diff --git a/transformer_engine/pytorch/csrc/extensions/cast.cpp b/transformer_engine/pytorch/csrc/extensions/cast.cpp
index acbe4753b..f3bf2a804 100644
--- a/transformer_engine/pytorch/csrc/extensions/cast.cpp
+++ b/transformer_engine/pytorch/csrc/extensions/cast.cpp
@@ -132,6 +132,14 @@ void multi_tensor_quantize_impl(const std::vector<TensorWrapper> &input_list,
     }
   }
 
+  // Check if all quantizers are MXFP8 (for multi-stream quantize)
+  bool with_mxfp8_multi_stream = false;
+  if (!with_fused_kernel) {
+    with_mxfp8_multi_stream = std::all_of(
+        quantizer_py_list.begin(), quantizer_py_list.end(),
+        [](const py::handle &q) { return detail::IsMXFP8Quantizers(q.ptr()); });
+  }
+
   // Launch TE kernel
   if (with_fused_kernel) {
     // Fused kernel for multi-tensor quantize
@@ -145,6 +153,20 @@ void multi_tensor_quantize_impl(const std::vector<TensorWrapper> &input_list,
       nvte_multi_cast_transpose(nvte_tensor_input_list.size(), nvte_tensor_input_list.data(),
                                 nvte_tensor_output_list.data(), at::cuda::getCurrentCUDAStream());
     });
+  } else if (with_mxfp8_multi_stream) {
+    // Multi-stream quantize for MXFP8
+    std::vector<NVTETensor> nvte_tensor_input_list;
+    std::vector<NVTETensor> nvte_tensor_output_list;
+    for (size_t i = 0; i < num_tensors; ++i) {
+      nvte_tensor_input_list.push_back(input_list[i].data());
+      nvte_tensor_output_list.push_back(output_list[i].data());
+    }
+    QuantizationConfigWrapper quant_config;
+    NVTE_SCOPED_GIL_RELEASE({
+      nvte_multi_tensor_quantize(nvte_tensor_input_list.data(), nvte_tensor_output_list.data(),
+                                 quant_config, num_tensors,
+                                 at::cuda::getCurrentCUDAStream());
+    });
   } else {
     // Quantize kernels individually
     for (size_t i = 0; i < num_tensors; ++i) {