Defence software has a debt problem.
Weapons platforms, C2 systems, fire control — written in Ada, Fortran, COBOL, VHDL. Languages from the 70s and 80s. The engineers who know them are retiring. Nobody new is learning them.
Here's how bad it's gotten: the US DoD is actively retraining engineers in Ada rather than Rust. Not because Ada is better. Because they can't hire enough Rust engineers who also hold the right clearances. They've given up on modernisation and started teaching a dead language instead. That's not a skills gap. That's a structural forcing function.
The translation problem isn't just a talent issue — it's a safety issue. A bug in a missile fire-control algorithm isn't a sprint ticket. Manual translation by scarce Ada/Fortran specialists takes 12–18 months per module. One logic error in the translated output can kill a programme. Or worse. So MoDs default to expensive contractors or don't modernise at all, accepting mounting technical debt and declining NATO interoperability at the same time.
Code Metal (Boston) proved this is a venture-scale category. $200M raised, $1.25B valuation, 18 months. The US Air Force and L3Harris are customers. Their insight: LLM generation alone is not enough. AI hallucinates. In safety-critical code, a hallucination is catastrophic. The moat is pairing generation with formal verification — not just outputting new code, but mathematically proving the translated output is functionally equivalent to the original. They're now an "Awardable" vendor in the US CDAO marketplace.
Verification is the moat, not generation. LLM code generation is commoditising fast — GitHub Copilot, Cursor, Claude. Proving the output is correct is the hard, defensible, non-commoditisable part. And it compounds. Every codebase you translate adds edge cases to your verification suite. More edge cases mean higher confidence thresholds. Higher confidence thresholds mean higher switching costs. It's a data flywheel, but for proving software is safe.
Here's the European opportunity.
Code Metal has Raytheon (RTX) on its cap table. That's not a footnote — it's a structural disqualifier for European sovereign procurement. No European MoD is handing weapons system source code to a vendor backed by an American prime. It's not even a policy debate. It just won't happen.
And there's no European equivalent. Not at venture scale.
What does exist: Europe has accidentally pre-built the verification infrastructure layer. TrustInSoft (Paris, ~$7M raised) verifies C/C++/Rust using Frama-C — Thales became a customer in 2025, and they extended to Rust in November. AbsInt (Saarbrücken) ships CompCert — the world's only formally verified optimising C compiler, used by Airbus on A380 flight controls since 2003. Ferrous Systems (Berlin) built Ferrocene, the first TÜV SÜD-qualified Rust compiler for safety-critical environments, now partnered with TrustInSoft.
The verification stack exists. What's missing is the AI translation engine that sits on top of it.
The GTM architecture is dual-use by design. MoD and prime contracts de-risk the technology and build the verification track record. Commercial — automotive OEMs, semiconductor fabs like ASML and NXP — provides the volume and margins to scale. Same product, two very different procurement timelines. The defence contract makes you trusted. The commercial contract makes you profitable.
The window is opening now. The EDF 2026 Work Programme earmarks €40M explicitly for replacing legacy defence systems — deadline September 2026. EUDIS 2026 offers up to €2.7M per project for European SMEs, zero dilution. The funding infrastructure exists before the companies do. That gap doesn't stay open for long.
The company I'm looking for: European-incorporated, on-prem deployable for air-gapped environments, formal verification as proprietary IP, no US prime anywhere near the cap table. Founding team out of formal methods research — INRIA, TU Delft, ETH Zurich — or ex-prime engineers who lived this problem from the inside and know exactly which programme office will pay first.
If you're building this — ties@keenventurepartners.com