diff --git a/include/libyuv/row.h b/include/libyuv/row.h
index 77e646c54..c6efa5336 100644
--- a/include/libyuv/row.h
+++ b/include/libyuv/row.h
@@ -566,6 +566,7 @@ extern "C" {
 #if !defined(LIBYUV_DISABLE_SVE) && defined(__aarch64__)
 #define HAS_I444TOARGBROW_SVE2
 #define HAS_I422TOARGBROW_SVE2
+#define HAS_I444ALPHATOARGBROW_SVE2
 #endif
 
 // The following are available on AArch64 platforms:
@@ -1067,6 +1068,13 @@ void I444AlphaToARGBRow_NEON(const uint8_t* src_y,
                              uint8_t* dst_argb,
                              const struct YuvConstants* yuvconstants,
                              int width);
+void I444AlphaToARGBRow_SVE2(const uint8_t* src_y,
+                             const uint8_t* src_u,
+                             const uint8_t* src_v,
+                             const uint8_t* src_a,
+                             uint8_t* dst_argb,
+                             const struct YuvConstants* yuvconstants,
+                             int width);
 void I422AlphaToARGBRow_NEON(const uint8_t* src_y,
                              const uint8_t* src_u,
                              const uint8_t* src_v,
diff --git a/source/convert_argb.cc b/source/convert_argb.cc
index 9d64ca556..ebedf3c18 100644
--- a/source/convert_argb.cc
+++ b/source/convert_argb.cc
@@ -2334,6 +2334,11 @@ int I444AlphaToARGBMatrix(const uint8_t* src_y,
     }
   }
 #endif
+#if defined(HAS_I444ALPHATOARGBROW_SVE2)
+  if (TestCpuFlag(kCpuHasSVE2)) {
+    I444AlphaToARGBRow = I444AlphaToARGBRow_SVE2;
+  }
+#endif
 #if defined(HAS_I444ALPHATOARGBROW_MSA)
   if (TestCpuFlag(kCpuHasMSA)) {
     I444AlphaToARGBRow = I444AlphaToARGBRow_Any_MSA;
@@ -7053,6 +7058,11 @@ static int I420AlphaToARGBMatrixBilinear(
     }
   }
 #endif
+#if defined(HAS_I444ALPHATOARGBROW_SVE2)
+  if (TestCpuFlag(kCpuHasSVE2)) {
+    I444AlphaToARGBRow = I444AlphaToARGBRow_SVE2;
+  }
+#endif
 #if defined(HAS_I444ALPHATOARGBROW_MSA)
   if (TestCpuFlag(kCpuHasMSA)) {
     I444AlphaToARGBRow = I444AlphaToARGBRow_Any_MSA;
@@ -7263,6 +7273,11 @@ static int I422AlphaToARGBMatrixLinear(const uint8_t* src_y,
     }
   }
 #endif
+#if defined(HAS_I444ALPHATOARGBROW_SVE2)
+  if (TestCpuFlag(kCpuHasSVE2)) {
+    I444AlphaToARGBRow = I444AlphaToARGBRow_SVE2;
+  }
+#endif
 #if defined(HAS_I444ALPHATOARGBROW_MSA)
   if (TestCpuFlag(kCpuHasMSA)) {
     I444AlphaToARGBRow = I444AlphaToARGBRow_Any_MSA;
diff --git a/source/row_sve.cc b/source/row_sve.cc
index 139931367..9862c3a4b 100644
--- a/source/row_sve.cc
+++ b/source/row_sve.cc
@@ -161,6 +161,54 @@ void I422ToARGBRow_SVE2(const uint8_t* src_y,
       : "cc", "memory", YUVTORGB_SVE_REGS);
 }
 
+void I444AlphaToARGBRow_SVE2(const uint8_t* src_y,
+                             const uint8_t* src_u,
+                             const uint8_t* src_v,
+                             const uint8_t* src_a,
+                             uint8_t* dst_argb,
+                             const struct YuvConstants* yuvconstants,
+                             int width) {
+  uint64_t vl;
+  asm("cnth     %[vl]                                   \n"
+      "ptrue    p0.b                                    \n" YUVTORGB_SVE_SETUP
+      "subs     %w[width], %w[width], %w[vl]            \n"
+      "b.lt     2f                                      \n"
+
+      // Run bulk of computation with an all-true predicate to avoid predicate
+      // generation overhead.
+      "ptrue    p1.h                                    \n"
+      "1:                                               \n" READYUV444_SVE
+      "ld1b     {z19.h}, p1/z, [%[src_a]]               \n"  // A
+      "add      %[src_a], %[src_a], %[vl]               \n" I4XXTORGB_SVE
+          RGBTORGBA8_SVE
+      "subs     %w[width], %w[width], %w[vl]            \n"
+      "st2h     {z16.h, z17.h}, p1, [%[dst_argb]]       \n"
+      "add      %[dst_argb], %[dst_argb], %[vl], lsl #2 \n"
+      "b.ge     1b                                      \n"
+
+      "2:                                               \n"
+      "adds     %w[width], %w[width], %w[vl]            \n"
+      "b.eq     99f                                     \n"
+
+      // Calculate a predicate for the final iteration to deal with the tail.
+      "whilelt  p1.h, wzr, %w[width]                    \n" READYUV444_SVE
+      "ld1b     {z19.h}, p1/z, [%[src_a]]               \n"  // A
+      I4XXTORGB_SVE RGBTORGBA8_SVE
+      "st2h     {z16.h, z17.h}, p1, [%[dst_argb]]       \n"
+
+      "99:                                              \n"
+      : [src_y] "+r"(src_y),                               // %[src_y]
+        [src_u] "+r"(src_u),                               // %[src_u]
+        [src_v] "+r"(src_v),                               // %[src_v]
+        [src_a] "+r"(src_a),                               // %[src_a]
+        [dst_argb] "+r"(dst_argb),                         // %[dst_argb]
+        [width] "+r"(width),                               // %[width]
+        [vl] "=&r"(vl)                                     // %[vl]
+      : [kUVCoeff] "r"(&yuvconstants->kUVCoeff),           // %[kUVCoeff]
+        [kRGBCoeffBias] "r"(&yuvconstants->kRGBCoeffBias)  // %[kRGBCoeffBias]
+      : "cc", "memory", YUVTORGB_SVE_REGS);
+}
+
 #endif  // !defined(LIBYUV_DISABLE_SVE) && defined(__aarch64__)
 
 #ifdef __cplusplus