On October 27, 2026, Oracle said it is expanding its cloud AI chip deals with Nvidia to multi-gigawatt scale using the newest Vera Rubin systems, while also adding Google’s tensor processing units (TPUs) for the first time and planning to offer chips from Cerebras Systems and SambaNova Systems. The moves give enterprise customers more silicon options for training and inference workloads on Oracle Cloud Infrastructure (OCI). Technology teams evaluating cloud AI capacity should review contract terms, delivery timelines and networking choices this week, as the company is also substituting some of its own Acceleron networking chips in Nvidia-based systems.
Oracle is diversifying its AI hardware lineup at the same time it scales Nvidia capacity, a shift that gives buyers more choice but also introduces new integration and procurement variables for technology teams.
- What changed? Oracle expanded its Nvidia partnership to multi-gigawatt Vera Rubin deployments and announced it will offer Google TPUs, Cerebras and SambaNova chips on OCI.
- When? Announced October 27, 2026 during Oracle AI World; Google TPUs and some inference chips targeted for 2027 availability.
- Who is affected? Enterprises running or planning AI training and inference workloads on Oracle Cloud Infrastructure.
- What to do this week? Inventory current OCI AI contracts, ask Oracle about roadmap timing for non-Nvidia options, and assess networking differences if using Acceleron chips.
What did Oracle announce on October 27?
Oracle and Nvidia said the cloud provider will expand to multi-gigawatt scale with Nvidia’s newest Vera Rubin AI servers. Oracle also stated it will begin offering Google’s TPU chips for the first time and plans to make Cerebras and SambaNova systems available. Even on Nvidia-based systems, Oracle said it will use some of its own in-house Acceleron networking chips instead of certain Nvidia networking components. The companies did not disclose the financial value of the expanded Nvidia deal.
How does the expanded lineup work for customers?
Customers can access different accelerators depending on workload type. Nvidia systems remain the core for large-scale training, while Google TPUs, Cerebras wafer-scale systems and SambaNova chips are positioned more for inference (generating responses after a model is trained). Oracle noted that customers can scale CPU and GPU resources independently in some configurations, and that the inference market is more competitive than training. Availability for Google TPUs and Cerebras is expected in 2027 according to company statements.
Why is Oracle doing this now?
Demand for AI compute continues to outstrip supply of any single vendor’s chips. By adding Google, Cerebras and SambaNova options while still scaling Nvidia, Oracle aims to reduce customer lock-in risk and attract workloads that prefer alternative architectures. The company has previously highlighted capital constraints across the AI infrastructure build-out; diversifying silicon sources is one way to expand capacity without relying solely on one supplier’s production schedule.
What should technology operators do this week?
Review existing OCI agreements for AI instances and note any exclusive or preferred-vendor clauses. Ask your Oracle account team for concrete delivery windows on Vera Rubin capacity and for the 2027 timeline on TPUs and Cerebras. Evaluate whether your inference workloads would benefit from lower-latency or different-cost options versus standard Nvidia GPUs. If you use networking-sensitive clusters, confirm whether Acceleron substitution changes latency or management tools.
What should teams watch next?
Watch for formal availability dates and pricing for Google TPUs and Cerebras on OCI. Monitor whether other hyperscalers follow with similar multi-vendor chip offerings. Track Nvidia’s own production ramp for Vera Rubin systems, because Oracle’s multi-gigawatt plans still depend on that supply. Any further updates on Oracle’s capital expenditure guidance for fiscal 2027 will also signal how aggressively the company intends to fund the expansion.
FAQ
Did Oracle disclose the dollar value of the Nvidia expansion?
No. The companies did not provide a financial figure. Nvidia’s CEO has previously said Vera Rubin systems can generate substantial revenue per gigawatt for the chipmaker, but that is not a contract value for Oracle.
When will Google TPUs be available on OCI?
Oracle indicated it will begin offering them, with broader deployment expected in 2027.
Is this a move away from Nvidia?
No. Oracle is scaling Nvidia capacity at the same time it adds alternatives. Executives have described Nvidia chips as still the most fungible option for many workloads.
What is the difference between training and inference chips in this context?
Training builds or updates large models and is still dominated by Nvidia. Inference runs the trained model to answer queries and has more competing hardware options, which is why Cerebras and SambaNova are being positioned there.
Does the Acceleron networking change affect existing customers?
It applies to certain new Nvidia-based systems. Teams should confirm with Oracle whether their specific configurations will use Acceleron or Nvidia networking and what that means for tools and performance.
Last Updated: October 28, 2026.
Related: Technology hub · Technology News
Discover more from Kurums | Business Intelligence
Subscribe to get the latest posts sent to your email.