Oracle CTO Larry Ellison revealed his company provided Nvidia GPUs for Elon Musk “to bring up the first available version” of its AI model Grok – saying “they got that up and running but boy did they want a lot more GPUs than we gave them; we're in the process of getting them more.”
(On December 7 Musk said Elon Musk said his AI startup xAI is rolling out Grok for Premium+ subscribers of social media platform X.)
Ellison made the comments in a Q&A with analysts late Monday, during a Q2 earnings call – as Oracle reported quarterly revenue of $12.9 billion and as CEO Safra Katz said that cloud revenue run rate is $20 billion “and cloud services demand continues to grow at unprecedented levels.”
Ellison, a former friend of the controversial world’s richest man, fell out with him over AI safeguards – according to Walter Isaacson’s Musk biography. (In a podcast last month, Musk said, "I would like to be friends again with Larry. I haven't seen him in ages. We were friends for a very long time. The breaking of the friendship was over OpenAI," he added.)
But Musk’s need for hardware to train its new large language model saw him tap Oracle for help, Ellison suggested. He made the comments as Oracle claimed that demand was outstripping supply when it came to the company’s cloud services, with Oracle CFO Safra Katz saying on the fiscal Q2 earnings call that the company had “missed out on “hundreds of millions of dollars that we would have been able to recognize if our capacity was available” – as it continues to build out its cloud footprint.
Although economic headwinds mean many CIOs are tightening their belts when it comes to cloud spending, Oracle says its OCI cloud expansion (backed by a $9 billion CapEx figure this year) meant that it was scrambling to secure all the hardware it needed as it eyed the opportunity, including for cloud migrations of the widely used Oracle Database. (Per Katz, on the earnings call: “... not a toy, it's a mission-critical system. If it just disappeared at companies, the whole planet would come to a standstill.")
Ellison added: "There's this gold rush toward building the world's greatest large language model. And we are doing our best to keep – give our customers what we can this quarter and then dramatically increase our ability to give them more and more capacity each succeeding quarter... In the next few months, we will turn on 20 new Oracle Cloud data centers co-located with and connected to Microsoft Azure as a part of our joint multi-cloud initiative. These 20 new multi-cloud data centers will house over 2,000 full racks of Exadata Database Machines, designed to meet pent-up demand for the Oracle Cloud database" he said on the call.
See also: Opinion - The Big Hallucination
GPUs are particularly sought after at the moment.
In September 2023 the chairman of Taiwanese semiconductor maker said that their short supply was caused by constraints of its chip-on-wafer-on-substrate (CoWoS) packaging capacity and that this shortfall was expected to last through 2024 due to slow expansion of CoWoS capacity at TSMC.
"It is not the shortage of AI chips, it is the shortage of our CoWoS capacity," said Mark Liu, the chairman of TSMC, in a conversation with Nikkei at Semicon Taiwan. "Currently, we cannot fulfill 100% of our customers' needs, but we try to support about 80%. We think this is a temporary phenomenon. After our expansion of [advanced chip packaging capacity], it should be alleviated in one and a half years."