Code Metal raises $125M to modernize defense software with AI

The Boston startup Code Metal announced attracting $125 million in investment — and this news speaks not only to the ambitions of one company, but also to a massive shift in the defense industry's approach to its technological debt. For decades, military contractors have operated software systems written in COBOL, Fortran, and other languages of the past century, keeping them alive through the efforts of rare specialists whose average age steadily grows. Now AI takes on a task that people have postponed for far too long.

The problem of legacy code in the defense sector is one of the most underestimated technological threats of our time. Critical control systems, logistical platforms, targeting software — all of this often runs on code that is thirty, forty, or even fifty years old. Attempts to rewrite it manually traditionally end the same way: projects drag on for years, costs multiply several times over budget, and new bugs reproduce or even exacerbate the original problems. Code Metal positions itself precisely at this point of failure.

The key distinction of the company's approach is not the speed of translation, but verification. Most tools based on large language models can rewrite code from one language to another with enviable speed, but cannot guarantee that the result is identical in behavior to the original. For consumer applications, a small discrepancy is a bug report. For a missile system control unit or a target recognition algorithm — this is a catastrophe. Code Metal claims that its AI system does not simply translate code, but mathematically confirms the equivalence of the new code to the old: every branch of logic, every boundary condition, every side effect passes formal verification.

This direction — so-called formal verification — has existed in academia for a long time, but traditionally required colossal manual effort and deep mathematical expertise. Code Metal bets that modern AI models now allow automating this process at industrial scale for the first time. If this really works as the company claims, we're talking about a qualitative change in what is even possible when migrating legacy systems.

The $125 million investment round is a substantial signal to the market. The defense technology sector is experiencing a real investment boom after several years when Silicon Valley ostentatiously distanced itself from military contracts. The success of Anduril, Shield AI, and other defense-tech startups has reformatted the industry's perception: today defense technologies are again considered an attractive venture direction. Code Metal arrives on this wave with a product that addresses a specific, measurable pain point — not an abstract "security" or "efficiency."

For defense contractors themselves, the stakes are extremely high. The U.S. Department of Defense has spent years trying to modernize its technological base through programs like JEDI and its successors, but the infrastructure layer — the very software on outdated languages — remains a painful bottleneck. Finding a Fortran programmer capable of understanding forty-year-old code without documentation becomes harder each year. Automated migration with a guarantee of correctness — this is not just convenience, it's a matter of long-term combat readiness.

Code Metal faces obvious skeptics as well. Formal verification of complex systems is a task that computer science has not definitively solved even outside the AI context. Real defense code often contains intentional or accidental dependencies on specific behavior of hardware, the operating system, or the compiler — nuances that are extremely difficult for any automated tool to capture. Proving the technology works on actual military systems, not demonstration projects — this is a task the startup still has to solve publicly and convincingly.

Nevertheless, the very emergence of Code Metal with serious funding marks an important moment: the industry has finally acknowledged that AI is ready to tackle not only writing new code, but also paying off decades of accumulated technological debt. If this bet proves correct, the coming years will show how the very understanding of what it means to maintain critical software systems changes — not only in defense, but in energy, aviation, finance. Legacy code exists everywhere. The problem is universal. Code Metal simply started with the most demanding customer.