Amid an onslaught of demand from tech giants rushing to launch and scale their artificial intelligence units, a global chip crunch persists.
On a call with investors Monday (Dec. 11) to go over quarterly earnings, Oracle chairman Larry Ellison reiterated how hot things remain in the race to supply AI models with the massive amounts of computing power necessary to train and operate them.
“We got enough Nvidia GPUs for Elon Musk’s company, xAI, to bring up the first version, the first available version of their large language model called Grok,” said Ellison, whose company is one of the world’s largest software makers. “They got that up and running. But boy, did they want a lot more. Boy, did they want a lot more GPUs than we gave them.”
Nvidia and Oracle have a longstanding partnership that combines the former’s hardware with the latter’s software to help advance AI cloud computing development. As Quartz’s Michelle Cheng has reported, the high-performance, $40,000 Nvidia H100 chip has become one of the most sought-after pieces of hardware in Silicon Valley. AI models need hundreds to thousands of the chips to work properly; in June, Ellison said Oracle was spending “billions” on them.
Grok pulls source material from Musk’s X, the social media platform formerly known as Twitter, and has a tendency to include jokes and parallel-reality, alternative-facts “hallucinations” in its output. That means some experts question whether it can be taken seriously as an AI, but the need for computing is real.
“There’s this gold rush towards building the world’s greatest large language model,” Ellison told investors. “And we are doing our best to give our customers what we can this quarter and then dramatically increase our ability to give them more and more capacity each succeeding quarter.”
Artificial intelligence, real constraints
But one analyst on the call, Deutsche Bank’s Brad Zelnick, questioned whether Oracle would be able to meet the demand it’s seeing. Zelnick pointed out that it’s taken a few years for the company to build 66 cloud data centers, which makes it unclear how Oracle will stay on track to build 100 more. (Zelnick, who has a “buy” rating on Oracle stock, wrote in a research note that artificial intelligence, including the company’s own generative AI products, wasn’t the only thing straining cloud computing supply.)
Executives acknowledged that they were leaving money on the table, even if they were grabbing it hand over fist. “We had capacity this quarter in the hundreds of millions of dollars more that was just sitting there waiting to take it,” said Oracle CEO Safra Catz on the call. “Instead of taking small pieces or smaller pieces, we decided to focus on the bigger parts and tried to also treat our customers fairly and work with them to meet their needs.”
Oracle profits for the third quarter came in slightly higher than anticipated, at $2.5 billion on a somewhat lower than expected $12.9 billion in revenue. The market didn’t like what it saw, and shares had fallen about 10% in Tuesday trading.