Oracle makes Nvidia Vera Rubin systems available on its cloud
Oracle has made Vera Rubin NVL72 systems and Vera CPUs available on its cloud, with Google and Cerebras hardware planned for 2027.
Loading… Բեռնվում է…
Oracle has made Vera Rubin NVL72 systems and Vera CPUs available on its cloud, with Google and Cerebras hardware planned for 2027.
Oracle announced general availability of Nvidia Vera Rubin systems and Vera processors on Oracle Cloud Infrastructure (OCI) on 27 October, The Register said. Google and Cerebras hardware remain planned additions for 2027, rather than available alternatives.
Oracle and Nvidia also said on 27 October that Oracle would expand to “multi-gigawatt scale” using Vera Rubin servers for superintelligence (SI), Reuters said in an article carried by CNA. The expansion follows Oracle’s September earnings call, when the company said customers were increasingly paying for chips and maintained its capital-spending target of about $90 billion to $95 billion for fiscal 2027.
| Hardware | OCI status |
|---|---|
| Vera Rubin NVL72 systems | Generally available. |
| Vera central processing units (CPUs) | Generally available. |
| HGX Rubin NVL8 systems | Planned; Oracle has not specified an availability date. |
Oracle says it combines Vera Rubin graphics processing units (GPUs) and Vera CPUs with its Acceleron networking and SI-optimised data centres to keep GPUs productive and help customers expand efficiently and securely.
The W27 keeps the Galaxy Z Fold 8 Ultra’s core hardware, with dark-red or dark-blue finishes, golden accents and 16GB of working memory in both storage versions.
2 min
The Next Platform said on 27 October that Google’s tensor processing units (TPUs) and Cerebras hardware would arrive on OCI in 2027. Oracle says the Google agreement will make OCI the first large-scale cloud platform outside Google’s own cloud to offer TPUs, with capacity increasing over several years.
| Supplier | Hardware and intended work | OCI availability |
|---|---|---|
| Oracle’s first TPU offering. Google’s purpose-built machine-learning chips are designed for training large language models, further training after that initial stage, and inference for reasoning and software agents. | Planned from 2027. | |
| Cerebras Systems | CS-4, a rack-scale SI computing system based on three wafer-sized chips. The semiconductor and supercomputer developer’s offering targets inference. | Planned for 2027. |
| SambaNova Systems | Fifth-generation SN50 inference processors from the Intel-backed company. Its Reconfigurable Dataflow Unit design aims to improve data movement and computation. | Planned; no OCI start date announced. |
Cerebras introduced CS-4 in August. The company says the hardware can deliver up to 30 times the speed of GPUs, measured in tokens—small units of text—generated per second for each user. That is Cerebras’s performance claim, not an independent test result.
SambaNova released SN50 in February. The processors are designed to work alongside GPUs: GPUs handle “prefill,” or processing the input, while SambaNova’s processors handle “decode,” or generating the response.
Training is the stage in which a model learns from data. Inference uses the trained model to answer requests, making it a different computing task from the initial learning process.
Mahesh Thiagarajan, president of Oracle Cloud Infrastructure, said different SI tasks need different computing resources. Offering a choice would let customers select infrastructure for each task’s performance, throughput—the amount of work processed—and cost.
Cerebras chief executive Andrew Feldman described the agreement as a major expansion of its top-tier customer base. He said Cerebras systems were being deployed at large scale in other companies’ data centres.
Oracle said even its Nvidia-based systems would replace some Nvidia networking chips with its in-house Acceleron technology. Oracle describes Acceleron as a suite of networking and infrastructure technologies, not just chips. It combines dedicated networks, more direct paths for data and a network interface card, which connects a server to a network. Oracle says its software brings together networking, storage and security functions.
Oracle and Nvidia did not disclose the agreement’s value, and Oracle did not explain how it would finance the purchases. Nvidia chief executive Jensen Huang previously said Vera Rubin systems could generate up to $40 billion in Nvidia revenue per gigawatt. That was an estimate of potential revenue, not a valuation of Oracle’s agreement.
Oracle has not given a schedule for the multi-gigawatt expansion. The company plans to hold an investor day during the week of 27 October.
The satellite operator expects access to more than twice OneWeb’s current bandwidth in return for spending in 2027–2034, with initial IRIS² services expected from 2030.
3 min
The black, gold and green edition follows September’s sold-out preorders. Nintendo’s replenished launch-day stock also quickly sold out.
1 min