diff --git a/mlir/include/mlir/Dialect/GPU/GPUOps.td b/mlir/include/mlir/Dialect/GPU/GPUOps.td
--- a/mlir/include/mlir/Dialect/GPU/GPUOps.td
+++ b/mlir/include/mlir/Dialect/GPU/GPUOps.td
@@ -773,4 +773,25 @@
   let assemblyFormat = "(`[` $asyncDependencies^ `]`)? attr-dict";
 }
 
+def GPU_WaitOp : GPU_Op<"host_wait", [GPU_AsyncOpInterface]> {
+  let summary = "GPU async blocking wait operation";
+  let description = [{
+    The `host_wait` operation blocks until all ops producing the async
+    dependency tokens finished execution.
+
+    All dependent memory operations are visible to the host once this op
+    completes.
+  }];
+
+  let arguments = (ins Variadic<GPU_AsyncToken>:$asyncDependencies);
+  let results = (outs);
+
+  let extraClassDeclaration = [{
+      // This op never returns an async token.
+      Value asyncToken() const { return Value(); }
+  }];
+
+  let assemblyFormat = "`[` $asyncDependencies `]` attr-dict";
+}
+
 #endif // GPU_OPS
diff --git a/mlir/test/Dialect/GPU/ops.mlir b/mlir/test/Dialect/GPU/ops.mlir
--- a/mlir/test/Dialect/GPU/ops.mlir
+++ b/mlir/test/Dialect/GPU/ops.mlir
@@ -160,4 +160,13 @@
     %2 = gpu.create_token[%0, %1]
     return
   }
+
+  func @async_host_wait() {
+    // CHECK-LABEL: func @async_host_wait
+    // CHECK: %[[t0:.*]] = gpu.create_token
+    %0 = gpu.create_token
+    // CHECK: gpu.host_wait[%[[t0]]]
+    gpu.host_wait[%0]
+    return
+  }
 }