diff --git a/python/sglang/srt/layers/attention/dsv4/indexer.py b/python/sglang/srt/layers/attention/dsv4/indexer.py
index 93f026a5c..0f92d65d4 100644
--- a/python/sglang/srt/layers/attention/dsv4/indexer.py
+++ b/python/sglang/srt/layers/attention/dsv4/indexer.py
@@ -32,7 +32,7 @@ from sglang.srt.model_executor.runner_backend_utils.tc_piecewise_cuda_graph impo
 )
 from sglang.srt.runtime_context import get_parallel
 from sglang.srt.state_capturer.indexer_topk import get_global_indexer_capturer
-from sglang.srt.utils import add_prefix, is_cuda, is_hip
+from sglang.srt.utils import add_prefix, is_cuda, is_hip, is_xpu
 from sglang.srt.utils.common import is_sm120_supported
 
 if TYPE_CHECKING:
@@ -668,6 +668,12 @@ class C4IndexerBackendMixin:
                 fn = fp8_paged_mqa_logits_torch_sm120
             else:
                 fn = fp8_paged_mqa_logits_torch
+        elif is_xpu():
+            from sgl_kernel import fp8_paged_mqa_logits_triton
+
+            # TODO: switch from triton to SYCL when OOM is resolved
+
+            fn = fp8_paged_mqa_logits_triton
         else:
             from deep_gemm import fp8_paged_mqa_logits as fn
 
diff --git a/python/sglang/srt/layers/attention/dsv4/metadata.py b/python/sglang/srt/layers/attention/dsv4/metadata.py
index d26d1ccf5..c5335b0e0 100644
--- a/python/sglang/srt/layers/attention/dsv4/metadata.py
+++ b/python/sglang/srt/layers/attention/dsv4/metadata.py
@@ -7,7 +7,7 @@ from typing import TYPE_CHECKING, Any, List, Optional
 import torch
 
 from sglang.srt.environ import envs
-from sglang.srt.utils import is_hip
+from sglang.srt.utils import is_hip, is_xpu
 
 if TYPE_CHECKING:
     pass
@@ -122,6 +122,7 @@ class PagedIndexerMetadata:
     def __post_init__(self):
         if (
             envs.SGLANG_FP8_PAGED_MQA_LOGITS_TORCH.get()
+            or is_xpu()
             or envs.SGLANG_OPT_USE_AITER_INDEXER.get()
         ):
             self.deep_gemm_metadata = None