Code Metal just landed a $1.25 billion valuation. Its bet is that the hardest AI coding work is not writing new apps, it’s safely rewriting the old ones.
Boston-based Code Metal raised $125 million in a Series B led by Salesforce Ventures, with Accel, B Capital, and J2 Ventures also participating, according to WIRED. The company, founded in 2023, told WIRED it’s already profitable and generating positive cash flow.
The startup translates legacy code into modern targets, then verifies the output still behaves correctly, a must-have in regulated environments where a “small” translation bug can trigger costly revalidation. Code Metal says early customers include the US Air Force, L3Harris, RTX, and Toshiba. In that same report, CEO Peter Morales said the company is tackling “tentpole problems” where modernization can’t come with translation errors.
What Code Metal actually does
Code Metal’s platform translates software written in higher-level languages such as Python, Julia, Matlab, and C++ into lower-level or hardware-specific languages such as Rust, VHDL, and Nvidia CUDA, per WIRED. The appeal is speed plus portability, especially when teams need software to run on new architectures without rebuilding everything by hand. Investor appetite for practical AI tooling is also visible in big funding totals, including a record $150 billion AI startup haul in 2025.
How it tries to avoid ‘AI wrote a bug’
Code Metal generates test harnesses during translation to check whether the converted code behaves as expected, WIRED reported. Morales also said the system can decline to complete a translation when it can’t verify correctness.
That “prove it” posture is showing up across developer tooling, from near-instant coding models to the broader field of generative AI startups.
Also read: This $14B startup is building a ‘brain’ for all robots.