diff --git a/torch/csrc/inductor/aoti_runtime/arrayref_tensor.h b/torch/csrc/inductor/aoti_runtime/arrayref_tensor.h
index a864dbf45bf4..436ed3f01f2f 100644
--- a/torch/csrc/inductor/aoti_runtime/arrayref_tensor.h
+++ b/torch/csrc/inductor/aoti_runtime/arrayref_tensor.h
@@ -154,6 +154,10 @@ class MiniArrayRef final {
 
 using MiniIntArrayRef = MiniArrayRef<int64_t>;
 
+static_assert(
+    sizeof(MiniIntArrayRef) == sizeof(void*) + sizeof(size_t),
+    "changing the size of MiniArrayRef breaks ABI compatibility!");
+
 inline bool is_contiguous_strides_for_shape(
     int64_t ndim,
     const int64_t* strides_ptr,
@@ -189,8 +193,7 @@ class ArrayRefTensor {
         sizes_(sizes),
         strides_(strides),
         device_type_(device_type),
-        device_idx_(device_idx),
-        numel_(arr.size()) {
+        device_idx_(device_idx) {
     assert(sizes.size() == strides.size());
     assert(is_contiguous_strides_for_shape(
         sizes.size(), strides.data(), sizes.data()));
@@ -242,7 +245,7 @@ class ArrayRefTensor {
   }
 
   auto numel() const {
-    return numel_;
+    return arrayRef_.size();
   }
 
   void set_arrayref(MiniArrayRef<T> new_arrayref) {
@@ -257,9 +260,17 @@ class ArrayRefTensor {
   MiniArrayRef<const int64_t> strides_;
   int32_t device_type_ = 0;
   int32_t device_idx_ = 0;
-  int32_t numel_ = 0;
+  // We continue to zero-initialize this field in case we repurpose
+  // the space later; having predictable contents can only help.
+  int32_t unusedDoNotRemoveForABICompatibility_ = 0;
 };
 
+static_assert(
+    sizeof(ArrayRefTensor<int>) ==
+        3 * sizeof(MiniIntArrayRef) + 3 * sizeof(int32_t) +
+            (alignof(ArrayRefTensor<int>) > 4 ? sizeof(int32_t) : 0),
+    "changing the size of ArrayRefTensor breaks ABI compatibility!");
+
 inline AtenTensorHandle reinterpret_tensor_wrapper(
     AtenTensorHandle self,
     int64_t ndim,