Oracle founder and Chairman Larry Elllison has revealed the cloud computing giant is spending billions of dollars on GPU and computer chips from three main tech companies without any mention of Intel, the emblematic x86 firm.
According to a report by Reuters, Ellison said, “This year, Oracle will buy GPUs and CPUs from three companies, We will buy GPUs from Nvidia, and we’re buying billions of dollars of those. We will spend three times that on CPUs from Ampere and AMD. We still spend more money on conventional compute.”
If Oracle purchased $2 billion worth of GPU from Nvidia, it would put AMD and Ampere’s spending at $6 billion to reach a total of $8 billion. That is close to the $8.7 billion the company reported for its latest 12-months capital expenditure (ending May 2023), almost double what it spend over the same period (ending May 2022) and more than four times the CAPEX for the FY 2021.
Oracle wants to position itself as a major AI cloud computing player as it looks to grapple market share from the big three (AWS, Microsoft and Google). Its 2023 CAPEX represented 17% of its annual revenue, a sharp jump from 10.6% the year before and more than 3X its CAPEX from 2021.
The Ampere connection
The announcement was made at an Ampere Computing event; Oracle invested more than $800 million in Ampere Computing, a startup that designs Arm-based chip for the server market and was founded by Renée James, a former Intel President. Oracle also announced that its database products will run on Ampere Computing, a significant blow to the hegemony of x86 architecture.
Intel was not namechecked during the event and is set to be the big loser as AMD capitalizes on its archrival inability to launch many-core processors to swoop in and secure another huge win. Like Ampere Computing, AMD has focussed on producing processors with a large number of physical cores; it launched its EPYC 9754 “Bergamo” CPU, its third flagship server processor that offers 128 cores (and 256 threads), earlier in June 2023 alongside a 128-core/128-thread processor, the 9754S which is clearly aimed at the growing threats that Ampere Computing and homegrown Arm-based processors (e.g. the Graviton from AWS) are becoming.
It is likely that Oracle has purchased significant amounts of AMD’s new Instinct MI300 accelerator which combines CPU, GPU and a lot of high bandwidth memory.
What it means for businesses
The multi-billion dollar deal with Nvidia will help Oracle secure tens of thousands of specialised so-called Tensor Core GPUs (like the H100 “Hopper” or the A100), each costing around $40,000 a pop (but likely less in bulk). This will allow the US-based cloud computing giant to add generative AI to its business offerings, similar to what Microsoft has been quietly doing with Azure (and Google on Google Cloud).
Oracle also partnered with AI platform specialist, Cohere, to help it deliver generative AI using up to 16,000 Nvidia H100 GPUs per cluster, something that the company says will “enable the acceleration of large language models (LLM) training while simultaneously reducing the cost”.
One more major player means more competition, which in turn, should help lower prices in the long run and foster innovation across the entire spectrum of organizations, from startups to Fortune 100 companies.