Oracle is expanding its artificial intelligence infrastructure by deepening its partnership with Nvidia while introducing chips from Google and several other suppliers to its cloud platform. The move is designed to give businesses more flexibility in selecting computing hardware for AI model training, inference and increasingly complex AI-agent applications.
Announced on October 27, 2026, the expansion reflects a broader shift in the cloud computing industry as providers seek alternatives and complementary technologies to meet growing demand for AI processing power.
Oracle Scales Up Nvidia Infrastructure
Oracle plans to expand its use of Nvidia's latest Vera Rubin AI server systems to a multi-gigawatt scale through its Oracle Cloud Infrastructure platform. The systems are designed to support demanding AI workloads, including training advanced models and running them in production.
AI model training requires substantial computing capacity to process large datasets and adjust model parameters. Inference, meanwhile, involves using trained models to generate responses, analyse information or complete tasks. Both activities are becoming increasingly important as businesses deploy AI applications at scale.
Oracle's expanded collaboration with Nvidia is intended to provide the infrastructure needed for these workloads. The companies have not disclosed the financial value of the expanded agreement.
The platform will also support computing requirements associated with AI agents, which can execute tools, test code and retrieve information between model calls. These tasks can place different demands on processors compared with traditional AI model training.
Google TPUs to Join Oracle Cloud in 2027
One of the most significant developments is Oracle's agreement to offer Google's Tensor Processing Units (TPUs) through its cloud platform beginning in 2027.
TPUs are specialised processors developed by Google to accelerate machine-learning workloads. They are designed to support AI training and inference, offering an alternative to general-purpose graphics processing units used widely across the industry.
The agreement will make Oracle the first major cloud provider outside Google's own cloud platform to offer Google TPUs to customers. Deployment is expected to increase over several years rather than occur all at once.
The arrangement could be particularly useful for organisations that want to run AI workloads close to their existing enterprise data without moving large volumes of information between cloud providers.
By offering multiple types of accelerators within one cloud environment, Oracle aims to help customers select computing resources according to performance requirements, latency and operating costs.
More Suppliers Expand the Choice of AI Chips
Oracle is also broadening its infrastructure options with specialised processors from other companies.
These include inference-focused chips from Cerebras Systems and SambaNova, alongside Positron AI's inference hardware. Oracle has also announced plans to deploy OpenAI's Jalapeño inference accelerators, developed in partnership with Broadcom.
The suppliers target different aspects of AI computing. Some processors are designed to accelerate large-scale inference, while others focus on specialised architectures intended to improve performance for particular workloads.
The expanded portfolio reflects the growing diversity of AI hardware. Although Nvidia remains a major provider of AI accelerators, cloud companies are increasingly exploring additional architectures to improve efficiency and give customers more choices.
Competition Shifts Towards Cost and Efficiency
As demand for AI services rises, companies are paying greater attention to the cost of running models, energy consumption and the ability to scale infrastructure efficiently.
Different AI workloads may perform better on different processors, meaning that a single hardware architecture may not be optimal for every task. Offering several chip options could help cloud customers allocate resources more effectively.
However, alternative chips must demonstrate reliable performance, competitive costs and compatibility with existing software before they can become widely adopted. Integrating and supporting multiple architectures also adds complexity for cloud providers.
Oracle's strategy combines a larger Nvidia deployment with a broader selection of specialised AI processors. The long-term success of the approach will depend on customer adoption, infrastructure availability and the practical benefits delivered by each platform.
The expansion highlights how competition in AI infrastructure is increasingly about more than raw processing power. Access to diverse chips, efficient deployment and predictable operating costs are becoming important considerations as businesses move AI applications from experimentation into everyday operations.
