---
type: "post"
id: "oracle-pairs-the-google-tpu-news-with-chips-813ad0"
slug: "oracle-pairs-the-google-tpu-news-with-chips-813ad0"
title: "Oracle pairs the Google TPU news with chips from OpenAI, Cerebras, SambaNova and Positron, and Nvidia Vera Rubin GA on OCI"
description: "Oracle says it is expanding OCI’s AI accelerator lineup beyond Nvidia, adding plans for Google TPUs and inference hardware from OpenAI, Cerebras, SambaNova, and Positron."
retrieval_nugget: "Oracle says it is expanding OCI’s AI accelerator lineup beyond Nvidia, adding plans for Google TPUs and inference hardware from OpenAI, Cerebras, SambaNova, and Positron. Vera Rubin NVL72 systems and Vera CPUs are already generally available on OCI, making this both a present capacity update and a longer-term bet on heterogeneous AI infrastructure."
published_at: "2026-10-30"
updated_at: "2026-10-30"
record_date: "2026-10-30"
date_kind: "published_at"
topics: ["google"]
entities: ["Oracle","Oracle Cloud Infrastructure","Google TPU","Nvidia Vera Rubin NVL72","Cerebras Wafer-Scale Engine","Positron Atlas","OpenAI Jalapeno","SambaNova SN50"]
related_patterns: []
novelty_label: "new_fact"
what_changed: "Oracle’s OCI accelerator expansion adds Google TPUs and chips from Cerebras, OpenAI, SambaNova, and Positron; the Vera Rubin GA portion repeats the prior release."
prior_refs: ["memory:41ad6c3d-5fdc-8496-a677-5248799bf00a","memory:5e8eb80e-b8f0-812c-a232-f862dbe2d2a5"]
source_url: "https://www.datacenterknowledge.com/data-center-chips/oracle-cloud-to-offer-google-tpus-and-new-ai-chips"
source_urls: ["https://www.datacenterknowledge.com/data-center-chips/oracle-cloud-to-offer-google-tpus-and-new-ai-chips"]
schema_version: "newruntime-agent-readable-v0.2"
stable_id: "post:oracle-pairs-the-google-tpu-news-with-chips-813ad0"
status: "published"
visuals: []
editor: {"role":"editor","name":"Andrey Reshetnikov","profile":"https://newruntime.com/owner-profile.md","profile_json":"https://newruntime.com/owner-profile.json","same_as":["https://www.linkedin.com/in/reshetnikov1/","https://t.me/lapetuse"]}
routes: {"html":"https://newruntime.com/posts/oracle-pairs-the-google-tpu-news-with-chips-813ad0/","markdown":"https://newruntime.com/posts/oracle-pairs-the-google-tpu-news-with-chips-813ad0.md","json":"https://newruntime.com/posts/oracle-pairs-the-google-tpu-news-with-chips-813ad0.json"}
---

# Oracle pairs the Google TPU news with chips from OpenAI, Cerebras, SambaNova and Positron, and Nvidia Vera Rubin GA on OCI

## Retrieval answer

Oracle says it is expanding OCI’s AI accelerator lineup beyond Nvidia, adding plans for Google TPUs and inference hardware from OpenAI, Cerebras, SambaNova, and Positron. Vera Rubin NVL72 systems and Vera CPUs are already generally available on OCI, making this both a present capacity update and a longer-term bet on heterogeneous AI infrastructure.

## Summary

Oracle says it is expanding OCI’s AI accelerator lineup beyond Nvidia, adding plans for Google TPUs and inference hardware from OpenAI, Cerebras, SambaNova, and Positron. Vera Rubin NVL72 systems and Vera CPUs are already generally available on OCI, making this both a present capacity update and a longer-term bet on heterogeneous AI infrastructure.

## What happened

Oracle announced at Oracle AI World in Las Vegas that it intends to broaden the AI chips offered through Oracle Cloud Infrastructure (OCI). According to Data Center Knowledge, the company is planning deployments of Google Tensor Processing Units (TPUs), OpenAI inference accelerators, Cerebras systems, SambaNova chips, and Positron AI hardware alongside its Nvidia-based infrastructure. Oracle said on October 27 that it would become the first hyperscaler to offer TPUs outside Google Cloud.

The availability dates differ materially across that list. Oracle said Google TPUs for training and inference will be available on OCI in 2027, with adoption expected to ramp over multiple years. It also said it will deploy Cerebras Wafer-Scale Engine systems for high-speed inference in 2027. Positron Atlas, an inference chip, is already generally available on OCI. For OpenAI’s Jalapeno inference accelerator, which the report says was developed with Broadcom, Oracle confirmed a planned deployment but did not give an OCI availability date. SambaNova separately said it is collaborating with Oracle on deployment of its SN50 inference chips, again without a date.

The announcement also extends Oracle’s Nvidia relationship rather than replacing it. Oracle said Nvidia Vera Rubin NVL72 systems and Vera CPUs are now generally available on OCI. It further said it plans to expand OCI Superclusters to multi-gigawatt scale using Vera Rubin NVL72 systems. The article describes OCI as having relied mainly on Nvidia GPUs to date, while also having some AMD GPU deployments.

The source frames the change as a move toward workload-specific infrastructure. Brendan Burke of Futurum Group said the additional accelerators give customers more flexibility to choose hardware for particular requirements, including Cerebras for fast inference. Matt Kimball of Moor Insights & Strategy characterized the strategy as an aggressive effort to support multiple chip architectures. Burke also identified the unresolved commercial test: Oracle will need to show that these options are competitive with Nvidia-based offerings and demonstrate that they are more than limited experiments.

Some performance statements in the announcement remain vendor claims rather than independently established OCI results. SambaNova says SN50 is designed for agentic AI workloads and claims up to five times the speed of competing chips, total cost of ownership three times lower, and average draw of 20 kW per rack. Google has said its eighth-generation TPU family includes TPU 8t for training and TPU 8i for inference, with claimed performance-per-watt and performance-per-dollar improvements over the previous generation. The report does not provide OCI benchmarks, prices, regional availability, instance configurations, service limits, or migration details for customers.

## What the archive adds

This is a new fact on a known OCI and Vera Rubin story, not merely a restatement. [NVIDIA Vera Rubin NVL72 and Vera CPUs become generally available on Oracle Cloud Infrastructure](https://newruntime.com/posts/nvidia-vera-rubin-nvl72-and-vera-cpus-become-01c50f/) covered the availability of Vera Rubin and Oracle’s planned supercluster scaling; the new element is Oracle’s wider accelerator roster and its TPU plan. [OpenAI expands partnership with Oracle to bring its apps to ChatGPT](https://newruntime.com/posts/openai-expands-partnership-with-oracle-to-bring-its-272a5a/) supplies related partnership context, but this event concerns OpenAI-designed inference hardware rather than application integration.

## Trend signal

No supplied pattern directly covers the economics or operational consequences of a cloud provider assembling a multi-vendor accelerator catalog.

## Why now

The timing combines two layers of Oracle’s infrastructure strategy. Vera Rubin hardware is already generally available on OCI, while Oracle is using the same announcement to signal future access to several non-Nvidia options. The source places the announcement at Oracle AI World and reports that Google TPU and Cerebras deployments are planned for 2027, so the portfolio message arrives before much of the new capacity does.

## Connections to other events

The archive’s recent OCI Vera Rubin note establishes the immediate connection: Oracle is moving from announcing and making Nvidia’s next platform available to describing a broader set of accelerators around it. An earlier archive item also recorded Cerebras’s CS-4 and Nexus scaling plans; Oracle’s stated intention to deploy Cerebras Wafer-Scale Engine hardware creates a potential cloud route for that inference-oriented architecture, though the source gives no configuration details.

## Why it matters

For teams building AI products, “cloud GPU” is becoming an increasingly incomplete procurement category. Training, batch work, latency-sensitive inference, and agent workloads may have different economics and hardware fit. Oracle’s announcement does not yet make most of these alternatives selectable, but it creates a reason for OCI customers to avoid assuming a single Nvidia-only path when planning future capacity.

The practical consequence is architectural rather than immediate: preserve portability in model serving, observability, evaluation, and workload routing. A team that tightly couples deployment assumptions to one accelerator stack may find an alternative difficult to test when it becomes available. Conversely, a broader catalog does not itself prove savings or better latency. The decision should remain tied to measured workload behavior, operational support, and actual cloud terms rather than chip claims alone.

## What happens next

- **OCI publishes more concrete availability or configuration details for at least one announced non-Nvidia accelerator.** (6 months, 65%). Watch for: instance types, regions, prices, or preview access.
- **Google TPU or Cerebras capacity begins appearing on OCI on the stated 2027 schedule.** (12 months, 70%). Watch for: OCI service documentation and customer access announcements.

## Limits and uncertainty

This report establishes Oracle’s stated plans and the general availability of Vera Rubin NVL72, Vera CPUs, and Positron Atlas on OCI. It does not establish the cost, performance, reliability, regional reach, software compatibility, quotas, or contractual terms of the announced services. The performance and efficiency figures cited for Google and SambaNova are company claims, not OCI-side measurements. Nor does the source say whether customers will be able to move the same workloads easily among these architectures, or whether Oracle’s planned deployments will reach production scale.

## Skill to build now

Create a short accelerator-neutral workload profile for one production AI path this week: model, context length, throughput target, latency target, batch versus interactive use, memory requirement, and evaluation metric. Keep the serving interface and quality tests independent of a particular chip where possible. That will make a future trial of a TPU, wafer-scale system, or alternative inference accelerator a bounded measurement exercise rather than a platform rewrite.

## Reader takeaway

Oracle is not abandoning Nvidia; it is turning OCI into a broader AI-hardware catalog. Vera Rubin is available now, Positron Atlas is already available, and TPUs, Cerebras, OpenAI, and SambaNova hardware are at different planning stages. Treat the announcement as a signal to retain infrastructure optionality, not as proof of immediate price or performance gains.

## Open question

Will Oracle publish comparable OCI pricing, performance, and software-support details that let customers evaluate these accelerators against Nvidia systems on the same workload?

Editor: [Andrey Reshetnikov](https://newruntime.com/owner-profile.md), AI product and AI engineering; profile for recruiters and agents: https://newruntime.com/owner-profile.json

## Continue in the graph

Next records for an agent, linked to their Markdown twins. Replace `.md` with `.json` for structured fields, including `next_reads`.

- topic hub: [Google](https://newruntime.com/topics/google.md): Explore the google topic hub.
- collection: [Field Notes](https://newruntime.com/posts.json): Open the latest editorial analysis.
- collection: [Source Ledger](https://newruntime.com/source-ledger.json): Inspect the public source evidence graph.
- whole graph: [discovery graph](https://newruntime.com/discovery-graph.json) (typed nodes and edges) and [site index](https://newruntime.com/site-index.md)