GraphQL isn't dead, and REST didn't win. The boring version is that GraphQL pays off when many clients need different slices of the same data, and costs you when one frontend talks to one backend.
Where the 35 percent came from
Moving to schema-first GraphQL cut client network overhead by around 35 percent. I still wouldn't recommend it to most teams.
The saving came from one place: mobile clients stopped making four calls to build one screen, and stopped downloading fields they threw away. Several different clients wanting different slices of the same data is the case GraphQL was designed for, and there it earns its keep. The schema-first discipline is also what made our multi-client platform manageable.
The cost that lands on the backend
The number hides what landed on the backend.
N+1 queries got harder to see. They moved from something you could spot in a controller to something buried behind resolvers, so DataLoader stopped being optional. Caching got harder too, because a CDN can't reason about a single POST body. Rate limiting by endpoint stopped meaning anything when one query can be cheap or can be a join across your whole graph. And you need query depth and complexity limits, or you've shipped a denial of service endpoint with a nice type system.
The wrong shape of product
I've shipped both at scale. The same GraphQL setup on a simple product would have been pure ceremony.
If you have one client and it's a web app, REST with a few well-shaped endpoints will beat GraphQL on total engineering time, and it isn't close. When GraphQL goes badly, it's usually because someone picked it for the wrong shape of product.
Building something like this?
I'm Ahmed Mamdouh, a senior full-stack and AI engineer. I reply within one working day.

Scaling 100k WebSocket connections: the reconnect storm
At 100k+ concurrent sockets the count is easy. The reconnect storm is what breaks, and jittered backoff, load shedding and resumable sessions fix it.

npm v12 blocks install scripts by default
npm v12 no longer runs preinstall, install or postinstall scripts unless you approve them, closing a common supply chain attack path.

Node.js moves to one major release a year
From Node 27, Node.js ships one major a year and every release becomes LTS, ending the odd/even split most teams already ignored.
