diff --git a/cmake/deps.txt b/cmake/deps.txt
index e1870bf2df0cf..f8e5fb7f8ede0 100644
--- a/cmake/deps.txt
+++ b/cmake/deps.txt
@@ -47,7 +47,7 @@ protoc_mac_universal;https://github.com/protocolbuffers/protobuf/releases/downlo
 psimd;https://github.com/Maratyszcza/psimd/archive/072586a71b55b7f8c584153d223e95687148a900.zip;1f5454b01f06f9656b77e4a5e2e31d7422487013
 pthreadpool;https://github.com/google/pthreadpool/archive/dcc9f28589066af0dbd4555579281230abbf74dd.zip;533a77943203ef15ca608bcd9dbe2c94da7451d2
 pybind11;https://github.com/pybind/pybind11/archive/refs/tags/v2.13.6.zip;f780292da9db273c8ef06ccf5fd4b623624143e9
-pytorch_cpuinfo;https://github.com/pytorch/cpuinfo/archive/877328f188a3c7d1fa855871a278eb48d530c4c0.zip;9152d4bf6b8bde9f19b116de3bd8a745097ed9df
+pytorch_cpuinfo;https://github.com/pytorch/cpuinfo/archive/403d652dca4c1046e8145950b1c0997a9f748b57.zip;30b2a07fe4bae8574f89176e56274cacdd6d135b
 re2;https://github.com/google/re2/archive/refs/tags/2024-07-02.zip;646e1728269cde7fcef990bf4a8e87b047882e88
 safeint;https://github.com/dcleblanc/SafeInt/archive/refs/tags/3.0.28.zip;23f252040ff6cb9f1fd18575b32fa8fb5928daac
 tensorboard;https://github.com/tensorflow/tensorboard/archive/373eb09e4c5d2b3cc2493f0949dc4be6b6a45e81.zip;67b833913605a4f3f499894ab11528a702c2b381
diff --git a/cmake/vcpkg-ports/cpuinfo/portfile.cmake b/cmake/vcpkg-ports/cpuinfo/portfile.cmake
index 80192840ee9b0..5aa5aebe50f3d 100644
--- a/cmake/vcpkg-ports/cpuinfo/portfile.cmake
+++ b/cmake/vcpkg-ports/cpuinfo/portfile.cmake
@@ -6,8 +6,8 @@ endif()
 vcpkg_from_github(
     OUT_SOURCE_PATH SOURCE_PATH
     REPO pytorch/cpuinfo
-    REF 877328f188a3c7d1fa855871a278eb48d530c4c0
-    SHA512 b6d5a9ce9996eee3b2f09f39115f7ae178fe4d4814cc35b049a59d04a82228e268aa52d073c307ccb56a427428622940e1c77f004c99851dfca0d3a5d803658b
+    REF 403d652dca4c1046e8145950b1c0997a9f748b57
+    SHA512 f7cd6dc44bd1120af610cae1337ed4c0f557ba78d2de9c73fed350fa3dfe9512643a1619ae55f5a540c6316a87d641856cca27297bb8766e48f39b7b7a59da1f
     HEAD_REF master
     PATCHES
         patch_cpuinfo_h_for_arm64ec.patch
diff --git a/cmake/vcpkg-ports/cpuinfo/vcpkg.json b/cmake/vcpkg-ports/cpuinfo/vcpkg.json
index f1ccda72679b1..76486eceecf12 100644
--- a/cmake/vcpkg-ports/cpuinfo/vcpkg.json
+++ b/cmake/vcpkg-ports/cpuinfo/vcpkg.json
@@ -1,7 +1,7 @@
 {
   "name": "cpuinfo",
-  "version-date": "2025-10-23",
-  "port-version": 4,
+  "version-date": "2025-11-18",
+  "port-version": 5,
   "description": "CPU INFOrmation library (x86/x86-64/ARM/ARM64, Linux/Windows/Android/macOS/iOS)",
   "homepage": "https://github.com/pytorch/cpuinfo",
   "license": "BSD-2-Clause",
diff --git a/onnxruntime/contrib_ops/cpu/quantization/dynamic_quantize_matmul.cc b/onnxruntime/contrib_ops/cpu/quantization/dynamic_quantize_matmul.cc
index ceb498372a6fc..2bba0adcd987c 100644
--- a/onnxruntime/contrib_ops/cpu/quantization/dynamic_quantize_matmul.cc
+++ b/onnxruntime/contrib_ops/cpu/quantization/dynamic_quantize_matmul.cc
@@ -1,7 +1,6 @@
 // Copyright (c) Microsoft Corporation. All rights reserved.
 // Licensed under the MIT License.
 
-#include "core/common/cpuid_info.h"  // for CPUIDInfo::GetCPUIDInfo().HasArm_SME2()
 #include "core/common/narrow.h"
 #include "core/common/safeint.h"
 #include "core/mlas/inc/mlas.h"
@@ -213,9 +212,7 @@ class DynamicQuantizeMatMul final : public MatMulIntegerToFloatBase {
         }
       }
 
-      // Currently, MlasDynamicQGemmBatch() and associated functions require SME2 or else they are no-ops.
-      // We check that here too before attempting to use them.
-      if (!CPUIDInfo::GetCPUIDInfo().HasArm_SME2()) {
+      if (!MlasIsDynamicQGemmAvailable()) {
         can_use_dynamic_quant_mlas_ = false;
       }
 
diff --git a/onnxruntime/core/common/cpuid_info.cc b/onnxruntime/core/common/cpuid_info.cc
index ab8ab0b326292..f5d3b93b395e2 100644
--- a/onnxruntime/core/common/cpuid_info.cc
+++ b/onnxruntime/core/common/cpuid_info.cc
@@ -237,9 +237,9 @@ void CPUIDInfo::ArmLinuxInit() {
 #elif defined(_WIN32)  // ^ defined(__linux__)
 
 void CPUIDInfo::ArmWindowsInit() {
-  // Read MIDR and ID_AA64ISAR1_EL1 register values from Windows registry
+  // Read MIDR register values from Windows registry
   // There should be one per CPU
-  std::vector<uint64_t> midr_values{}, id_aa64isar1_el1_values{};
+  std::vector<uint64_t> midr_values{};
 
   // TODO!! Don't support multiple processor group yet!!
   constexpr int MAX_CORES = 64;
@@ -272,17 +272,7 @@ void CPUIDInfo::ArmWindowsInit() {
       break;
     }
 
-    uint64_t id_aa64isar1_el1_value;
-    data_size = sizeof(id_aa64isar1_el1_value);
-
-    // CP 4031 corresponds to ID_AA64ISAR1_EL1 register
-    if (::RegGetValueA(HKEY_LOCAL_MACHINE, processor_subkey, "CP 4031", RRF_RT_REG_QWORD,
-                       nullptr, &id_aa64isar1_el1_value, &data_size) != ERROR_SUCCESS) {
-      break;
-    }
-
     midr_values.push_back(midr_value);
-    id_aa64isar1_el1_values.push_back(id_aa64isar1_el1_value);
   }
 
   // process midr_values
@@ -308,22 +298,31 @@ void CPUIDInfo::ArmWindowsInit() {
     }
   }
 
-  has_arm_neon_i8mm_ = std::all_of(
-      id_aa64isar1_el1_values.begin(), id_aa64isar1_el1_values.end(),
-      [](uint64_t id_aa64isar1_el1_value) {
-        // I8MM, bits [55:52]
-        return ((id_aa64isar1_el1_value >> 52) & 0xF) != 0;
-      });
-
-  has_arm_neon_dot_ = (IsProcessorFeaturePresent(PF_ARM_V82_DP_INSTRUCTIONS_AVAILABLE) != 0);
-
 #if defined(CPUINFO_SUPPORTED)
   if (pytorch_cpuinfo_init_) {
+    has_arm_neon_dot_ = cpuinfo_has_arm_neon_dot();
     has_fp16_ = cpuinfo_has_arm_neon_fp16_arith();
-    // cpuinfo_has_arm_i8mm() doesn't work on Windows yet. See https://github.com/pytorch/cpuinfo/issues/279.
-    // has_arm_neon_i8mm_ = cpuinfo_has_arm_i8mm();
-    has_arm_sve_i8mm_ = cpuinfo_has_arm_sve() && has_arm_neon_i8mm_;
+
+    // Note:
+    // cpuinfo is using IsProcessorFeaturePresent(PF_ARM_V82_FP16_INSTRUCTIONS_AVAILABLE):
+    // https://github.com/pytorch/cpuinfo/blob/403d652dca4c1046e8145950b1c0997a9f748b57/src/arm/windows/init.c#L224-L225
+    // However, on some systems (notably, a Windows ARM64 CI build agent), cpuinfo_has_arm_neon_fp16_arith() started to
+    // return false in the newer cpuinfo version that uses IsProcessorFeaturePresent(). Perhaps the newer
+    // PF_ARM_V82_FP16_INSTRUCTIONS_AVAILABLE constant is not supported yet in the Windows version on those systems.
+    // To avoid regressing in fp16 instructions detection, we fall back to what cpuinfo used to do, i.e., use the
+    // detection of dot product instructions:
+    // https://github.com/pytorch/cpuinfo/blob/877328f188a3c7d1fa855871a278eb48d530c4c0/src/arm/windows/init.c#L206-L209
+    // This workaround can be removed when cpuinfo_has_arm_neon_fp16_arith() works correctly on all the Windows
+    // versions that we want to support.
+    if (!has_fp16_) {
+      has_fp16_ = has_arm_neon_dot_;
+    }
+
+    has_arm_neon_i8mm_ = cpuinfo_has_arm_i8mm();
+    has_arm_sve_i8mm_ = cpuinfo_has_arm_sve() && cpuinfo_has_arm_i8mm();
     has_arm_neon_bf16_ = cpuinfo_has_arm_neon_bf16();
+    has_arm_sme_ = cpuinfo_has_arm_sme();
+    has_arm_sme2_ = cpuinfo_has_arm_sme2();
   }
 #endif  // defined(CPUINFO_SUPPORTED)
 }
@@ -397,4 +396,4 @@ CPUIDInfo::CPUIDInfo() {
 #endif
 #endif  // defined(CPUIDINFO_ARCH_ARM)
 }
-}  // namespace onnxruntime
\ No newline at end of file
+}  // namespace onnxruntime
diff --git a/onnxruntime/core/common/cpuid_info.h b/onnxruntime/core/common/cpuid_info.h
index 9c40627b5cd1b..ca9315c7ef95d 100644
--- a/onnxruntime/core/common/cpuid_info.h
+++ b/onnxruntime/core/common/cpuid_info.h
@@ -171,4 +171,4 @@ class CPUIDInfo {
   uint32_t vendor_id_;
 };
 
-}  // namespace onnxruntime
\ No newline at end of file
+}  // namespace onnxruntime
diff --git a/onnxruntime/core/mlas/inc/mlas.h b/onnxruntime/core/mlas/inc/mlas.h
index 9d98a15d8457a..248c6d74e6cbd 100644
--- a/onnxruntime/core/mlas/inc/mlas.h
+++ b/onnxruntime/core/mlas/inc/mlas.h
@@ -634,6 +634,7 @@ MlasGemm(
 {
     MlasGemmBatch(Shape, &DataParams, 1, ThreadPool);
 }
+
 /**
  * @brief Parameters that define the shape of a dynamically quantized GEMM operation.
  *
@@ -646,6 +647,7 @@ struct MLAS_GEMM_DYN_QUANT_SHAPE_PARAMS {
     size_t N = 0;                  /**< Column size of matrix B */
     size_t K = 0;                  /**< Column size of matrix A and Row size of matrix B */
 };
+
 /**
  * @brief Parameters that define the data buffers and layout for a dynamic quant GEMM.
  *
@@ -680,6 +682,14 @@ MlasDynamicQGemm (
     MlasDynamicQGemmBatch(Shape, DataParams, 1, ThreadPool);
 }
 
+/**
+ * @brief Determines whether a dynamic quantized GEMM implementation is available on the current platform.
+ *
+ * MlasDynamicQGemm() and MlasDynamicQGemmBatch() should only be called if this function returns true.
+ */
+bool
+MLASCALL
+MlasIsDynamicQGemmAvailable();
 
 //
 // Symmetric QGEMM has limited buffer overrun.
diff --git a/onnxruntime/core/mlas/lib/qgemm.cpp b/onnxruntime/core/mlas/lib/qgemm.cpp
index a1c2e467188f7..b3eb5893ba7e7 100644
--- a/onnxruntime/core/mlas/lib/qgemm.cpp
+++ b/onnxruntime/core/mlas/lib/qgemm.cpp
@@ -201,6 +201,17 @@ MlasGemmBatch(
     });
 }
 
+bool
+MLASCALL
+MlasIsDynamicQGemmAvailable()
+{
+#if defined(USE_KLEIDIAI) && !defined(_MSC_VER)
+    return ArmKleidiAI::UseSME2;
+#else
+    return false;
+#endif
+}
+
 void
 MLASCALL
 MlasDynamicQGemmBatch (
@@ -211,7 +222,7 @@ MlasDynamicQGemmBatch (
 ) {
 #if defined(USE_KLEIDIAI) && !defined(_MSC_VER)
     //No fallback and putting in guards. This implementation is SME2 specific.
-    if(ArmKleidiAI::UseSME2){
+    if (ArmKleidiAI::UseSME2) {
         ArmKleidiAI::MlasDynamicQGemmBatch(Shape, DataParams, BatchN, ThreadPool);
     }
 #endif
@@ -336,7 +347,7 @@ MlasDynamicQgemmPackBSize(
 #if defined(USE_KLEIDIAI) && !defined(_MSC_VER)
     //No fallback available
     //TODO: Insert Override
-    if(MLAS_CPUIDINFO::GetCPUIDInfo().HasArm_SME()){//Still require this since no override
+    if (ArmKleidiAI::UseSME2) {  //Still require this since no override
         bytes = ArmKleidiAI::MlasDynamicQgemmPackBSize(N, K);
     }
 #endif
@@ -407,7 +418,7 @@ Return Value:
         ~(BufferAlignment - 1);
     // If this gemm B argument is used in a dynamically quantization gemm operation we can optimize for
     // this use case. Concat both packed representations for later decision. This allows for cases later
-    // where we still have the prepack at the cost of some memory otherwise we can use the qgemm quantization 
+    // where we still have the prepack at the cost of some memory otherwise we can use the qgemm quantization
     // for better performance
     return AlignedBytesRequired + MlasDynamicQgemmPackBSize(N, K);
 }
@@ -425,7 +436,7 @@ MlasDynamicQgemmPackB(
 {
 #if defined(USE_KLEIDIAI) && !defined(_MSC_VER)
     //No fallback
-    if(MLAS_CPUIDINFO::GetCPUIDInfo().HasArm_SME()){//Still require this since no override
+    if (ArmKleidiAI::UseSME2) {  //Still require this since no override
         ArmKleidiAI::MlasDynamicQgemmPackB(N, K, B, Scales, Bias, PackedB);
     }
 #endif
diff --git a/onnxruntime/test/mlas/unittest/test_dynamic_qgemm.cpp b/onnxruntime/test/mlas/unittest/test_dynamic_qgemm.cpp
index bebff37ad8460..8a9de636c835f 100644
--- a/onnxruntime/test/mlas/unittest/test_dynamic_qgemm.cpp
+++ b/onnxruntime/test/mlas/unittest/test_dynamic_qgemm.cpp
@@ -7,8 +7,8 @@
 // Currently this test only applies to KleidiAI Guard against it running in any other situation
 #if defined(USE_KLEIDIAI) && !defined(_MSC_VER)
 
+#include "mlas.h"
 #include "test_util.h"
-#include "core/mlas/lib/mlasi.h"  // for MLAS_CPUIDINFO
 
 class MlasDynamicQgemmTest {
  private:
@@ -20,9 +20,8 @@ class MlasDynamicQgemmTest {
 
  public:
   void Test(size_t M, size_t N, size_t K, size_t BatchSize) {
-    // Currently, MlasDynamicQGemmBatch() and associated functions require SME2 or else they are no-ops.
-    if (!MLAS_CPUIDINFO::GetCPUIDInfo().HasArm_SME2()) {
-      GTEST_SKIP() << "MlasDynamicQGemmBatch() requires ARM64 SME2 but it was not detected. Skipping test.";
+    if (!MlasIsDynamicQGemmAvailable()) {
+      GTEST_SKIP() << "MlasDynamicQGemmBatch() is not supported on this platform. Skipping test.";
     }
 
     // Setup buffers for holding various data