- Over 600 AI agents embedded across Oracle Fusion Applications for finance, HR, supply chain, sales, marketing, and service.
- New AI Data Platform and AI Database 26ai introduce vector search, built-in agents, privacy controls, and post-quantum security.
- OCI expansions include AMD MI450 AI superclusters and Zettascale10 with up to 16 zettaFLOPS and 800,000 NVIDIA GPUs.
- Agent Marketplace, Agent Studio, Dedicated Region25, and Oracle Acceleron target regulated sectors and low-latency, zero-trust networking.
With Las Vegas as the stage, Oracle AI World 2025 brought together more than 16,000 attendees to explore how artificial intelligence is being woven into day‑to‑day business operations. The event centered on practical adoption rather than hype, highlighting a stack that spans applications, data, and infrastructure built to run sensitive workloads at scale.
Across two days, Oracle outlined a roadmap that couples new data and database offerings with large‑scale compute and application automation, culminating in over 600 AI agents integrated into Oracle Fusion Applications. Announcements also included a specialized marketplace, an expanded developer studio, and cloud advances such as OCI Dedicated Region25 and the Oracle Acceleron network architecture aimed at regulated industries and regional deployments, including Latin America.
Day 1: Infrastructure, data, and supercomputing
Oracle introduced the AI Data Platform, a unified environment that connects leading generative models to enterprise data, applications, and workflows. It features automated ingestion, vector indexing for retrieval, and an integrated Agent Hub to manage and operationalize agents across business processes.
The new AI Database 26ai builds on Oracle Database with native vector search, embedded AI agents, advanced analytics, and enhanced privacy safeguards. It emphasizes openness (tooling compatibility) and even post‑quantum security measures to future‑proof sensitive data handling.
On compute, Oracle and AMD unveiled next‑generation public AI superclusters featuring AMD Instinct MI450 Series accelerators. Configurations scale to up to 50,000 GPUs, enabling large‑batch training and fine‑tuning of sophisticated models across domains.
Complementing that, OCI Zettascale10 was presented as a high‑end AI cluster tier with up to 16 zettaFLOPS of performance backed by as many as 800,000 NVIDIA GPUs. The target: massive training jobs and multi‑tenant research workloads that benefit from extreme throughput and interconnect.
Day 2: Automation in applications and agent ecosystem
The spotlight shifted to business automation with 600+ AI agents embedded directly in Oracle Fusion Applications. These agents operate natively inside finance, HR, supply chain, sales, marketing, and service flows, acting under the same security and permissions that govern human users rather than as bolt‑on copilots.
Oracle expanded its Agent Marketplace and Agent Studio to accelerate build‑out and customization. The ecosystem supports multiple foundation models, including OpenAI, Anthropic, Cohere, Google’s Gemini, Meta, and xAI, so organizations can select models that fit cost, governance, and performance needs.
In a notable commercial stance, Oracle positions these capabilities as part of the product rather than a premium add‑on, effectively charging no add‑on fee for embedded agents. That said, the economics of large‑scale AI (compute, storage, and the cost of GPU inference) could prompt practical consumption controls or tiers as adoption ramps up.
Adoption signals included training and community metrics: more than 32,000 professionals certified on Agent Studio and a hackathon where 66 companies built 109 agents in a single day. Demonstrations showed agents that flag accounting anomalies without manual intervention and others that generate purchase orders with optional approval gates.
Use cases across industries and the human‑in‑the‑loop
Customer examples spanned multiple sectors. Milwaukee Tool described how standardizing on Fusion is preparing the ground for broader AI adoption across operational workflows and decision support.
Berkshire Hathaway Energy consolidated six subsidiaries onto a single Fusion instance, enabling shared data models for predictive maintenance and demand forecasting, and ultimately speeding executive and field‑level decisions.
MGM Resorts and Helzberg Diamonds reported cycle‑time reductions in service and back‑office processes while keeping a human‑in‑the‑loop for customer‑facing touchpoints—a balance between automation and brand experience.
Additional examples presented by Oracle’s industry leadership included hospitals cutting documentation time by a 41% reduction and validating data in healthcare, construction firms correcting billing errors via agents, and financial institutions accelerating fraud investigation workflows with auditability.
A coherent stack: OCI, data, and applications
Speakers emphasized a top‑to‑bottom integration: infrastructure led by Clay Magouyrk (OCI), data and database advances from Juan Loaiza and T. K. Anand (AI Database 26ai and AI Data Platform), and application integration under Steve Miranda. The takeaway was a cohesive stack where what the infra team builds and data team trains is directly consumed by business apps.
Such tight coupling offers efficiency but hinges on data quality and access control. If governance falters, agents can degrade in accuracy or reliability, underscoring the need for rigorous governance across data lineage, permissions, and model lifecycle management.
For deployment flexibility, OCI Dedicated Region25 aims to deliver more than 200 cloud and AI services on premises or in‑country with public‑cloud parity, while Oracle Acceleron introduces a next‑gen network fabric designed for lower latency, higher throughput, and zero‑trust security. Both target regulated industries and geographic requirements, including customers across Latin America.
In competitive context, Oracle’s approach contrasts with UI‑centric copilots from Microsoft Fabric, SAP BTP, or Salesforce Einstein. The strategy here is to embed AI deep into operational systems rather than to emphasize front‑end assistants.
Governance, measurement, and sustainability
Enterprises evaluating these capabilities will want observable metrics—accuracy, latency, intervention rates, and uptime—reported consistently. Oracle has not yet published standard benchmarks or production counts for every agent, leaving room for customers to define accuracy and latency SLOs aligned with their risk posture.
As agent portfolios expand, organizations should enforce audit trails, policy‑based approvals, and human‑oversight checkpoints. Without disciplined traceability, automation can become opaque, complicating compliance and root‑cause analysis.
Finally, the operating model must remain sustainable, balancing performance and cost through resource scheduling, caching strategies, and potential usage guardrails. Clear capacity planning will help teams right‑size spend while preserving service levels.
Oracle AI World 2025 sketched a pragmatic blueprint: 600+ embedded agents inside Fusion, a modern data foundation via AI Data Platform and AI Database 26ai, and ambitious OCI upgrades for high‑end training and regulated deployments—an integrated play that will rise or fall on data discipline, governance, and real‑world business outcomes.
