Musk’s Rivals Take Jabs at His GPU Cluster, Nvidia’s Cuda Weakness, Enterprise AI Use Cases: Summit Highlights

I’m still buzzing from getting to meet so many of you yesterday at our first-ever AI Summit. We got to talk all things LLMs, chips and AI applications with leaders such as Reid Hoffman, one of the earliest investors in OpenAI and a Microsoft board director; Chris Lattner, who helped Google launch their tensor processing units and TensorFlow software; and Naveen Rao, the former CEO of MosaicML, whose acquisition by Databricks was one of the first big generative AI deals.

Elon Musk might not have enjoyed yesterday’s summit as much as I did, though, as the X owner’s AI ambitions were the subject of more than a few not-so-subtle jabs during talks. During my conversation with Hoffman, the OpenAI investor let some of the air out of Musk’s “Colossus cluster” balloon—the 100,000-H100-strong supercomputing cluster Musk says he completed in just four months. (That may or may not be completely true, as we discussed on Tuesday.)

Hoffman said that such a cluster is “table stakes” for the kind of conversational AI Musk and his model developer, xAI, wants to pursue. In other words, the cluster—despite being the biggest of its kind—only helps him catch up to rivals like OpenAI and Anthropic.