The 30 second answer
With a fixed budget, I'd look at what each request costs in CPU and memory. In the video, Node.js used all 48 cores and about 10 to 12 GB of RAM at half a million. The rewrite in a native language reached about 1.2 million with roughly 10% CPU idle and half a gigabyte of RAM. Without a fixed budget, adding servers behind a load balancer is usually cheaper than a rewrite.
The two options he names. Spend more (a bigger instance with 64 cores) or make each request cheaper. The budget was fixed, so he rewrote the backend in Rust on his own HTTP framework. The backend was about 5,000 lines, and the AI-assisted rewrite took about a week, keeping the same file structure, routes and SQL so the code stayed readable.
The results he reports.
- A 60-second read test with 2,400 connections and 48 workers averaged about 1.2 million requests per second, 70 million requests in total. The views table and the log each gained about 70 million entries.
- The server process showed roughly 3,000% CPU and the machine kept about 10% CPU idle, so the app stayed usable during the test.
- Compared with Node.js, throughput doubled (about 1 million against about half a million), response time was about two times better, max response time was about a third of a second against over a second, and memory was about half a gigabyte against roughly 10 to 12 GB (48 Node.js workers each with its own V8 runtime).
What he explicitly does not claim. That this is Rust magic. He says any native language (C, C++, Go, Rust) would show the same kind of gain because it does less work per request. He puts Express, Flask, Django and Spring Boot in Java near the bottom of his chart, and says Go's Fiber, Vert.x in Java or Drogon in C++ could also reach 1 million. For CPU-heavy work per request he expects Rust and C++ to pull ahead of Go and Java. These are his benchmark impressions, not measurements shown in the video.
The senior framing. Scaling isn't linear in difficulty: an index gave 10x early on, but going from 50,000 to half a million needed a stream, a consumer and caching, and the last 2x needed a rewrite. A rewrite costs development time, so I'd justify it with cost per request at the target load. With a flexible budget, horizontal scaling behind a load balancer is usually the first move.
Alternate interview scenarios
"You're a Java and Spring Boot shop. Do you rewrite in Rust?"
Not by default. First profile where the CPU goes per request: serialization, logging, framework overhead, allocation and GC. A reactive or event-loop Java stack is an option the speaker himself names (Vert.x). Rewrite only the hot path, and only when the hardware cost of scaling out is larger than the engineering cost.