PodcastThe Neuron: AI Explained

Inside CoreWeave’s AI Infrastructure With Chen Goldberg

Written by an Ethmos research agent · shared with you

CoreWeave frames AI infrastructure as an integrated system where compute, cooling, security and power — not just GPU count — determine capability and constrain scaling.

ShowThe Neuron: AI Explained

Episode dateSeptember 23, 2026

Runtime40 min

FeaturingChen Goldberg, Corey Noles

Vera Rubin cluster live: CoreWeave brought a multi-rack NVIDIA Vera Rubin NVL72 cluster online, linking hundreds of Rubin GPUs, paired with custom tools Velvi (liquid-cooling valve manager) and Recy (rack manager feeding CoreWeave's Mission Control dashboard).
Infrastructure now IS the product: Goldberg says AI infra can no longer be made boring—compute, network, storage and security must behave as one supercomputer, unlike portable cloud-native workloads.
Kubernetes still dominant but modified: over 90% of CoreWeave's AI workloads run on Kubernetes, but CoreWeave runs it on bare metal with heavier observability to handle multi-node AI clusters.
Customer base broadening: beyond OpenAI and Meta, CoreWeave now serves Caterpillar and physical-AI/robotics customers, and acquired Weights & Biases, OpenPipe, and Marimo to move up the developer stack.
Security burden distributed: Goldberg says CoreWeave treats every employee as playing a role in security, as agent tool access raises risks from both legitimate builders and malicious actors.
Scale metrics disclosed: CoreWeave had roughly 1.5 gigawatts active power and ~3.7 gigawatts contracted as of Q2, with power/energy cited as the next innovation frontier.

Deep dive

The Vera Rubin buildout

CoreWeave's newly announced multi-rack NVIDIA Vera Rubin NVL72 deployment links hundreds of Rubin GPUs into a scale-out system for training, inference, and agentic workloads. Goldberg frames it less as a raw hardware upgrade and more as an expanded innovation surface, where CoreWeave layered in two homegrown tools: Velvi, a valve manager for liquid cooling, and Recy, a rack manager that feeds signals into CoreWeave's "Mission Control" dashboard, giving engineers visibility into how infrastructure performance affects running jobs. She notes best practices carried over from prior large-cluster builds, but says Rubin's dual optimization for both training and inference arrives as customer demand for capacity and experimentation keeps rising.

The full analysis is a free trial away.

An Ethmos agent found this conversation, listened, and wrote this brief. Yours can do the same — across the shows you follow and the thousands you don't.

Read free — start your trial

TMTB readers: 14-day free trial + 25% off your first month

Also inside: 01 Beyond GPU count · 02 Agents, latency, and security · 03 Moving up the stack · 04 Kubernetes' durability and physical AI

This page is an AI-generated summary and analysis prepared with Ethmos and shared by an Ethmos user. It does not reproduce the original programming. The underlying episode and its recording remain the property of their respective creators, and all show and company names are the property of their respective owners.

AI summary. May contain errors. Not investment advice.