Can Java Overtake Python in AI? Insights from the 2025 Azul Report

According to The New Stack, 2025 will be the last year Python leads AI, and Java is poised to take over.

Simon Ritter, Vice President of Engineering at Azul Systems, cites research and a developer survey indicating that Java could erode Python's AI dominance within a year and a half.

Ritter explains that while Python’s simplicity and rich library ecosystem make it popular among AI researchers, enterprises favor Java for its scalability, performance, and suitability for large‑scale, mission‑critical AI applications.

Enterprise challenges

Ritter warns that as enterprises rely more on AI, they will face obstacles if they cannot run Java and Python side‑by‑side, requiring extensive rewrites and redesigns to keep Java AI applications future‑proof.

Java’s advantage lies in projects like Project Panama, which eases native library integration, and Project Babylon, which aims to enable direct GPU usage from Java without code changes.

Azul’s survey shows 50% of respondents building AI features use Java, surpassing Python and JavaScript.

Java’s suitable use cases

The report highlights Java’s scalability, extensive libraries, and seamless integration with existing enterprise systems as key strengths for AI development.

JavaML is identified as the most commonly used Java AI library, and 72% of surveyed developers say their compute consumption must increase to support AI‑enabled Java applications.

Can Java truly take over?

IDC analyst Arnal Dayaratna believes Java could become the leading AI development language due to its unmatched enterprise‑grade capabilities.

Conversely, Omdia analyst Brad Shimmin doubts Java will surpass Python, noting Python’s performance gains, thriving ecosystem (PyTorch, Pandas), and developer preference, especially for generative AI.

Some developers argue the Java community needs more language innovations to better serve AI workloads.

Oracle’s efforts

Oracle’s senior VP Georges Saab emphasizes Java’s strong typing, memory safety, and extensive tooling as natural fits for emerging AI technologies.

He points to projects such as Babylon (GPU programming), Valhalla (value types), and Amber (sealed types, records, pattern matching) that aim to improve Java’s performance and expressiveness for AI training and inference.

Saab also mentions langchain4j, which allows large language model outputs to be directly mapped into Java record types, enabling developers to stay within the Java ecosystem for AI solutions.

Overall findings

Azul’s report, based on responses from over 2,000 Java professionals worldwide, shows 99% of surveyed organizations actively use Java, with nearly 70% of applications built on or running on the JVM.

Security and DevOps

The survey highlights DevOps productivity issues: 62% cite dead code, 33% spend over half their time handling false‑positive Java security alerts, and 49% still encounter Log4j vulnerabilities in production.

CEO Scott Sellers notes that Java’s continued dominance drives interest in Oracle alternatives, cloud‑optimized strategies, and AI innovation within enterprise environments.