Hypercubic Brings Modern AI to the COBOL Mainframes That Still Run the Fortune 500

Hypercubic, a Fall 2025 Y Combinator-backed startup, is tackling one of the most valuable and least glamorous technology problems in the global economy: the decades-old COBOL codebases running on mainframes that continue to power the core systems of banks, insurers, airlines, telecoms, and retailers. The company is building an AI-native platform to maintain, understand, and gradually modernize those systems – without forcing the risky, failure-prone rewrites that have defeated generations of enterprise IT projects.

Founded in 2025 by Sai Gurrapu and Aayush Naik, Hypercubic is starting from a set of widely acknowledged facts that have, until now, been impossible to act on. Roughly 70% of Fortune 500 companies still rely on mainframes to run their mission-critical business applications. Those systems, many of them written between the 1960s and 1990s, continue to process trillions of dollars of transactions, insurance claims, airline reservations, and phone bills every year. And they have grown increasingly opaque as the developers who wrote and maintained them have retired or left the workforce, often taking their undocumented institutional knowledge with them.

Replacing that software has been a strategic priority at most large enterprises for twenty years. It has also been a serial disaster. Full-scale rewrites from COBOL to Java or cloud-native stacks routinely run years over schedule and hundreds of millions of dollars over budget, and occasionally fail spectacularly in production. The core difficulty is not the language itself but the accretion: decades of business logic, regulatory requirements, and edge cases encoded into hundreds of thousands or even millions of lines of COBOL, with comments missing, variable names cryptic, and behavior entangled with the quirks of the underlying mainframe operating system.

Hypercubic’s bet is that modern large language models are, for the first time, genuinely useful on this problem. AI models can now read and reason about COBOL at a level that enables real work – explaining what a module does, identifying dependencies, surfacing the implicit business rules inside a procedure, generating test cases, and producing candidate modernized implementations that engineers can review and harden. The result is not an overnight rewrite but a COBOL modernization platform that gradually makes legacy systems more understandable, more testable, and more tractable to migrate – one subsystem at a time, at a pace enterprise IT leaders can actually manage.

That incremental approach matches how enterprise IT leaders actually talk about the problem. No CIO running a core banking platform is going to flip a switch from a 40-year-old COBOL system to a new cloud stack. What they will do is invest in tools that let their shrinking pool of mainframe engineers be more productive, that capture business logic before the remaining experts retire, and that make it possible to refactor and extract services over a multi-year horizon. Hypercubic is positioning itself as the software layer that makes that realistic.

The market is huge and unusually durable. Mainframe spending has, contrary to expectations, remained essentially flat for two decades, because the systems continue to be business-critical and nothing has been able to replace them. A two-person team is a modest starting point for a business of this scope, but Hypercubic’s founders are betting that AI has finally given them a usable handle on a problem that has resisted every prior generation of tooling.