diff --git a/tests/test_config.py b/tests/test_config.py
index 5c01d652a17a..f3ca866ae1e3 100644
--- a/tests/test_config.py
+++ b/tests/test_config.py
@@ -1301,6 +1301,20 @@ def test_scheduler_config_init():
         # InitVar does not become an attribute
         print(SchedulerConfig.default_factory().max_model_len)
 
+    # Regression: max_num_scheduled_tokens must reject non-positive values at
+    # construction time (matches behavior of sibling max_num_batched_tokens
+    # and the existing <= 0 check on the speculative-decoding path; see #44123).
+    sched_kwargs = dict(max_model_len=2048, is_encoder_decoder=False)
+    # None is the sentinel meaning "default to max_num_batched_tokens".
+    SchedulerConfig(max_num_scheduled_tokens=None, **sched_kwargs)
+    # Positive integers are accepted.
+    SchedulerConfig(max_num_scheduled_tokens=4096, **sched_kwargs)
+    # Zero and negative integers are rejected by the field constraint.
+    with pytest.raises(ValidationError):
+        SchedulerConfig(max_num_scheduled_tokens=0, **sched_kwargs)
+    with pytest.raises(ValidationError):
+        SchedulerConfig(max_num_scheduled_tokens=-1, **sched_kwargs)
+
 
 @pytest.mark.parametrize(
     (
diff --git a/vllm/config/scheduler.py b/vllm/config/scheduler.py
index fb6951ea7dd1..7257c23bf3ad 100644
--- a/vllm/config/scheduler.py
+++ b/vllm/config/scheduler.py
@@ -53,7 +53,7 @@ class SchedulerConfig:
     In real usage, this should be set in `EngineArgs.create_engine_config`.
     """
 
-    max_num_scheduled_tokens: int | None = None
+    max_num_scheduled_tokens: int | None = Field(default=None, ge=1)
     """Maximum number of tokens that the scheduler may issue in a single iteration.
     
     This is usually equal to max_num_batched_tokens, but can be smaller in cases