diff --git a/backends/vulkan/runtime/api/containers/Tensor.cpp b/backends/vulkan/runtime/api/containers/Tensor.cpp
index 351b920e805..2aa533f7aa8 100644
--- a/backends/vulkan/runtime/api/containers/Tensor.cpp
+++ b/backends/vulkan/runtime/api/containers/Tensor.cpp
@@ -564,14 +564,19 @@ size_t calculate_max_ubo_nbytes(
     const size_t min_nbytes_per_ubo,
     const utils::StorageType storage_type) {
   size_t ivec4_ubo_nbytes = utils::align_up(size_t(16), min_nbytes_per_ubo);
-  size_t uvec3_ubo_nbytes = utils::align_up(size_t(12), min_nbytes_per_ubo);
+  // TextureLimits has alignas(16) so sizeof(TextureLimits) == 16, not 12.
+  // Use 16 to match the actual sizeof used by metadata_ubo_impl().
+  size_t uvec3_ubo_nbytes = utils::align_up(size_t(16), min_nbytes_per_ubo);
   size_t int32_ubo_nbytes = utils::align_up(size_t(4), min_nbytes_per_ubo);
   if (storage_type == utils::kBuffer) {
     // sizes, strides, dim order, numel
     return 3 * ivec4_ubo_nbytes + int32_ubo_nbytes;
   }
-  // sizes, logical limits
-  return ivec4_ubo_nbytes + uvec3_ubo_nbytes;
+  // sizes, strides, dim_order, numel, logical_limits
+  // Ops like Linear and MatMul unconditionally request strides/numel UBOs on
+  // all tensors regardless of storage type, so texture tensors need the same
+  // metadata budget as buffer tensors plus logical_limits.
+  return 3 * ivec4_ubo_nbytes + int32_ubo_nbytes + uvec3_ubo_nbytes;
 }
 
 //
@@ -1161,9 +1166,10 @@ bool vTensor::is_contiguous() const {
 }
 
 size_t vTensor::get_max_ubo_nbytes(const size_t nbytes_per_ubo) const {
-  // For texture backed tensors, the metadata fields needed are:
-  // sizes, logical limits
-  size_t max_metadata_field_count = 2u;
+  // Ops like Linear and MatMul unconditionally request strides/numel UBOs on
+  // all tensors regardless of storage type, so texture tensors need the same
+  // metadata budget as buffer tensors plus logical_limits (5 fields total).
+  size_t max_metadata_field_count = 5u;
   if (storage_type() == utils::kBuffer) {
     // sizes, strides, dim order, numel
     max_metadata_field_count = 4u;
diff --git a/backends/vulkan/test/vulkan_compute_api_test.cpp b/backends/vulkan/test/vulkan_compute_api_test.cpp
index 20c9ac2b14f..261d3f72d01 100644
--- a/backends/vulkan/test/vulkan_compute_api_test.cpp
+++ b/backends/vulkan/test/vulkan_compute_api_test.cpp
@@ -920,6 +920,56 @@ TEST_F(VulkanComputeAPITest, tensor_layout_metadata_test_against_golden) {
   }
 }
 
+// Test that texture-backed tensors can serve all metadata UBO requests
+// (sizes, strides, dim_order, numel, logical_limits) without exceeding the
+// pre-allocated UBO budget. This is a regression test for an issue where
+// calculate_max_ubo_nbytes() only allocated 2 fields for texture tensors
+// (sizes + logical_limits), but operators like Linear/MatMul unconditionally
+// request strides_ubo() and numel_ubo() on all tensors regardless of storage
+// type, causing an assertion failure:
+//   "Uniform data allocation has exceeded Tensor uniform buffer size"
+TEST_F(VulkanComputeAPITest, texture_tensor_ubo_metadata_budget_test) {
+  // Create a texture-backed tensor (the default for most Vulkan ops)
+  std::vector<int64_t> sizes = {4, 8, 8};
+  vTensor texture_tensor = vTensor(
+      context(),
+      sizes,
+      vkapi::kFloat,
+      utils::StorageType::TEXTURE_3D,
+      utils::GPUMemoryLayout::TENSOR_CHANNELS_PACKED);
+
+  // These two UBOs are within the original 2-field texture budget:
+  // Field 1: sizes (ivec4)
+  EXPECT_NO_THROW(texture_tensor.sizes_ubo());
+  // Field 2: logical_limits (uvec3)
+  EXPECT_NO_THROW(texture_tensor.logical_limits_ubo());
+
+  // These UBOs exceed the original 2-field texture budget but are
+  // unconditionally requested by ops like Linear, MatMul, etc.
+  // Without the fix, these will trigger:
+  //   VK_CHECK_COND((uniforms_size_ + ubo_nbytes) <= max_ubo_nbytes_)
+  // Field 3: strides (ivec4) - FAILS without fix
+  EXPECT_NO_THROW(texture_tensor.strides_ubo());
+  // Field 4: numel (int32) - FAILS without fix
+  EXPECT_NO_THROW(texture_tensor.numel_ubo());
+  // Field 5: dim_order (ivec4) - FAILS without fix
+  EXPECT_NO_THROW(texture_tensor.dim_order_ubo());
+
+  // Also verify a buffer-backed tensor still works (should always have had
+  // enough budget for all 4+ fields)
+  vTensor buffer_tensor = vTensor(
+      context(),
+      sizes,
+      vkapi::kFloat,
+      utils::StorageType::BUFFER,
+      utils::GPUMemoryLayout::TENSOR_CHANNELS_PACKED);
+
+  EXPECT_NO_THROW(buffer_tensor.sizes_ubo());
+  EXPECT_NO_THROW(buffer_tensor.strides_ubo());
+  EXPECT_NO_THROW(buffer_tensor.dim_order_ubo());
+  EXPECT_NO_THROW(buffer_tensor.numel_ubo());
+}
+
 TEST_F(VulkanComputeAPITest, virtual_transpose_test) {
   std::vector<int64_t> sizes = {7, 9, 11, 13};
   // (dim0, dim1), new_sizes, new_dim_order, new_axis_map, new_packed_dim_idx