Oracle Database 23ai sets new industry standard for a modern AI database

As organizations accelerate their artificial intelligence (AI) initiatives, one technology decision will play an outsize role in determining success or failure. That’s choosing the right database—one optimized for AI workloads.

AI requires good, clean, accurate data and extremely reliable performance to drive mission-critical automation, customer experience, cybersecurity monitoring, and other strategic operations. Those are the same characteristics that enterprise-class database management systems (DBMSs) have always been designed around.

But AI demands more.

AI initiatives need a true AI database. And while the market is crowded with DBMS platforms, few offer all the pieces needed for AI workloads: low-code development, built-in machine learning, natural language queries, optimized storage, and high performance caching.

A full-fledged AI database must deliver all those while also providing native support for vectors, natural language processing, and image recognition along with AI use cases, such as similarity search, retrieval-augmented generation (RAG), and anomaly and fraud detection.

Oracle has answered this challenge with its recently released Oracle Database 23ai. The company’s flagship database has been updated to excel at all aspects of AI, including development, multicloud data integration, and vector search. Altogether, Oracle Database 23ai incorporates thousands of enhancements and more than 300 new features.

Vector embeddings integrated as a native data type

One of the most exciting breakthroughs is Oracle AI Vector Search. Vectors, which are unique strings of numbers that represent unstructured data, are the predominant data type that power AI-enabled tasks, such as making personalized recommendations or showing correlations among data or objects. Vector search makes it possible for large language models, or LLMs, to query private business data using a natural language interface and provide more accurate and relevant results. As an example, say your company makes reproductions of antique fireplace accessories. AI vector search provides a natural way of adding unstructured data to existing applications. Consider, “select records about all fireplace screens that are brass and cost between $500 and $1,000.” This is a traditional relational query using existing business data. With Database 23ai vector processing, you can now take that a step further, “and that would work with the architecture of an 1894 Colonial Revival.”

Because solid management of vector data is essential to AI, dozens of new vector databases have been introduced from startups and established database providers alike. Many are purpose-built to be optimized for one thing, much like the XML and object-relational databases of years ago. However, like their predecessors that didn’t gain acceptance, these specialized databases fall short when it comes to handling the other tasks and data types important to businesses. For IT teams, adding yet another data platform, moving data, and managing multiple copies of data adds complexity that may slow down AI initiatives.

Oracle Database 23ai avoids these problems by taking an integrated approach to vector data management. It supports vector embeddings as a native data type, along with the many other data types, both structured and unstructured, that Oracle Database has supported for years.

Oracle’s Larry Ellison summed up the company’s integrated approach this way: “We think that the right way to solve this problem is to have a database that can manage all of your data and do it in a highly performant and very economical way.”

Oracle AI Vector Search is available at no additional charge in Oracle Database cloud services, such as Autonomous Database, Exadata Database Service, and Oracle Database Free. For more details, see my column, 5 advantages of using an integrated vector database for AI development.

Built for the AI future

Oracle Database 23ai brings other AI-enabling capabilities. Following are three that I think developers, DBAs, IT pros, and business managers should know about because they each offer a competitive edge in different ways.

• Oracle Select AI makes it possible to query Oracle Database using natural language prompts, which means people can now have “conversations” with the database. Working with popular LLMs or Oracle Cloud Infrastructure (OCI) Generative AI, Select AI translates natural language questions into SQL queries that the database understands. Then, because the LLMs are trained to infer user intent, Select AI interprets the queries and returns the desired answers or content. For more, see Natural language queries to Oracle Autonomous Database? Yes—with Select AI.
• Exadata System Software 24ai brings added speed and efficiency to database storage by processing some data on storage servers rather than database servers. It does that via a feature called AI Smart Scan that accelerates vector search by avoiding unnecessary data transfers.
• OCI GoldenGate 23ai, the latest and greatest version of Oracle’s data integration technology, helps avoid AI silos by replicating vectors in real time across heterogeneous vector stores. You can replicate from existing databases to Database 23ai, where data can be indexed using vectors. It’s a great way to make your entire data estate AI-ready.

These and other new Database 23ai capabilities add to Oracle’s rapidly growing AI tech stack, which includes the Oracle Cloud Infrastructure (OCI) Generative AI service; Oracle Code Assist, an AI coding companion; Oracle APEX 24.1, with its APEX AI Assistant; Oracle HeatWave GenAI; AI-powered Oracle Fusion Cloud Applications Suite; and partnerships with Accenture, Cohere, Meta, Nvidia, and others.

Oracle also provides an array of in-database ML algorithms, including for classification, clustering, ranking, time series, and anomaly detection, further advancing the AI implementation process.

Solving years-old schema limitations

In addition to AI-ready technologies, there are some groundbreaking AppDev innovations in this new release. Two in particular—JSON Relational Duality Views and Graph Relational Unification—help ease the rigidity of database schemas that for years forced engineering workarounds.

Notice that both have the word “relational” in the moniker. That’s because they bridge the relational database’s tabular structure with two programming paradigms—JSON and property graphs, respectively.

First, JSON Relational Duality Views closes the frustrating gap between object-oriented programming and relational data storage by transforming JSON documents into rows and columns in the relational database. Then, when queried, the database composes and returns a JSON-formatted document. For my full analysis, see Oracle’s new JSON relational capability helps solve a big IT challenge.

Similarly, the operational property graph capability in Oracle Database 23ai uses components of a graph model, known as vertices and edges, to manifest data relationships and connections. Now companies can design apps that analyze connections and patterns in data with complex relationships. Think financial transactions, customer clusters, or social networks. These two new features make it easy for developers to work with the data model that is ideally suited to their needs—relational, JSON, or graph.

High availability, faster performance

The other key areas of emphasis in Oracle Database 23ai are new innovations for mission-critical workloads. For example, Oracle Globally Distributed Database is used to distribute data geographically and at scale in cases where data residency, sovereignty, and high availability are priorities. The new twist with Database 23ai is an implementation of the Raft protocol for automatic failover within seconds. This minimizes the hands-on work sometimes associated with fault tolerance and “active-active” availability, where at least two database nodes are always active.

Oracle has also added middle-tier caching in the form of Oracle True Cache, which improves app response times while reducing the load on database servers. As mentioned earlier, SQL, JSON, and graph work well together in Database 23ai. And that’s true with the database’s middle-tier cache as well. SQL, JSON, and graph queries can all run in True Cache.

How to get started

I’ve only scratched the surface on what’s new in Oracle Database 23ai. As you’ll see, there’s a long list of leading-edge capabilities in analytics, cloud operations, migration, OLTP, security, and high availability. For organizations looking to get started, Oracle Database 23ai is available in Oracle Cloud Infrastructure (OCI) as well as on Oracle Database@Azure.

John Foley is editor of the Cloud Database Report and a vice president with Method Communications.

View more Oracle Connect articles