diff --git a/vllm/distributed/kv_transfer/kv_connector/utils.py b/vllm/distributed/kv_transfer/kv_connector/utils.py
index 4f1ea1a0240c4..914ab91b1563c 100644
--- a/vllm/distributed/kv_transfer/kv_connector/utils.py
+++ b/vllm/distributed/kv_transfer/kv_connector/utils.py
@@ -11,7 +11,6 @@ from typing import TYPE_CHECKING, Literal
 import torch
 
 from vllm.attention.backends.abstract import AttentionBackend
-from vllm.attention.backends.registry import AttentionBackendEnum
 from vllm.config import get_current_vllm_config
 from vllm.distributed.kv_transfer.kv_connector.factory import KVConnectorFactory
 from vllm.logger import init_logger
@@ -251,9 +250,6 @@ class TpKVTopology:
             len(kv_cache_shape) == 5 and kv_cache_shape[0] == 1
         )
 
-        attn_backend = AttentionBackendEnum[self.attn_backend.get_name()]
-        self._use_pallas = attn_backend == AttentionBackendEnum.PALLAS
-
     @property
     def is_kv_layout_blocks_first(self) -> bool:
         return self._is_kv_layout_blocks_first
@@ -261,7 +257,7 @@ class TpKVTopology:
     @property
     def split_k_and_v(self) -> bool:
         # Whether to register regions for K and V separately (when present).
-        return not (self.is_mla or self._use_pallas or self.is_kv_layout_blocks_first)
+        return not (self.is_mla or self.is_kv_layout_blocks_first)
 
     @property
     def tp_size(self) -> int:
diff --git a/vllm/distributed/kv_transfer/kv_connector/v1/mooncake_connector.py b/vllm/distributed/kv_transfer/kv_connector/v1/mooncake_connector.py
index 9a15d3fa6ed09..38ce02a2fef76 100644
--- a/vllm/distributed/kv_transfer/kv_connector/v1/mooncake_connector.py
+++ b/vllm/distributed/kv_transfer/kv_connector/v1/mooncake_connector.py
@@ -499,7 +499,6 @@ class MooncakeConnectorWorker:
             total_num_kv_heads=self.model_config.get_total_num_kv_heads(),
             attn_backend=backend,
         )
-        self._use_pallas = self.kv_topo._use_pallas
 
         self.zmq_ctx = zmq.Context()
         self.async_zmq_ctx = zmq.asyncio.Context()
diff --git a/vllm/distributed/kv_transfer/kv_connector/v1/nixl_connector.py b/vllm/distributed/kv_transfer/kv_connector/v1/nixl_connector.py
index 757ca41e9844b..0f33cde7d3221 100644
--- a/vllm/distributed/kv_transfer/kv_connector/v1/nixl_connector.py
+++ b/vllm/distributed/kv_transfer/kv_connector/v1/nixl_connector.py
@@ -983,7 +983,6 @@ class NixlConnectorWorker:
             total_num_kv_heads=self.model_config.get_total_num_kv_heads(),
             attn_backend=backend,
         )
-        self._use_pallas = self.kv_topo._use_pallas
         self._physical_blocks_per_logical_kv_block = 1
 
     def _nixl_handshake(
@@ -1641,9 +1640,6 @@ class NixlConnectorWorker:
         # Num kv_heads > tp_size and P TP > D TP case, not supported
         assert not (tp_ratio < 0 and self.kv_topo.is_kv_replicated(remote_engine_id))
 
-        assert not self._use_pallas or tp_ratio == 1, (
-            "TPU (pallas_v1) DOES NOT support heterogeneous TP yet."
-        )
         kv_cache_layout = (
             self.kv_cache_layout
             if not self.use_host_buffer
@@ -1814,9 +1810,7 @@ class NixlConnectorWorker:
 
         if len(self.device_kv_caches) == 0:
             return
-        split_k_and_v = not (
-            self.use_mla or self._use_pallas or self.kv_topo.is_kv_layout_blocks_first
-        )
+        split_k_and_v = not (self.use_mla or self.kv_topo.is_kv_layout_blocks_first)
         sample_cache = list(self.device_kv_caches.values())[0][0]
         for block_size_ratio, block_ids_list in block_ids_per_ratio.items():
             assert block_size_ratio > 1, "Only nP < nD supported currently."