This article is the third in "The Future of Java" series.

Every wave of enterprise technology has been defined not just by the frameworks and runtimes, but by the underlying models and data. In the early days of Java, the models were relational databases, and the standards were JDBC and JPA. Later, the models were business processes, and the standards were messaging and transaction APIs. Today, the models are large language models (LLMs), and the question is not whether enterprises will adopt them, but which models they will trust and how they will integrate them into existing application landscapes.

This is where IBM’s Granite models and the watsonx platform come in. For Java developers and enterprise architects, they represent the next step in the long tradition of combining open standards with trusted enterprise platforms. Granite models provide an open, transparent foundation for AI, while watsonx delivers the enterprise governance and lifecycle management that makes those models usable in production. When combined with Quarkus, LangChain4j, Jakarta EE, and Red Hat OpenShift AI, they form a complete story: one where enterprises can adopt AI confidently, without compromising on openness, security, or long-term stability.

The model question: Open versus closed models

The debate over open versus closed models is not new. Enterprises faced similar questions with databases, operating systems, and even Java itself. Proprietary models might deliver short-term advantages, but they also carry long-term risks. Vendor lock-in, opaque decision-making, and uncertain licensing terms can leave enterprises exposed.

Granite models are IBM’s answer to this challenge. They are fully open source LLMs, released with transparent training data, evaluation metrics, and performance benchmarks. Just as Jakarta EE provided an open standard for enterprise Java, Granite models provide an open foundation for enterprise AI. They are not black boxes; they are models enterprises can inspect, test, and improve.

For architects, this matters because it aligns AI adoption with the same principles that made Java successful. Openness is not just a philosophical choice, it is a practical strategy for ensuring that investments remain valuable over time.

Granite models in practice

Granite models are designed to be flexible. They can run in cloud environments, on-premises, or in hybrid setups. They are optimized for enterprise workloads, including text generation, classification, summarization, and retrieval-augmented generation. And because they are open, they can be fine-tuned on domain-specific data without hitting licensing roadblocks.

For Java developers, LangChain4j makes integration straightforward. A Quarkus application can connect to a Granite model through a simple API, use embeddings for semantic search, or build an agentic workflow that calls multiple models in sequence. Because langchain4j-cdi aligns with Jakarta CDI, the same programming model applies whether the application runs on Quarkus, WildFly, or another Jakarta EE-compatible runtime.

This consistency is critical. Developers can adopt Granite models today using Quarkus, confident that the integration patterns will mature into standardized APIs tomorrow. In other words, Granite models are not just another tool, but they are part of a larger ecosystem that includes both frameworks and standards.

IBM watsonx: The enterprise AI environment

If Granite models are the foundation, watsonx is the environment that makes them usable in production. Enterprises cannot run models in isolation. They need lifecycle management, version tracking, governance, monitoring, and security controls. Watsonx provides all of this.

At its core, watsonx is a platform for building, governing, and deploying AI models at scale. It integrates model registries, pipelines, monitoring dashboards, and policy enforcement into a single environment. For enterprises in regulated industries, this is not optional. You need to know which model was used to make a decision, what data it was trained on, and whether it complies with industry regulations. Watsonx ensures that this information is available and auditable.

For Java developers, this governance is invisible but essential. A Quarkus application that calls a Granite model through LangChain4j does not need to worry about whether the model is properly versioned or whether its use complies with internal policies. Watsonx handles those concerns. Developers focus on application logic, while the platform ensures compliance.

Standards as the common thread

What ties Granite and watsonx back to Java is the same principle that has always defined enterprise Java: standards. Jakarta EE and MicroProfile provide the programming models that let developers focus on business logic. Quarkus provides the innovation layer that makes those models usable with AI today. Granite and watsonx extend that story into the model and platform layer.

Standards are not just about APIs; they are about stability of knowledge. Developers who learn Jakarta CDI or Jakarta Persistence know that those skills will remain relevant across runtimes. In the same way, adopting Granite models and watsonx ensures that AI skills remain portable and aligned with open standards. You are not learning one provider’s proprietary API; you are learning patterns that will continue to matter as the ecosystem evolves.

This continuity is critical in the age of AI, where hype cycles move faster than ever. Without standards, developer skills risk becoming disposable. With standards and open models, knowledge scales across languages, platforms, and evolving implementation approaches.

IBM’s role in securing the future

IBM’s investment strategy is clear: innovation and standards must move together. On the framework side, IBM and Red Hat are advancing Quarkus and LangChain4j, giving Java developers the tools they need to build AI-infused applications today. On the standards side, they continue to drive Jakarta EE and MicroProfile forward, ensuring that proven patterns are codified for the long term. On the model side, Granite provides open, transparent alternatives. On the enterprise platform side, watsonx and Red Hat OpenShift AI ensure that those models can be deployed and governed at scale.

This is not just about delivering products. It is about securing the future of enterprise applications. Enterprises can experiment with AI today without putting themselves at risk tomorrow. They can innovate with Quarkus and Granite while staying aligned with Jakarta EE and watsonx. They can move fast without abandoning the discipline that has always defined enterprise Java.

Practical example: Document approval workflow

Consider a government agency that is building an AI-powered document approval system. The goal is to analyze incoming documents, classify them, and route them for human approval when necessary.

With Quarkus and LangChain4j, developers can build the workflow logic. Granite models provide the AI capabilities for classification and summarization. Watsonx ensures that every model used is tracked, versioned, and governed according to agency policies. Red Hat OpenShift AI provides the operational backbone, scheduling workloads on the right hardware and ensuring data governance across the pipeline.

From the developer’s perspective, this is just another Quarkus application. From the architect’s perspective, it is a secure, scalable, and compliant system that aligns with enterprise standards. From the decision maker’s perspective, it is an investment that will remain viable as AI patterns mature into Jakarta EE specifications.

The open path to trusted enterprise AI

Decision makers need to understand that models are not interchangeable commodities. The choice of model and platform defines the future viability of an AI strategy. Proprietary models might deliver quick wins, but they can also lock enterprises into opaque ecosystems that do not align with long-term needs.

Granite and watsonx provide a different path. They align AI adoption with the same principles that made Java successful: openness, interoperability, and long-term support. Combined with Quarkus, Jakarta EE, and Red Hat OpenShift AI, they provide a full-stack story where innovation and stability coexist.

This is not just about technology. It is about risk management. Choosing Granite and watsonx means choosing a future where AI investments remain valuable, portable, and governed. It means securing the enterprise’s future in the same way that adopting Java standards secured it two decades ago.

Securing the next decade of enterprise applications

AI is not just another tool to be bolted onto enterprise applications. It is a fundamental shift in how systems are built, deployed, and governed. For Java developers and architects, the question is not whether to adopt AI, but how to do it responsibly and sustainably.

Granite models provide the open foundation. Watsonx provides the governed environment. Quarkus and LangChain4j provide the innovation layer that makes AI accessible to Java developers. Jakarta EE and MicroProfile provide the standards that ensure those innovations remain stable and interoperable. Red Hat OpenShift AI provides the operational backbone that turns prototypes into production systems.

For software architects and IT decision makers, the message is clear. The future of enterprise applications will be defined by AI, but it does not have to be defined by risk. By investing in open standards, open models, and trusted platforms, enterprises can secure their future. Granite and watsonx are not just products, they are part of a larger ecosystem that ensures enterprise AI will be as stable, secure, and interoperable as enterprise Java has always been.

For an overview on what's happening at the standards level with Jakarta EE and AI, see the article What's Up With Open Standard Enterprise Java and AI.

Granite and watsonx: Open Models for the Next Generation of Enterprise AI with Java

Granite and watsonx: Open Models for the Next Generation of Enterprise AI with Java