Donny Greenberg, Runhouse | theCUBE + NYSE Wired: Media Week - Cyber & AI Innovators Summit
summary
Argues that inference tooling is largely solved while training at scale isn’t, and offers a platform that takes an existing Python training loop, lets engineers declare X GPUs, and auto-distributes code across multi-node GPU clusters with emphasis on fault-tolerant multi-node GPU execution. Mentions integration paths (Ray, DeepSpeed, Megatron, PyTorch), targets hybrid multi-cloud/on‑prem setups, and frames the core win as cutting months of infra lift so teams can pull the ML “slot machine” lever far more often.
AI-written summary · about
chapters (12)
- 0:00Welcome to theCUBE
- 0:28Introduction to Donny Greenberg
- 1:10Machine learning pre-Gen AI
- 3:16Infra as the pacing car
- 6:28Hadoop era of machine learning
- 8:55Runhouse core thesis
- 11:22Productionizing AI systems
- 12:24Defining systems mindset
- 17:51Data and infra bottlenecks
- 20:42Donny's journey to Runhouse
- 22:52Advice for aspiring ML engineers
- 26:20Closing remarks
