Teams facing slow, timeout-prone fan-out calls often reach for GraphQL as the fix, expecting the new API layer to solve what was really a sequential-calling problem underneath. This talk walks through a real rearchitecture where switching to GraphQL alone changed nothing — the same blocking, one-at-a-time calls to an external source were still there, just hidden behind a new schema. The actual fix came from three pieces working together: a DataLoader pattern to batch and deduplicate calls, batch-size tuning found through real timeout failures, and time-boxed virtual threads to bound how hard the system could hit a source it didn't control. The final design also made a deliberate tradeoff — choosing partial responses over waiting for everything, and surfacing what timed out through GraphQL's error channel instead of silently dropping it. This session covers what actually breaks when fan-out meets scale, and the concurrency discipline that fixes it — not the API layer wrapped around it.