Oracle founder Larry Ellison admitted during the company’s most recent earnings call that he had to beg NVIDIA CEO Jensen Huang to supply the company with its latest GPUs (via Barron’s): “I had dinner with Elon Musk and Jensen Huang at Nobu Palo Alto. You could describe that dinner as me and Elon begging Jensen for GPUs. Take our money. No, take more. You’re not taking enough. Take more of our money. Please. It worked. It worked,” Ellison said on the conference call.
$ORCL CTO Larry Ellison: “I went to dinner with @elonmusk and Jensen Huang. And that dinner was Oracle and me and Elon begging Jensen for GPUs. Take our money. No, no, take more. You’re not taking enough. Take more of our money.” pic.twitter.com/atBUJaULCSSeptember 14, 2024
The dinner was very productive and appears to have been money well spent for Ellison and Oracle, who recently announced the creation of a Zettascale AI supercluster consisting of 131,072 Nvidia GB200 NVL72 Blackwell GPUs, delivering 2.4 ZettaFLOPS of AI performance — more powerful than Musk and xAI’s Memphis Supercluster, which currently features 100,000 Nvidia H100 AI GPUs.
Oracle’s AI plans require vast amounts of electricity, which is why the company already has permission to build three modular nuclear reactors to meet the facilities’ power needs. But installing reactors at data centers would likely take years, so in the meantime the company could follow Musk’s lead and use giant mobile generators to bolster local power supplies when needed.
Despite being smaller than other large technology companies offering data center services, such as Amazon Web Services, Microsoft Azure, and Google Cloud, Oracle Cloud Infrastructure (OCI) has a distinct advantage over these giants. According to a Barron’s report, OCI offers greater flexibility and can meet the specific requirements of some customers. It can also provide offline servers that run on its own network infrastructure to maximize security for the most demanding clients.
But despite its size, Oracle is forging ahead with its AI investments. Ellison said it will cost $100 billion to train the most advanced AI models that will emerge over the next three years, echoing Anthropoc CEO Dario Amodei’s thoughts on the subject. And OCI seems keen to be one of the leaders in AI processing. “There’s always going to be somebody who’s better than everybody else in this space. There are multiple people trying, and it’s a race,” Ellison said on the conference call. He later added, “It’s a big deal to get there first.”
