diff --git a/ibis/backends/duckdb/tests/snapshots/test_client/test_default_backend/out.sql b/ibis/backends/duckdb/tests/snapshots/test_client/test_default_backend/out.sql
deleted file mode 100644
index 06687fb622f1..000000000000
--- a/ibis/backends/duckdb/tests/snapshots/test_client/test_default_backend/out.sql
+++ /dev/null
@@ -1,3 +0,0 @@
-SELECT
-  SUM(t0.a) AS "Sum(a)"
-FROM ibis_pandas_memtable_fw3sdos5brerlgtmbkopvh334m AS t0
\ No newline at end of file
diff --git a/ibis/backends/duckdb/tests/test_client.py b/ibis/backends/duckdb/tests/test_client.py
index 061ff3809aa2..f2cfa5619568 100644
--- a/ibis/backends/duckdb/tests/test_client.py
+++ b/ibis/backends/duckdb/tests/test_client.py
@@ -1,6 +1,8 @@
 from __future__ import annotations
 
 import os
+import subprocess
+import sys
 
 import duckdb
 import pandas as pd
@@ -211,18 +213,27 @@ def test_insert_preserves_column_case(con):
     assert t1.count().execute() == 8
 
 
-def test_default_backend(snapshot):
-    df = pd.DataFrame({"a": [1, 2, 3]})
-    t = ibis.memtable(df)
-    expr = t.a.sum()
+def test_default_backend():
+    # use subprocess to avoid mutating state across tests
+    script = """\
+import pandas as pd
+
+import ibis
 
-    # run this twice to ensure that we hit the optimizations in
-    # `_default_backend`
-    for _ in range(2):
-        assert expr.execute() == df.a.sum()
+df = pd.DataFrame({"a": [1, 2, 3]})
 
-    sql = ibis.to_sql(expr)
-    snapshot.assert_match(sql, "out.sql")
+t = ibis.memtable(df)
+
+expr = t.a.sum()
+
+# run twice to ensure that we hit the optimizations in
+# `_default_backend`
+for _ in range(2):
+    assert expr.execute() == df.a.sum()"""
+
+    assert ibis.options.default_backend is None
+    subprocess.run([sys.executable, "-c", script], check=True)
+    assert ibis.options.default_backend is None
 
 
 @pytest.mark.parametrize(