Wasmer rewrites Django backend in Rust, cuts compute and RAM sharply

Wasmer’s Rust rewrite story lands because it comes with a before-and-after bill, not a vague promise. The company says porting its Django backend to Rust cut compute from 220 CPUs to 24 and memory from 800 GB to 64 GB, while also shrinking p95 API latency from 120 ms to 30 ms. That is the kind of number that forces a real reality check on the usual “rewrite it in Rust” pitch.

The old backend had been in service for seven years, originally built to store package information when wapm.io was Wasmer’s only dynamic website. By the time of the rewrite, one person had spent about three months full time on the port, with AI assistance, and the work was brought in gradually as the new backend stabilized and more team members joined. Wasmer’s CEO and founder, Syrus Akbary, wrote the post, and his background matters: he is also the creator of Graphene-Python, so he knows the Python GraphQL world from the inside.

The savings did not come from language choice alone. Wasmer said the backend had become increasingly hard to maintain because Django’s async story did not fit a GraphQL-heavy service cleanly, Python’s gradual typing did not give the team the type safety it wanted, and the Python backend sat awkwardly beside Rust-based edge components the company already ran. Wasmer also said it lacked an in-house Python expert who could keep rescuing the codebase. That points to the bigger pattern behind many Rust migrations: the language swap is only the visible part, while architecture cleanup and ownership cleanup do a lot of the real work.

The operational numbers back that up. Average CPU utilization fell from 80% to 30%. Database connections dropped from thousands to hundreds. Queries ran five to 10 times faster. Worst-case startup time went from more than 60 seconds to about one second. Wasmer says the comparison was measured under similar traffic patterns before and after the migration, which makes the result look less like marketing and more like an engineering ledger.

Django’s own documentation helps explain why this sort of backend can become awkward. Its async guidance says persistent database connections should be disabled in async mode, and its database docs describe a synchronous model built around opening and reusing connections. A 2025 Django forum discussion even focused on async-capable database connections and cursors. None of that proves Wasmer’s exact bottleneck, but it does show where the friction lives when a GraphQL-heavy service outgrows a synchronous Python stack.

That is the useful takeaway for Rust teams: judge a rewrite by what it removes, not just what language it uses. In Wasmer’s case, the win looks less like “Rust is faster” and more like a stack that finally stopped carrying seven years of mismatch, connection churn, and operational weight.