diff --git a/src/diffusers/pipelines/pipeline_utils.py b/src/diffusers/pipelines/pipeline_utils.py
index a683973df5d7..24fe0eabfa6f 100644
--- a/src/diffusers/pipelines/pipeline_utils.py
+++ b/src/diffusers/pipelines/pipeline_utils.py
@@ -601,12 +601,20 @@ def module_is_offloaded(module):
     def device(self) -> torch.device:
         r"""
         Returns:
-            `torch.device`: The torch device on which the pipeline is located.
+            `torch.device`: The torch device on which the pipeline is located. When components are split across devices
+            (for example, text encoders on CPU while the denoising backbone runs on an accelerator), the accelerator
+            device is returned.
         """
         module_names, _ = self._get_signature_keys(self)
         modules = [getattr(self, n, None) for n in module_names]
         modules = [m for m in modules if isinstance(m, torch.nn.Module)]
 
+        # Prefer a non-CPU, non-meta component so a split pipeline reports the accelerator it computes on,
+        # rather than whichever component happens to sort first.
+        for module in modules:
+            if module.device.type not in ("cpu", "meta"):
+                return module.device
+
         for module in modules:
             return module.device
 
diff --git a/tests/pipelines/test_pipelines.py b/tests/pipelines/test_pipelines.py
index d975ef8bd2d8..5feef16871ee 100644
--- a/tests/pipelines/test_pipelines.py
+++ b/tests/pipelines/test_pipelines.py
@@ -1944,6 +1944,38 @@ def test_pipe_to(self):
         assert sd1.device.type == device_type
         assert sd2.device.type == device_type
 
+    @require_torch_accelerator
+    def test_pipe_device_split_across_devices(self):
+        unet = self.dummy_cond_unet()
+        scheduler = PNDMScheduler(skip_prk_steps=True)
+        vae = self.dummy_vae
+        bert = self.dummy_text_encoder
+        tokenizer = CLIPTokenizer.from_pretrained("hf-internal-testing/tiny-random-clip")
+
+        sd = StableDiffusionPipeline(
+            unet=unet,
+            scheduler=scheduler,
+            vae=vae,
+            text_encoder=bert,
+            tokenizer=tokenizer,
+            safety_checker=None,
+            feature_extractor=self.dummy_extractor,
+        )
+
+        device_type = torch.device(torch_device).type
+
+        # Text encoder stays on CPU while the denoising backbone runs on the accelerator. `text_encoder` sorts
+        # before `unet`/`vae`, so a first-component rule would report `cpu` here.
+        sd.unet.to(torch_device)
+        sd.vae.to(torch_device)
+
+        assert sd.text_encoder.device.type == "cpu"
+        assert sd.device.type == device_type
+
+        # With every component on CPU there is no accelerator to prefer.
+        sd.to("cpu")
+        assert sd.device.type == "cpu"
+
     def test_pipe_same_device_id_offload(self):
         unet = self.dummy_cond_unet()
         scheduler = PNDMScheduler(skip_prk_steps=True)