Oracle expands Nvidia deal, adds Google and Cerebras chips

Oracle ↗ 4.61% and Nvidia said on Tuesday, October 27, that they plan to expand Oracle Cloud Infrastructure (OCI) Superclusters built on Nvidia's Vera Rubin NVL72 systems to "multi-gigawatt scale." Both Vera Rubin NVL72 and Nvidia's Vera CPUs are now generally available on OCI, according to Nvidia. Oracle also announced that it is adding chips from Nvidia rivals. Google's Alphabet Inc. ↗ 1.30% tensor processing units (TPUs) will come to its cloud, along with inference hardware from Cerebras ↗ 8.56% and SambaNova Systems, Reuters Thomson Reuters Corporation ↘ 4.59% reported.nvidianews.nvidia.com+1

The companies did not say what the Nvidia deal is worth. Nvidia CEO Jensen Huang has said Vera Rubin systems can bring the chipmaker up to $40 billion in revenue per gigawatt, Reuters noted.reuters.com

Nvidia expansion

OCI Superclusters combine Nvidia NVL72 racks with Oracle's own Acceleron networking and Nvidia Spectrum-X Ethernet. OCI also plans to offer smaller HGX Rubin NVL8 servers, which pair eight Rubin GPUs with Intel ↗ 3.53% Xeon 6 processors, Nvidia said. The Vera CPUs are meant for the work AI agents do between model calls, such as running tools and code. OpenAI, which had early access, said it is "bringing Vera into production on OCI." Reuters reported that even Oracle's Nvidia-based systems will use Oracle's in-house Acceleron chips in place of some Nvidia networking parts.nvidianews.nvidia.com+1

Google, Cerebras and SambaNova

OCI will be the first hyperscaler outside Google to offer TPUs, with deployments starting in calendar 2027, Google Cloud said. Capacity will ramp up over several years, Investing.com reported. Cerebras inference on OCI is also planned for calendar 2027, Cerebras said. Oracle also said it will add OpenAI's Jalapeño chip, according to CNBC's coverage. A Cerebras spokesperson said the deal is material but that the company would not disclose its scale because it is in a quiet period.cnbc.com+3

"That's a massive expansion of our customer base of tier-one customers," Cerebras CEO Andrew Feldman told Reuters. Nvidia leads the market for training AI models, but the inference market is more open to competitors.reuters.com

Funding questions

Oracle made the announcements at its AI World conference in Las Vegas and is set to hold an investor day later this week. It has not said how it will pay for the chips. On last month's earnings call, Oracle said customers are increasingly covering chip costs and kept its capital spending target of about $90 billion to $95 billion for fiscal 2027, Reuters reported.reuters.com

"Different AI workloads have different compute needs," said Mahesh Thiagarajan, president of Oracle Cloud Infrastructure.