Pegging the entry cost for creating advanced AI models running in big data centers at $100 billion, Larry Ellison revealed at CloudWorld that Oracle plans to use three nuclear reactors to power the colossal gigawatt-plus data center the company is currently designing.
While Ellison has made a fortune — one that, following Oracle’s blowout Q1 numbers, is now valued at about $192 billion by announcing seemingly outrageous plans and then making them come true, his vision for a nuclear-powered gigawatt-plus data center strains the bounds of believability for even Ellison.
But hey — don’t take my word for that — Ellison himself admitted as much on Oracle’s fiscal-Q1 earnings call earlier this week.
“Let me say something that is going to sound really bizarre,” Ellison said during the Q&A portion of the call.
Ask AI Ecosystem Copilot about this analysis
“You’re probably saying, ‘Well, he says bizarre things all the time so why is he announcing this one? It must be really bizarre!’
“So we’re in the middle of designing a data center that’s north of a gigawatt and we’ve found the location…and they’ve already got building permits for three nuclear reactors — these are the small modular nuclear reactors — to power the data center.
“This is how crazy it’s getting,” Ellison said of this new AI data-center game where the table stakes are $100 billion.
A moment earlier, Ellison had laid out his view of how expensive — almost unfathomably so — these new data centers and their related costs are going to be.
“When I talk about building gigawatt or multigigawatt data centers for these AI models, these frontier models, the entry price for a real frontier model for someone that wants to compete in that area is around $100 billion,” Ellison said.
“Let me repeat: around $100 billion. That’s over the next four or five years for anyone who wants to play in that game.
“That’s a lot of money. And it doesn’t get easier. So there are not going to be a lot of those [players].”
Indeed, Ellison had noted earlier that the AI Model Wars are going to be not only stupendously expensive but also pretty much endless because the world will not be content with a small number of general-purpose models, and instead will trigger demand for more and more-specialized data centers.
And the goal will begin to center on the creation of not just massive data centers but rather the fruits of those behemoths: neural networks.
“Well, a lot of people think ‘My God, I send a kid to college and then I’m done — the training is over. Four years of training and then I can put the kid to work and they’ll be doing inferencing,’ ” Ellison said on the call.
“And that’s not true. This race goes on forever to build a better and better neural network. And the cost of that training gets to be astronomical.”
Final Thought
Ellison’s genius is that just as the rest of us begin to think we’ve finally got a good grasp on his plans, he catapults off in some other totally unexpected direction with outrageously bold — some are tempted to say “impossible” — ideas.
And then his vision becomes reality, and he ratchets up the craziness once again.
So for Larry Ellison, a gigawatt data center wasn’t crazy enough. And even a gigawatt-plus data center wasn’t crazy enough.
So he’s gone all-in with an idea that might seem preposterous today but will perhaps become the norm not too far down the road: nuclear-powered gigawatt-plus data centers.
Keep the bizarre ideas coming, Larry!
The AI Ecosystem Q2 2024 Report compiles the innovations, funding, and products highlighted in AI Ecosystem Reports from the second quarter of 2024. Download now for perspectives on the companies, innovations, and solutions shaping the future of AI.