Code Metal Raises $125 Million to Rewrite the Defense Industry’s Code With AI

Code Metal, a Boston-based startup that uses AI to write code and translate it into other programming languages ​​just closed a $125 million Series B funding round from new and existing investors. The news comes just a few months after the startup raised $36 million in Series A funding led by Accel.

Code Metal is part of a new wave of startups that aim to modernize the tech industry by using AI to generate code and translate it across programming languages. However, one of the lingering questions about AI-assisted code is whether the output is correct – and what the consequences might be if it isn't.

Over the past two years, companies like Antithesis, Code Rabbit, Synthesized, Theorem and Harness have all secured millions in venture capitalist backing for their approaches to automating, validating, testing and securing AI-generated code. These startups are selling the “picks and shovels” of the AI ​​gold rush — tech tools that serve a larger industry. While some of the methods behind their technology remain unproven, investors are willing to gamble that at least a few will pan out.

Code Metal, which was founded in 2023, has focused its efforts on code translation and code verification for the defense industry. It has L3Harris, RTX (formerly known as Raytheon), and the US Air Force as early customers. The startup also works with Japanese electronics company Toshiba, and says it is in talks with a major chip company to work on code portability across chip platforms, though the company l declined to say which.

The startup's software platform translates code from high-level programming languages ​​like Python, Julia, Matlab, and C++, to lower-level languages ​​or code that runs on specific hardware, like Rust, VHDL, and chip-specific languages ​​like Nvidia's CUDA.

Code Metal CEO Peter Morales, who previously worked at Microsoft and the MIT Lincoln Laboratory, says the market is starting to recognize “the big tentpole issues” in an industry that could be supported by AI-generated code in the not-so-distant future. One of those problems is porting old code into new applications. If a government agency or defense contractor needs coding work done quickly, but only has access to engineers who have specialized in a legacy programming language, he says that slows everyone down.

Morales quotes a recent post on X from well-known AI researcher Andrej Karpathy, who observed, among other things, the “rising momentum behind porting C to Rust.” Karpathy concluded: “It feels likely that we will end up rewriting large fractions of all the software that has ever been written many times.”

“That's all we do in one tweet,” Morales says.

One of Code Metal's investors, Yan-David Erlich, a general partner at B Capital, says the reality is that some of the code that controls essential communications infrastructure, and even satellites, “is old, it's cramped, it's written in programming languages ​​that people may not use anymore. It needs to be modernized.”

“But in the course of the translation,” added Erlich, “you can insert bugs – which is catastrophically problematic.”

That's where Code Metal says its proprietary tech comes in. Morales says that at each step of translation, Code Metal's software generates a series of test harnesses—a virtual container of data and tools—that evaluate the code and show customers along the way that it works. When asked about Code Metal's translation error rate, Morales says that it largely depends on how difficult the code conversion is, but that for the pipelines Code Metal is currently running, “there's no way to generate an error. The software will just say, 'There's no solution for this,' if we can't complete the translation.”