Should you rewrite it in Rust?

The three cases where Rust wins

• Compute cost dominates — If your cloud bill is majority CPU on a hot path, a Rust rewrite of that path routinely cuts instance count 3-10x. At $40K/month of compute, that's a real payback window. At $4K/month, it never pays.
• Tail latency is the product — Trading systems, ad bidding, real-time telemetry ingestion. No GC pauses and predictable memory behavior are the actual features — not raw throughput.
• Memory safety is a liability — Existing C or C++ code where use-after-free and buffer overflows are causing incidents or failing security review. Rust here replaces a genuinely unsafe foundation, not a merely slower one.

What a rewrite actually costs

Full rewrites of a mature service run 1.5-3x the original build cost, because you're re-deriving behavior that was never written down. Every undocumented edge case is a defect waiting to surface in production after cutover.

Concretely: a service that took two engineers six months to build typically takes two Rust-fluent engineers eight to fourteen months to replace faithfully, plus a parallel-run period. Budget $250K-$700K for anything non-trivial, and add hiring time — the senior Rust market is small and expensive in every U.S. metro.

The incremental path we actually recommend

Profile first and find the 5% of code burning 80% of the CPU. In our experience that's almost always serialization, parsing, image or vector processing, or a tight loop over a large collection.

Then extend rather than replace: PyO3 to call Rust from Python, napi-rs for Node, or a small standalone Rust service behind the existing API. You get the performance win in weeks, keep the business logic where your team can read it, and you learn whether your org can actually maintain Rust before betting the whole system on it.

If after six months the Rust component is well-maintained and the team is comfortable, widening its footprint is a decision you can now make with evidence instead of enthusiasm.

The organizational question nobody asks

Can you hire two more Rust engineers in your market within 90 days, and can your existing team review Rust pull requests competently? If either answer is no, the rewrite doesn't fail on performance — it fails when the one engineer who wrote it leaves and the service becomes untouchable. We've been called in to rescue exactly that situation more than once, and the fix is usually a partial rewrite back to a stack the team can staff.