diff --git a/django/db/backends/postgresql/compiler.py b/django/db/backends/postgresql/compiler.py
index 3b972b5ba5..c4080ac037 100644
--- a/django/db/backends/postgresql/compiler.py
+++ b/django/db/backends/postgresql/compiler.py
@@ -27,8 +27,8 @@ class InsertUnnest(list):
 
 class SQLInsertCompiler(BaseSQLInsertCompiler):
     def assemble_as_sql(self, fields, value_rows):
-        # Specialize bulk-insertion of literal non-array values through
-        # UNNEST to reduce the time spent planning the query.
+        # Specialize bulk-insertion of literal values through UNNEST to
+        # reduce the time spent planning the query.
         if (
             # The optimization is not worth doing if there is a single
             # row as it will result in the same number of placeholders.
@@ -36,15 +36,18 @@ class SQLInsertCompiler(BaseSQLInsertCompiler):
             # Lack of fields denote the usage of the DEFAULT keyword
             # for the insertion of empty rows.
             or any(field is None for field in fields)
+            # Fields that don't use standard internal types might not be
+            # unnest'able (e.g. array and geometry types are known to be
+            # problematic).
+            or any(
+                field.get_internal_type() not in self.connection.data_types
+                for field in fields
+            )
             # Compilable cannot be combined in an array of literal values.
             or any(any(hasattr(value, "as_sql") for value in row) for row in value_rows)
         ):
             return super().assemble_as_sql(fields, value_rows)
         db_types = [field.db_type(self.connection) for field in fields]
-        # Abort if any of the fields are arrays as UNNEST indiscriminately
-        # flatten them instead of reducing their nesting by one.
-        if any(db_type.endswith("]") for db_type in db_types):
-            return super().assemble_as_sql(fields, value_rows)
         return InsertUnnest(["(%%s)::%s[]" % db_type for db_type in db_types]), [
             list(map(list, zip(*value_rows)))
         ]
diff --git a/docs/releases/5.2.1.txt b/docs/releases/5.2.1.txt
index b17e63ebd1..0f95eda848 100644
--- a/docs/releases/5.2.1.txt
+++ b/docs/releases/5.2.1.txt
@@ -15,3 +15,7 @@ Bugfixes
 
 * Fixed a regression in Django 5.2 that caused unnecessary queries when
   prefetching nullable foreign key relationships (:ticket:`36290`).
+
+* Fixed a regression in Django 5.2 that caused a crash of
+  ``QuerySet.bulk_create()`` with nullable geometry fields on PostGIS
+  (:ticket:`36289`).
diff --git a/tests/gis_tests/geo3d/models.py b/tests/gis_tests/geo3d/models.py
index 456be077f0..a09c599b02 100644
--- a/tests/gis_tests/geo3d/models.py
+++ b/tests/gis_tests/geo3d/models.py
@@ -58,7 +58,7 @@ class SimpleModel(models.Model):
 
 
 class Point2D(SimpleModel):
-    point = models.PointField()
+    point = models.PointField(null=True)
 
 
 class Point3D(SimpleModel):
diff --git a/tests/gis_tests/geo3d/tests.py b/tests/gis_tests/geo3d/tests.py
index b37deabb46..6a9376f2f8 100644
--- a/tests/gis_tests/geo3d/tests.py
+++ b/tests/gis_tests/geo3d/tests.py
@@ -206,6 +206,10 @@ class Geo3DTest(Geo3DLoadingHelper, TestCase):
         lm.save()
         self.assertEqual(3, MultiPoint3D.objects.count())
 
+    def test_bulk_create_point_field(self):
+        objs = Point2D.objects.bulk_create([Point2D(), Point2D()])
+        self.assertEqual(len(objs), 2)
+
     @skipUnlessDBFeature("supports_3d_functions")
     def test_union(self):
         """