diff --git a/pkg/storegateway/bucket.go b/pkg/storegateway/bucket.go
index 5b539e1d642..82a540da602 100644
--- a/pkg/storegateway/bucket.go
+++ b/pkg/storegateway/bucket.go
@@ -562,17 +562,6 @@ func (s *BucketStore) Series(req *storepb.SeriesRequest, srv storepb.Store_Serie
 		err = status.Error(code, err.Error())
 	}()
 
-	if s.queryGate != nil {
-		tracing.DoWithSpan(srv.Context(), "store_query_gate_ismyturn", func(ctx context.Context, _ tracing.Span) {
-			err = s.queryGate.Start(srv.Context())
-		})
-		if err != nil {
-			return errors.Wrapf(err, "failed to wait for turn")
-		}
-
-		defer s.queryGate.Done()
-	}
-
 	matchers, err := storepb.MatchersToPromMatchers(req.Matchers...)
 	if err != nil {
 		return status.Error(codes.InvalidArgument, err.Error())
@@ -620,6 +609,16 @@ func (s *BucketStore) Series(req *storepb.SeriesRequest, srv storepb.Store_Serie
 		readers = newChunkReaders(chunkReaders)
 	}
 
+	// Wait for the query gate only after opening blocks. Opening blocks is usually fast (~1ms),
+	// but sometimes it can take minutes if the block isn't loaded and there is a surge in queries for unloaded blocks.
+	tracing.DoWithSpan(ctx, "store_query_gate_ismyturn", func(ctx context.Context, _ tracing.Span) {
+		err = s.queryGate.Start(ctx)
+	})
+	if err != nil {
+		return errors.Wrapf(err, "failed to wait for turn")
+	}
+	defer s.queryGate.Done()
+
 	var (
 		// If we are streaming the series labels and chunks separately, we don't need to fetch the postings
 		// twice. So we use these slices to re-use them. Each reuse[i] corresponds to a single block.