For years, Python has dominated the conversation around artificial intelligence development, commanding headlines and attracting the lion’s share of attention from data scientists and machine learning engineers. But beneath the surface, a far older and more entrenched programming language has been steadily carving out a significant role in enterprise AI: Java. According to a sweeping new survey from Azul, the company behind high-performance Java runtimes, nearly two-thirds of Java developers are now using the language to build AI applications — a figure that underscores a quiet but powerful shift in how large organizations are approaching artificial intelligence infrastructure.
The findings, published in Azul’s 2025 State of Java report, draw on responses from more than 2,000 Java professionals worldwide. The results paint a picture of a language that is not merely surviving in the age of AI but actively thriving, propelled by its deep roots in enterprise computing, its robust ecosystem of libraries, and the sheer scale of existing Java codebases that companies are reluctant — and often unable — to abandon.
Enterprise Inertia Meets AI Ambition
As reported by InfoWorld, the Azul survey found that 64% of respondents are leveraging Java for AI development work. The most commonly cited tools include JavaML, the Deep Java Library (DJL) developed by Amazon, and OpenCL for GPU-accelerated computing. These frameworks allow Java developers to build, train, and deploy machine learning models without leaving the ecosystem they already know — a critical advantage in large organizations where retraining thousands of developers on an entirely new language stack would be prohibitively expensive.
The appeal of Java in AI contexts is not primarily about raw performance in model training, where Python and its C/C++ underpinnings still hold advantages. Rather, Java’s strength lies in what happens after the model is built: deployment, scaling, integration with existing enterprise systems, and long-term maintenance. Java’s type safety, mature tooling, and battle-tested concurrency frameworks make it particularly well-suited for running inference workloads in production environments where reliability and uptime are non-negotiable.
The Library Ecosystem Expands
Amazon’s Deep Java Library has emerged as a particularly important piece of the puzzle. DJL provides a framework-agnostic interface that allows Java developers to work with models built in TensorFlow, PyTorch, MXNet, and other popular machine learning frameworks. This interoperability is crucial because it means organizations don’t have to choose between their existing Java infrastructure and the latest advances in AI research, which are typically published with Python implementations. DJL effectively serves as a bridge, letting Java applications consume and serve models regardless of their origin.
JavaML, meanwhile, offers a collection of machine learning algorithms implemented natively in Java, covering classification, clustering, feature selection, and other core tasks. For organizations that need to run machine learning workloads entirely within a Java Virtual Machine (JVM) environment — whether for regulatory, security, or architectural reasons — JavaML provides a self-contained solution. OpenCL support, for its part, enables Java applications to tap into GPU acceleration, narrowing the performance gap with Python-based toolchains that have long benefited from CUDA and other GPU computing frameworks.
Why Companies Aren’t Abandoning Java for Python
The persistence of Java in AI development reflects a broader truth about enterprise technology adoption: organizations rarely rip and replace their core infrastructure, even when newer alternatives appear superior for specific tasks. Java remains the backbone of financial services, healthcare IT, logistics, telecommunications, and government systems worldwide. According to various industry estimates, there are between 30 and 45 million Java developers globally, and the language consistently ranks among the top three most widely used programming languages in every major index, including TIOBE and Stack Overflow’s annual developer surveys.
For these organizations, the question is not whether Python is better for prototyping neural networks — it often is — but whether it makes sense to introduce an entirely separate technology stack for AI when Java can handle the production deployment side effectively. Many enterprises are adopting a hybrid approach: data scientists build and experiment with models in Python, while engineering teams deploy and operate those models in Java-based production systems. The Azul survey results suggest this pattern is becoming the norm rather than the exception.
Java’s Performance Story Gets Stronger
Recent developments in the Java platform itself are also making the language more competitive for computationally intensive AI workloads. Project Panama, an ongoing effort within the OpenJDK community, is designed to improve the connection between Java code and native libraries, making it easier to call into optimized C/C++ machine learning libraries from Java without the overhead and complexity of the Java Native Interface (JNI). The Vector API, introduced as an incubator feature in recent Java releases, enables developers to write code that takes advantage of SIMD (Single Instruction, Multiple Data) hardware instructions — the same kind of low-level parallelism that powers high-performance numerical computing.
Project Valhalla, another major initiative, promises to bring value types to Java, which could significantly reduce memory overhead for the kind of large-scale data processing that AI applications demand. And Project Loom, which introduced virtual threads in Java 21, has already begun transforming how Java applications handle concurrent workloads — a direct benefit for AI inference servers that must process thousands of requests simultaneously. Together, these platform-level improvements are systematically addressing the historical performance criticisms that have been leveled at Java in comparison to lower-level languages.
The Competitive Dynamics of AI Tooling
Java’s growing role in AI also has implications for the competitive dynamics among JDK vendors. Azul, which sells commercial Java runtimes and performance monitoring tools, has an obvious interest in promoting Java’s relevance in AI. But the company is far from alone in this effort. Oracle, which stewards the official Java Development Kit, has been investing heavily in AI-related capabilities and partnerships. Microsoft, through its involvement in the OpenJDK community and its Azure cloud platform, has been making it easier to run Java-based AI workloads at scale. Amazon, beyond its DJL contribution, offers Corretto, its own no-cost, production-ready distribution of OpenJDK, which is optimized for AWS environments where many AI workloads run.
The spring ecosystem, arguably the most widely used Java application framework, has also been moving aggressively into AI territory. Spring AI, a relatively new project under the Spring umbrella, provides abstractions for working with large language models, vector databases, and retrieval-augmented generation (RAG) patterns — the building blocks of modern generative AI applications. This means Java developers can now build sophisticated AI-powered applications using the same Spring patterns and conventions they’ve relied on for years to build web applications and microservices.
What the Numbers Signal for the Industry
The Azul survey’s findings arrive at a moment when enterprises across every sector are racing to embed AI capabilities into their existing products and operations. Unlike the previous waves of AI hype, which were concentrated in research labs and startups, the current generative AI boom is being driven largely by established enterprises seeking to augment their existing systems. These are precisely the organizations where Java is most deeply entrenched, and they are increasingly looking for ways to bring AI into their Java-based architectures rather than migrating away from them.
According to InfoWorld, the survey also found that cloud-native development practices continue to expand among Java shops, with containerization and Kubernetes adoption climbing steadily. This dovetails with AI deployment trends, as containerized microservices architectures are increasingly the preferred way to deploy machine learning models in production. Java’s strong support for containerized environments — including optimized startup times through GraalVM native images and CRaC (Coordinated Restore at Checkpoint) — positions it well for this convergence of cloud-native and AI-native development.
A Language That Refuses to Fade
The narrative around Java has long been one of a language perpetually on the verge of decline, always about to be overtaken by something newer and shinier. Kotlin, Scala, Go, Rust, and of course Python have all been positioned at various times as Java’s successor. And yet, year after year, Java not only persists but adapts. The language’s six-month release cadence, adopted in 2017, has allowed it to evolve far more rapidly than in its earlier years, incorporating modern language features like records, sealed classes, pattern matching, and virtual threads at a pace that would have been unthinkable a decade ago.
The AI era may ultimately prove to be not Java’s twilight but its renaissance. With nearly two-thirds of Java developers already engaged in AI work, the tools and frameworks maturing rapidly, and the JVM platform itself evolving to meet the demands of computationally intensive workloads, Java’s role in artificial intelligence is no longer a curiosity — it is a strategic reality that enterprise technology leaders ignore at their peril. The 30-year-old language, it turns out, still has plenty of new tricks to learn.
WebProNews is an iEntry Publication