diff --git a/python/sglang/srt/disaggregation/mooncake/utils.py b/python/sglang/srt/disaggregation/mooncake/utils.py
index 767276139..63e7bd29d 100644
--- a/python/sglang/srt/disaggregation/mooncake/utils.py
+++ b/python/sglang/srt/disaggregation/mooncake/utils.py
@@ -23,7 +23,7 @@ from sglang.srt.environ import envs
 logger = logging.getLogger(__name__)
 
 # Global constants for custom memory pool types
-SUPPORTED_MOONCAKE_CUSTOM_MEM_POOL_TYPES = ["NVLINK", "BAREX"]
+SUPPORTED_MOONCAKE_CUSTOM_MEM_POOL_TYPES = ["NVLINK", "BAREX", "INTRA_NVLINK"]
 
 
 def init_mooncake_custom_mem_pool(
@@ -55,6 +55,8 @@ def init_mooncake_custom_mem_pool(
                 from mooncake.allocator import BarexAllocator
 
                 allocator = BarexAllocator.get_allocator(device)
+            elif custom_mem_pool_type == "INTRA_NODE_NVLINK":
+                return False, None, None
             else:
                 # This should not happen due to the enable_custom_mem_pool check above
                 raise ValueError(
diff --git a/python/sglang/srt/disaggregation/utils.py b/python/sglang/srt/disaggregation/utils.py
index 816acc871..b3786f881 100644
--- a/python/sglang/srt/disaggregation/utils.py
+++ b/python/sglang/srt/disaggregation/utils.py
@@ -11,6 +11,7 @@ import numpy as np
 import torch
 import torch.distributed as dist
 
+from sglang.srt.environ import envs
 from sglang.srt.utils import is_npu
 
 if TYPE_CHECKING:
@@ -97,6 +98,8 @@ class MetadataBuffers:
         elif self.custom_mem_pool:
             # TODO(shangming): Fix me (use 'cuda') when nvlink_transport of Mooncake is bug-free
             device = "cpu"
+        elif envs.SGLANG_MOONCAKE_CUSTOM_MEM_POOL.get() == "INTRA_NODE_NVLINK":
+            device = "cuda"
         with (
             torch.cuda.use_mem_pool(self.custom_mem_pool)
             if self.custom_mem_pool