Eliminating $50B in Technical Debt: Fujitsu Unveils AI-First COBOL Refactoring

For decades, the global financial and administrative sectors have been held hostage by COBOL. Despite being over 60 years old, Common Business-Oriented Language still powers approximately 80% of in-person banking transactions and 95% of ATM swipes worldwide. Today, Fujitsu announced a landmark Generative AI service that promises to automate the daunting task of refactoring these legacy systems into modern, cloud-native Java architectures.

The Problem: A Vanishing Workforce and Spaghettified Code

The "COBOL crisis" is two-fold. First, the original authors of these systems are retiring, leaving a critical talent gap. Second, these systems have undergone decades of "patchwork" maintenance, resulting in spaghettified logic that is nearly impossible to document manually. Fujitsu’s AI service, part of its Kozuchi AI platform, utilizes a large language model (LLM) fine-tuned on half a century of enterprise code to understand not just the syntax, but the business logic buried in these ancient files.

Traditional modernization attempts often fail because they try to "lift and shift" the code, leading to buggy, unmaintainable translations. Fujitsu’s approach is different: the AI first generates a comprehensive "business logic map" and design documentation. Once human architects verify the logic, the AI performs a "clean-room" refactoring into Java, ensuring that the new code follows modern microservices patterns rather than simply mirroring the legacy procedural structure.

Benchmarks: 80% Reduction in Modernization Time

Early pilot projects with major Japanese banks have yielded staggering results. Fujitsu reports an 80% reduction in the time required for the initial discovery and documentation phase—a process that used to take months of manual forensic accounting of the code. The subsequent refactoring and testing phases saw a 60% increase in efficiency, primarily due to the AI's ability to automatically generate unit tests for the newly created Java modules.

The AI also excels at identifying "dead code"—segments of the legacy system that are no longer used but still consume compute resources. In one case study, Fujitsu’s AI identified that nearly 25% of a client’s COBOL codebase was obsolete, allowing for a much leaner and more efficient cloud deployment after refactoring. This "pruning" effect is critical for organizations looking to reduce their data center footprint and operational costs.

Security and Compliance in the AI-Refactor Era

For enterprise clients, security is the paramount concern. Fujitsu has addressed this by offering the modernization service as an on-premise or sovereign cloud solution. This ensures that the sensitive business logic and proprietary data contained within the COBOL files never leave the client's secure environment during the training or inference phases of the AI model. This "privacy-first" approach is essential for the highly regulated financial and government sectors.

The service also includes an automated "Compliance Checker" that scans the newly generated Java code for common security vulnerabilities (OWASP Top 10) and ensures adherence to industry-specific regulatory standards like PCI-DSS and GDPR. By baking security into the automated refactoring process, Fujitsu is mitigating the risks associated with manual human error during massive system overhauls.

Conclusion: The End of Technical Debt

Fujitsu’s Generative AI service isn't just a tool; it's a structural solution to one of the most persistent bottlenecks in the global economy. By turning the "impossible task" of COBOL modernization into a repeatable, automated process, they are enabling a massive wave of innovation in sectors that have been stagnant for decades. As legacy technical debt is finally liquidated, we can expect a surge in the speed and quality of digital services in banking, insurance, and public administration worldwide.