Oracle Expands OCI Accelerator Options for Enterprise AI

Oracle’s position in the expanding AI market reflects two competing trends. The company is benefiting from strong demand for AI infrastructure while also facing questions about the enormous investment required to support that demand and the possibility of an AI-market correction.

Oracle’s sales increased from $57.4 billion in fiscal 2025 to $67.4 billion in fiscal 2026. Oracle Cloud Infrastructure (OCI) revenue reached $7.4 billion in the first quarter of fiscal 2027, a 121 percent year-over-year increase. Remaining performance obligations, which represent contracted services not yet delivered or recognized as revenue, reached $664 billion, an increase of $209 billion from the previous year.

That growth requires substantial spending. Oracle’s capital expenditures reached $55.7 billion in fiscal 2026 and could exceed $90 billion in fiscal 2027. The company is investing heavily in datacenters to meet demand for AI services, even as public resistance to datacenter construction and AI grows. Oracle is not alone in this spending, but its investment levels have made it a frequent example in discussions about the risks of the current AI expansion.

Expanding the Technology Portfolio

Beyond the financial figures, Oracle is expanding the AI infrastructure and services available through OCI. The company is competing with major cloud providers including Microsoft, Google, and Amazon Web Services.

At Oracle AI World 2026 in Las Vegas, the company highlighted enterprise concerns around data security, workplace automation, and the underlying architecture needed to support AI workloads. The second day focused more directly on OCI infrastructure, including additional hardware options for inference and other AI tasks.

The expansion reflects Oracle’s stated approach of allowing enterprise and high-performance computing customers to use the software and hardware suited to their workloads. Nathan Thomas, senior vice president and product manager for OCI, said Oracle intends to work with a broad range of accelerator providers rather than limit customers to a single hardware option. The company already works extensively with Nvidia and AMD and plans to continue supporting multiple accelerator platforms.

Google TPUs on OCI

Oracle announced that Google’s Tensor Processing Units (TPUs) will become available through OCI in 2027. This will be the first time Google’s specialized AI chips have been offered in a hyperscale cloud environment outside Google’s own infrastructure.

The chips are designed for large language model training and post-training, as well as inference for reasoning and agentic AI workloads. Google released its eighth-generation TPUs in April. The generation is divided into two specialized products: the TPU 8t for training and the TPU 8i for inference.

Cerebras Accelerators

OCI also will deploy Cerebras’ inference hardware in 2027. Cerebras develops wafer-scale processors that place compute and memory on the same chip. In August, the company introduced the CS-4 AI accelerator, a rack-scale system built around three Wafer Scale Engine 3 Turbo chips.

Cerebras says the CS-4 can deliver up to 30 times the tokens-per-second-per-user speed of GPUs. The system is intended to provide another option for customers running high-performance inference workloads in OCI.

SambaNova RDUs

Another planned OCI option is SambaNova’s fifth-generation SN50 Reconfigurable Dataflow Unit (RDU), which was released in February. The inference-focused accelerators are designed to work alongside GPUs. GPUs handle prefill operations, while the RDUs handle decoding.

Deep Trust AI Architecture

Oracle also introduced its Deep Trust AI Architecture. The architecture will be incorporated into Oracle AI Database, Autonomous AI Database, and APEX AI Application Generator.

The system is intended to protect organizational data as AI applications access sensitive information, AI agents interact with external systems, and attackers use AI in their operations. It is designed to enforce business rules, improve the accuracy of AI-generated answers, and control interactions between AI systems and enterprise environments.

Oracle’s approach is to place this control layer where the data resides. The design is aligned with Oracle’s Maximum Availability Architecture, which provides guidance for building systems that remain available during failures.

Oracle database executives Juan Loaiza and Maria Colgan described Deep Trust AI Architecture as an effort to establish trust at the architectural level instead of adding it separately to each application. Because the controls are enforced in the underlying infrastructure, they can apply to both existing applications and newly developed systems.

More AI Features for Enterprise Work

Oracle is also expanding its partnership with OpenAI. The work connects OpenAI models, ChatGPT, and the Codex developer tool with Oracle applications, data, and cloud infrastructure. The integration is intended to support tasks such as application development and the detection and remediation of security vulnerabilities.

Oracle AI Workspace includes ChatGPT plug-ins that allow enterprises to connect OpenAI tools with Oracle Fusion applications, NetSuite, AI Database, and OCI. These integrations make Oracle capabilities available to ChatGPT users and to agent-based systems.

New AI features in Oracle’s Fusion portfolio are designed to let users describe desired outcomes while software manages the underlying processes. Mike Sicilia, Oracle’s co-CEO, said AI can reduce the need for users to navigate multiple applications, locate information, follow process steps, and determine which person should act next. Instead, users can describe what they want to accomplish, with the system selecting how to complete the work based on its understanding of the business.

Oracle also introduced applications powered by Fusion Claw, the execution runtime behind the Fusion Agentic Applications suite. The company announced 25 new agentic applications across its Fusion software suites, including human resources, enterprise resource planning, supply chain, and customer management. Example tasks include scheduling deliveries, evaluating customer service, and managing payroll. Customers can also build their own agentic applications.

Fusion Companion allows users to interact with Fusion applications through ChatGPT and other AI tools. It interprets user requests and automatically selects and coordinates the agents, workflows, approvals, and other steps needed to complete the task.

Oracle also announced Fusion Setup and Configuration Registry, a public Git repository intended to simplify and accelerate the implementation of Fusion applications. It packages Oracle expertise and best practices as reusable AI skills that agents can apply to automate setup and configuration work that is currently performed manually.