Java, AI Codeshare in Race to Modernize Legacy Apps

Should Generative AI ever become the electric-vehicle motor that directs and perhaps executes enterprise workflow for any organization in the world, Oracle wants to ensure that Java is the battery behind this new fleet of mobility solutions.

At its recent JavaOne event in Redwood Shores, that kind of interdependency was unmistakable with companies like Uber, Nvidia and Microsoft touting the use of Java for development of next-generation AI applications.

While the programming language has been the workhorse for enterprise computing since its 1995 debut, Oracle is shining new light on its strategic importance with the relaunch of JavaOne, now in its second year as a standalone event following the vendor’s move to spin it out separately after decades of folding it into Oracle OpenWorld, CloudWorld and now AI World after buying Java as part of its purchase of Sun Microsystems in 2010.

Before hundreds of developers, the three-day event unveiled a raft of enhancements including the latest Java Development Kit 26 and the resurrected Project Detroit that makes Java compatible with its competing language Python, as well as Java Verified Portfolio designed to streamline Java licensing.

The attendees caught a glimpse of Java and AI moving in tandem at Uber with the rideshare service optimizing deep-learning performance and internal resources by making double-digit CPU reduction possible with each iteration of Michelangelo, one of the most advanced machine-learning systems for understanding riders and drivers that has been in use at Uber since 2016.

The Uber internal end to end AI platform is now capable of training 20,000 models per month and clocking 40 million peak predictions per second, after running tools like JDK 8, 11 and 21 through the years as the building blocks for data management, model training and sharing as well as reusing matching engine that forecasts and accommodates the needs of cities, drivers and riders.

With the use of Java and other tools, Nvidia has been able to run its cuVS, an open-source library for developers to enhance data mining and semantic search workloads, such as recommender systems and retrieval-augmented generation (RAG), according to Ikroop Dhillon, director of developer relations at Nvidia. The programming language makes AI vector search run faster within the Java ecosystem that now turns to Nvidia for their latest Gen AI projects with speed and stability among their top concerns.

As the world’s most valuable company better known for its semiconductor products, Nvidia actually places greater emphasis on its software development efforts than hardware, Dhillon said. In other words, it cares more about to what extent its developers – especially those that are steeped in Java – can readily access tools that ultimately help them achieve more with their Gen AI projects.

At Microsoft, Azure now handles 500 billion messages per day on Apache Kafka clusters written in Java, in addition to five trillion messages per day all using Java as the backbone for such massive amounts of computation.

A diamond sponsor of JavaOne, Microsoft has had its fair share of coopetition with Java, or Oracle for that matter, through the years, but the consensus appears that both sides stand to gain if enterprise developers are sticking with Java and by extension the Microsoft developer ecosystem that covers everything from Azure Foundry to Github since Java has always been touted as the tool that compiles once and runs anywhere.

All these partnerships to ensure Java stays competitive with open-source alternatives from Python to Ruby, not to mention debunking the myth that vibe coding – made easier by agents and prompts – is turning programming either into a trivial task or a commodity that could be discarded quickly.

The staying power of Java with its better and more predictable memory consumption for large-scale enterprise applications than Python, coupled with its proven stability and performance, should buttress the case for anyone trying to position Java as not just the right tool for Cloud and AI applications, but also the catalyst for accelerating coding and modernizing legacy systems.

Increasingly, what’s at stake for all the above partners is to find ways to keep Java open, portable and robust enough for next-generation enterprise applications, a tall order given the hyper-competitive environment where the specter of making wrong moves could have serious ramifications.

During the same week of JavaOne, OpenAI also announced the acquisition of Astral, a startup behind a viral Python coding tool used by millions of developers. Meanwhile, Anthropic has been making big waves with its Claude Code, another widely popular coding tool for millions of AI developers. All these pose serious challenges to Java, or the future of enterprise computing for that matter.

On staying open by enabling the coexistence of Java and Python, Georges Saab (senior vice president, Oracle Java Platform and chair, OpenJDK governing board) said the decision to resurrect Detroit – first proposed in 2018 and then abandoned for various reasons – underscores the fact instead of “chasing the stack,” which could mean performance degradation, a more pragmatic approach is to ensure the best runtime implementation of each language will be consistent and available since they will be compatible, rendering obsolete the need for any of its customers to reskill or retool themselves in order to run Python apps.

As part of the new Java Verified Portfolio for making Java licensing more accountable and auditable, Oracle is also making widely available Helidon- previously supported just for WebLogic and Coherence customers – and bringing back JavaFX, a popular tool for running apps on different operating systems.

Again, the moves are expected to boost the retention of enterprise customers that are likely to stick around longer given that there are more open and portable tools for them to sustain the intrinsic value of Java now that many of their legacy systems need to be modernized.

Oracle reiterated that the premise is not just about retaining current/existing customers, but rather any organization or individual developer who wants to become an OCI or Java SE customer by leveraging Java Verified Portfolio.

By most measures, Java has been revolutionary after paving the way for the creation of 73 billion active Java Virtual Machines including 51 billion Cloud-based Java Virtual Machines that run millions of Java programs everywhere, practically touching every single website on the client side. Estimated to be used in 97% of enterprise computers, Java’s importance on the backend is more evident given its 30-year-plus history and how instrumental it has been in handling enterprise-class web applications.

The rub is that many of these applications are not designed for the modern era and rewriting them will not be easy. Simon Martinelli, a Swiss-based Java consultant, said these modernization projects for such mission-critical processes like billing and point-of-sale applications in a multi-channel eCommerce environment could take years to pull off.

Similarly, an attendee who is a Java developer with an aerospace company specializing in flight planning systems, said shrinking IT budgets, coupled with the speed and performance of some of these AI applications, now are putting more pressure on them to figure out the most cost-effective way to modernize their legacy applications and whether Java can deliver for them in a realistic timeline against competing tools.

To that end, the recurring theme of JavaOne has been “Java is everywhere AI needs to be,” in addition to the pledge from Oracle’s long-standing stewardship of Java that revolves around integrity by default and that Java is to help write code developers can trust, a deciding factor that could make or break a deal in the AI era.

List of Oracle Java Platform Customers

Source: APPS RUN THE WORLD Technographics Platform, March 2026

• Industry Events
• Artificial Intelligence (AI)
• Feature