diff --git a/vllm/envs.py b/vllm/envs.py
index c416da755..d77c1e9d9 100755
--- a/vllm/envs.py
+++ b/vllm/envs.py
@@ -288,11 +288,16 @@ def use_aot_compile() -> bool:
     from vllm.model_executor.layers.batch_invariant import (
         vllm_is_batch_invariant,
     )
+    from vllm.platforms import current_platform
     from vllm.utils.torch_utils import is_torch_equal_or_newer
 
     default_value = (
         "1"
-        if is_torch_equal_or_newer("2.10.0.dev") and not disable_compile_cache()
+        if is_torch_equal_or_newer("2.10.0.dev")
+        and not disable_compile_cache()
+        # Disabling AOT_COMPILE for CPU
+        # See: https://github.com/vllm-project/vllm/issues/32033
+        and not current_platform.is_cpu()
         else "0"
     )