Which is better and why: Apollo Router or Apollo Gateway, for a high-throughput production system?

For a high-throughput production system, Apollo Router is the stronger choice, and the reasoning holds up on more than just raw benchmarks.

Its Rust runtime delivers meaningfully higher throughput and lower, more consistent latency under load than the Node.js-based Gateway, largely because Rust's memory model avoids the variance that garbage collection introduces at high concurrency. It also natively supports federated subscriptions, which Gateway doesn't handle robustly, and its Rhai/coprocessor extension model is where Apollo is investing new capability.

The counter-case: if a graph is genuinely small, latency-insensitive, and has significant existing investment in JS-based Apollo Server plugins that haven't been ported, staying on Gateway a bit longer may be pragmatic in the short term. But for anything actually hitting throughput or latency ceilings — which is exactly the "high-throughput" case in the question — Router is the recommended default, and Apollo's own guidance points toward eventually migrating regardless of current scale.