu contrasts Flash’s decorated Python functions with repeated GitHub, Docker, registry, and GPU-provisioning cycles, then demonstrates multi-model image composition.

Affiliations reflect each recorded session, not necessarily current employment.

Company sources · checked 2026-08-27