Elon Musk’s xAI won’t be renewing its server rental agreement with the cloud storage giant Oracle, reportedly after the firm deemed Musk’s specification for a supercomputer it was building him technological impossible.
Per Reuters, xAI had been renting Nvidia AI-equipped H100 graphics processing units (GPUs) from Oracle, but, in a tweet on Twitter, Musk said the company was now building its own system to ensure ‘[the] fastest time to completion’.
xAI and Oracle
Reuters was told by a source close to Oracle that the server capacity it rented out to xAI had already been reassigned to another company.
But this is no secret. xAI’s twitter account (via The Information) discussed last month’s news that Oracle have signed an agreement with Microsoft to provide Nvidia-powered servers for OpenAI workloads.
xAI and Oracle do still have an agreement to allow xAI to train its AI models in Oracle’s cloud infrastructure, and, per that X tweet (we’re not made of money), xAI still rents around 16,000 Nvidia chips from Oracle.
Said tweet also confirms xAI’s plans to build its own data center in Memphis, Tennessee powered by Nvidia chips bought directly from Dell and Supermicro, plus a 100,000 GPU strong supercomputer for the third version of Grok, Twitter’s very own ‘spicy’ AI chatbot.
Musk plans for the second version of Grok to launch next month.
More from TechRadar Pro
https://cdn.mos.cms.futurecdn.net/UhpgU7eT2dhfTc52jvyvM-1200-80.jpg
Source link
luke.hughes@futurenet.com (Luke Hughes)
