How IBM Built Their 6,000 GPU Fortress

0:43

IBM Research engineered a massive H100 cluster from the ground up to train Granite Foundation models. Inside the infrastructure bet that's reshaping enterprise AI.

Indexed Metadata

SEO / AIO

TypeYouTube Clip

Duration43s

Transcript115 words

Sources1

Qualityacceptable

YouTube Clip

Manifest JSON Transcript TXT

Source Context

Building a 6,000 GPU Cluster: The Reality of AI Infrastructure

Create your own short

Generate a short with captions, transcript metadata, and a public video page when you choose to share it.

Create from a prompt, source, or feed

Attach transcript and manifest data

Publish or keep private

Statuscompleted

CreatedOct 2, 2026

Transcript

115 words

So I run all of our AI infrastructure at IBM Research, or my team does. But we have been, our main task lately has been, or since the last few years, has been training the Granite Foundation family of models. That was what gave me, hey, go build out our infrastructure. So we went out and built a large H100 cluster. And we did it ourselves, got the space, souped the nuts. It was a huge task. And what was large? This was 6,000 GPUs, but this was Dell's second XE9680 deployment, so it was Israel 1. We came out, and then they went up to Tennessee for that other very large cluster. Memphis. Yeah.

Discover Pricing Login

Product

Solutions

Resources

Earn

Tools

Legal