diff --git a/src/strands/event_loop/event_loop.py b/src/strands/event_loop/event_loop.py
index 71165926..02a56a1c 100644
--- a/src/strands/event_loop/event_loop.py
+++ b/src/strands/event_loop/event_loop.py
@@ -136,6 +136,7 @@ def event_loop_cycle(
     metrics: Metrics
 
     # Retry loop for handling throttling exceptions
+    current_delay = INITIAL_DELAY
     for attempt in range(MAX_ATTEMPTS):
         model_id = model.config.get("model_id") if hasattr(model, "config") else None
         model_invoke_span = tracer.start_model_invoke_span(
@@ -168,7 +169,7 @@ def event_loop_cycle(
 
             # Handle throttling errors with exponential backoff
             should_retry, current_delay = handle_throttling_error(
-                e, attempt, MAX_ATTEMPTS, INITIAL_DELAY, MAX_DELAY, callback_handler, kwargs
+                e, attempt, MAX_ATTEMPTS, current_delay, MAX_DELAY, callback_handler, kwargs
             )
             if should_retry:
                 continue
diff --git a/tests/strands/event_loop/test_event_loop.py b/tests/strands/event_loop/test_event_loop.py
index 8c46e009..734457aa 100644
--- a/tests/strands/event_loop/test_event_loop.py
+++ b/tests/strands/event_loop/test_event_loop.py
@@ -11,6 +11,13 @@
 from strands.types.exceptions import ContextWindowOverflowException, EventLoopException, ModelThrottledException
 
 
+@pytest.fixture
+def mock_time():
+    """Fixture to mock the time module in the error_handler."""
+    with unittest.mock.patch.object(strands.event_loop.error_handler, "time") as mock:
+        yield mock
+
+
 @pytest.fixture
 def model():
     return unittest.mock.Mock()
@@ -157,8 +164,8 @@ def test_event_loop_cycle_text_response(
     assert tru_stop_reason == exp_stop_reason and tru_message == exp_message and tru_request_state == exp_request_state
 
 
-@unittest.mock.patch.object(strands.event_loop.error_handler, "time")
 def test_event_loop_cycle_text_response_throttling(
+    mock_time,
     model,
     model_id,
     system_prompt,
@@ -191,6 +198,53 @@ def test_event_loop_cycle_text_response_throttling(
     exp_request_state = {}
 
     assert tru_stop_reason == exp_stop_reason and tru_message == exp_message and tru_request_state == exp_request_state
+    # Verify that sleep was called once with the initial delay
+    mock_time.sleep.assert_called_once()
+
+
+def test_event_loop_cycle_exponential_backoff(
+    mock_time,
+    model,
+    model_id,
+    system_prompt,
+    messages,
+    tool_config,
+    callback_handler,
+    tool_handler,
+    tool_execution_handler,
+):
+    """Test that the exponential backoff works correctly with multiple retries."""
+    # Set up the model to raise throttling exceptions multiple times before succeeding
+    model.converse.side_effect = [
+        ModelThrottledException("ThrottlingException | ConverseStream"),
+        ModelThrottledException("ThrottlingException | ConverseStream"),
+        ModelThrottledException("ThrottlingException | ConverseStream"),
+        [
+            {"contentBlockDelta": {"delta": {"text": "test text"}}},
+            {"contentBlockStop": {}},
+        ],
+    ]
+
+    tru_stop_reason, tru_message, _, tru_request_state = strands.event_loop.event_loop.event_loop_cycle(
+        model=model,
+        model_id=model_id,
+        system_prompt=system_prompt,
+        messages=messages,
+        tool_config=tool_config,
+        callback_handler=callback_handler,
+        tool_handler=tool_handler,
+        tool_execution_handler=tool_execution_handler,
+    )
+
+    # Verify the final response
+    assert tru_stop_reason == "end_turn"
+    assert tru_message == {"role": "assistant", "content": [{"text": "test text"}]}
+    assert tru_request_state == {}
+
+    # Verify that sleep was called with increasing delays
+    # Initial delay is 4, then 8, then 16
+    assert mock_time.sleep.call_count == 3
+    assert mock_time.sleep.call_args_list == [call(4), call(8), call(16)]
 
 
 def test_event_loop_cycle_text_response_error(