You won’t believe what Larry Ellison and Elon Musk said to Nvidia CEO Jensen Huang

Larry Ellison is the president Oracle (ORCL -0.75%)which is currently building some of the world’s fastest and most efficient data centers for artificial intelligence (AI) development. Elon Musk, on the other hand, runs adze (TSLA 2.25%)which is building AI-powered self-driving software for its electric vehicles. He also runs SpaceX, X (formerly Twitter) and a new AI start-up called xAI.

Ellison and Musk need tens of thousands of graphics processing units (GPUs) for their data centers to bring AI to life and Nvidia (NVDA 0.94%) provides the best chips in the industry.

At Oracle’s Sept. 12 financial analyst meeting, Ellison told the audience that he and Musk recently went to dinner with Nvidia CEO Jensen Huang at Nobu restaurant in Palo Alto. The two, who are among the richest people on Earth, found themselves begging Huang for something money just can’t buy right now. Here’s how it went.

Image source: Getty Images.

The GPU arms race

Oracle currently has 162 data centers, either active or under construction, but believes that number could eventually exceed 2,000 as demand for computing power from AI developers increases. Some of Oracle’s largest data centers have clusters of more than 32,000 GPUs, but next year the company will offer a 131,072 GPU cluster from Nvidia’s latest Blackwell line.

Oracle has designed a unique RDMA (Direct Random Memory Access) networking technology that can move data from one point to another faster than traditional Ethernet networks, and because developers pay for computing power by the minute, this can significantly reduce the costs. This is why leading AI start-ups like OpenAI, Cohere and even Musk’s xAI use Oracle infrastructure.

In its recent fiscal 2025 first quarter (ended July 31), Oracle’s Cloud Infrastructure (OCI) segment generated $2.2 billion in revenue, a staggering 45% year-over-year growth. However, it could grow even faster if not for supply constraints — in other words, Oracle simply can’t get its hands on enough GPUs for its data centers.

Not only is Oracle battling other cloud giants Microsoft, Amazonand Alphabet for GPU allocations from Nvidia, but tech companies like Tesla and Meta platforms they also consume the supply to develop AI for their own purposes. Tesla is trying to bring a cluster of 50,000 GPUs online this year to improve its self-driving software, which requires a substantial amount of computing power.

Meta, on the other hand, used around 16,000 of Nvidia’s flagship H100 GPUs to train its Llama 3.1 large language model (LLM), but the company plans to increase its capacity to a staggering 600,000 H100 equivalents by at the end of this year. That will pave the way for Llama 4, which CEO Mark Zuckerberg says could set a benchmark for the industry in 2025.

Ellison and Musk are begging for more GPUs

Please take our money…take more of it. You don’t take enough. … We need you to take more money from us. Please.

— Ellison and Musk’s comments to Jensen Huang at the dinner, according to Ellison.

Ellison and Musk were practically begging Huang for more GPUs, but no amount of money in the world can buy the numbers they need right now because Nvidia simply can’t keep up with demand. Oracle and Tesla aren’t even Nvidia’s biggest customers!

Oracle spent $6.9 billion on capital expenditures (capex) during fiscal 2024 (ended April 30) and expects to spend double that’s in fiscal year 2025. Most of the money will go toward buying chips and building data centers. Tesla plans to spend over $10 billion on capex this calendar year as a whole, which will also go toward the 50,000 GPU cluster I mentioned earlier.

These figures are modest compared to what other tech giants spend. Microsoft has earmarked $55.7 billion for capital expenditures in fiscal 2024 (ended June 30) and plans to spend even more in fiscal 2025. Amazon’s capex, on the other hand, could exceed $60 billion of dollars in calendar 2024.

It’s no surprise, then, that Nvidia generated $26.3 billion in data center revenue in its latest fiscal 2025 quarter (ended July 28), up 154 ​​percent from the year-ago period. Ellison says the wave of AI spending could continue over the next 10 years as companies and nation states battle for technological supremacy when it comes to AI, so Nvidia’s data center revenue likely has plenty of growth left in tank.