diff --git a/pkg/telemetry/serve.go b/pkg/telemetry/serve.go
index 038790b5ff..ace3849658 100644
--- a/pkg/telemetry/serve.go
+++ b/pkg/telemetry/serve.go
@@ -29,6 +29,9 @@ func NewServeProvider(ctx context.Context) (provider *Provider, otelEnabled bool
 
 	otelCfg := appConfig.OTEL
 	hasRegisteredProcessors := HasRegisteredSpanProcessors()
+
+	handleUnusedEndpoint(&otelCfg)
+
 	if otelCfg.Endpoint == "" && !otelCfg.EnablePrometheusMetricsPath && !hasRegisteredProcessors {
 		return nil, false, nil
 	}
@@ -74,3 +77,20 @@ func NewServeProvider(ctx context.Context) (provider *Provider, otelEnabled bool
 
 	return p, true, nil
 }
+
+// handleUnusedEndpoint enables tracing by default when an OTLP endpoint is
+// configured but both tracing and metrics are disabled, so the server can start
+// normally instead of crashing with a fatal validation error.
+func handleUnusedEndpoint(otelCfg *config.OpenTelemetryConfig) {
+	if otelCfg.Endpoint == "" {
+		return
+	}
+	tracingOff := otelCfg.TracingEnabled == nil || !*otelCfg.TracingEnabled
+	metricsOff := otelCfg.MetricsEnabled == nil || !*otelCfg.MetricsEnabled
+	if tracingOff && metricsOff {
+		slog.Warn("OTLP endpoint is configured but tracing and metrics are both disabled; enabling tracing by default",
+			"endpoint", otelCfg.Endpoint)
+		enabled := true
+		otelCfg.TracingEnabled = &enabled
+	}
+}