diff --git a/include/grpcpp/channel.h b/include/grpcpp/channel.h
index 5e67c642a47..65cc46d151f 100644
--- a/include/grpcpp/channel.h
+++ b/include/grpcpp/channel.h
@@ -114,7 +114,7 @@ class Channel final : public ::grpc::ChannelInterface,
   // with this channel (if any). It is set on the first call to CallbackCQ().
   // It is _not owned_ by the channel; ownership belongs with its internal
   // shutdown callback tag (invoked when the CQ is fully shutdown).
-  ::grpc::CompletionQueue* callback_cq_ = nullptr;
+  std::atomic<CompletionQueue*> callback_cq_{nullptr};
 
   std::vector<
       std::unique_ptr<::grpc::experimental::ClientInterceptorFactoryInterface>>
diff --git a/include/grpcpp/server.h b/include/grpcpp/server.h
index 7f367dc5988..a69e64b462f 100644
--- a/include/grpcpp/server.h
+++ b/include/grpcpp/server.h
@@ -380,7 +380,7 @@ class Server : public ServerInterface, private GrpcLibraryCodegen {
   // with this server (if any). It is set on the first call to CallbackCQ().
   // It is _not owned_ by the server; ownership belongs with its internal
   // shutdown callback tag (invoked when the CQ is fully shutdown).
-  CompletionQueue* callback_cq_ ABSL_GUARDED_BY(mu_) = nullptr;
+  std::atomic<CompletionQueue*> callback_cq_{nullptr};
 
   // List of CQs passed in by user that must be Shutdown only after Server is
   // Shutdown.  Even though this is only used with NDEBUG, instantiate it in all
diff --git a/src/cpp/client/channel_cc.cc b/src/cpp/client/channel_cc.cc
index 56dee0b769a..a9f208394ec 100644
--- a/src/cpp/client/channel_cc.cc
+++ b/src/cpp/client/channel_cc.cc
@@ -57,12 +57,13 @@ Channel::Channel(const std::string& host, grpc_channel* channel,
 
 Channel::~Channel() {
   grpc_channel_destroy(c_channel_);
-  if (callback_cq_ != nullptr) {
+  CompletionQueue* callback_cq = callback_cq_.load(std::memory_order_relaxed);
+  if (callback_cq != nullptr) {
     if (grpc_iomgr_run_in_background()) {
       // gRPC-core provides the backing needed for the preferred CQ type
-      callback_cq_->Shutdown();
+      callback_cq->Shutdown();
     } else {
-      CompletionQueue::ReleaseCallbackAlternativeCQ(callback_cq_);
+      CompletionQueue::ReleaseCallbackAlternativeCQ(callback_cq);
     }
   }
 }
@@ -243,25 +244,33 @@ class ShutdownCallback : public grpc_experimental_completion_queue_functor {
 ::grpc::CompletionQueue* Channel::CallbackCQ() {
   // TODO(vjpai): Consider using a single global CQ for the default CQ
   // if there is no explicit per-channel CQ registered
+  CompletionQueue* callback_cq = callback_cq_.load(std::memory_order_acquire);
+  if (callback_cq != nullptr) {
+    return callback_cq;
+  }
+  // The callback_cq_ wasn't already set, so grab a lock and set it up exactly
+  // once for this channel.
   grpc::internal::MutexLock l(&mu_);
-  if (callback_cq_ == nullptr) {
+  callback_cq = callback_cq_.load(std::memory_order_relaxed);
+  if (callback_cq == nullptr) {
     if (grpc_iomgr_run_in_background()) {
       // gRPC-core provides the backing needed for the preferred CQ type
 
       auto* shutdown_callback = new ShutdownCallback;
-      callback_cq_ =
+      callback_cq =
           new ::grpc::CompletionQueue(grpc_completion_queue_attributes{
               GRPC_CQ_CURRENT_VERSION, GRPC_CQ_CALLBACK,
               GRPC_CQ_DEFAULT_POLLING, shutdown_callback});
 
       // Transfer ownership of the new cq to its own shutdown callback
-      shutdown_callback->TakeCQ(callback_cq_);
+      shutdown_callback->TakeCQ(callback_cq);
     } else {
       // Otherwise we need to use the alternative CQ variant
-      callback_cq_ = CompletionQueue::CallbackAlternativeCQ();
+      callback_cq = CompletionQueue::CallbackAlternativeCQ();
     }
+    callback_cq_.store(callback_cq, std::memory_order_release);
   }
-  return callback_cq_;
+  return callback_cq;
 }
 
 }  // namespace grpc
diff --git a/src/cpp/server/server_cc.cc b/src/cpp/server/server_cc.cc
index 2844bfa5137..d8f5af5e35f 100644
--- a/src/cpp/server/server_cc.cc
+++ b/src/cpp/server/server_cc.cc
@@ -930,14 +930,16 @@ Server::~Server() {
       for (const auto& value : sync_req_mgrs_) {
         value->Shutdown();
       }
-      if (callback_cq_ != nullptr) {
+      CompletionQueue* callback_cq =
+          callback_cq_.load(std::memory_order_relaxed);
+      if (callback_cq != nullptr) {
         if (grpc_iomgr_run_in_background()) {
           // gRPC-core provides the backing needed for the preferred CQ type
-          callback_cq_->Shutdown();
+          callback_cq->Shutdown();
         } else {
-          CompletionQueue::ReleaseCallbackAlternativeCQ(callback_cq_);
+          CompletionQueue::ReleaseCallbackAlternativeCQ(callback_cq);
         }
-        callback_cq_ = nullptr;
+        callback_cq_.store(nullptr, std::memory_order_release);
       }
     }
   }
@@ -1256,14 +1258,15 @@ void Server::ShutdownInternal(gpr_timespec deadline) {
 
   // Shutdown the callback CQ. The CQ is owned by its own shutdown tag, so it
   // will delete itself at true shutdown.
-  if (callback_cq_ != nullptr) {
+  CompletionQueue* callback_cq = callback_cq_.load(std::memory_order_relaxed);
+  if (callback_cq != nullptr) {
     if (grpc_iomgr_run_in_background()) {
       // gRPC-core provides the backing needed for the preferred CQ type
-      callback_cq_->Shutdown();
+      callback_cq->Shutdown();
     } else {
-      CompletionQueue::ReleaseCallbackAlternativeCQ(callback_cq_);
+      CompletionQueue::ReleaseCallbackAlternativeCQ(callback_cq);
     }
-    callback_cq_ = nullptr;
+    callback_cq_.store(nullptr, std::memory_order_release);
   }
 
   // Drain the shutdown queue (if the previous call to AsyncNext() timed out
@@ -1330,25 +1333,33 @@ grpc::ServerInitializer* Server::initializer() {
 grpc::CompletionQueue* Server::CallbackCQ() {
   // TODO(vjpai): Consider using a single global CQ for the default CQ
   // if there is no explicit per-server CQ registered
+  CompletionQueue* callback_cq = callback_cq_.load(std::memory_order_acquire);
+  if (callback_cq != nullptr) {
+    return callback_cq;
+  }
+  // The callback_cq_ wasn't already set, so grab a lock and set it up exactly
+  // once for this server.
   grpc::internal::MutexLock l(&mu_);
-  if (callback_cq_ != nullptr) {
-    return callback_cq_;
+  callback_cq = callback_cq_.load(std::memory_order_relaxed);
+  if (callback_cq != nullptr) {
+    return callback_cq;
   }
   if (grpc_iomgr_run_in_background()) {
     // gRPC-core provides the backing needed for the preferred CQ type
     auto* shutdown_callback = new grpc::ShutdownCallback;
-    callback_cq_ = new grpc::CompletionQueue(grpc_completion_queue_attributes{
+    callback_cq = new grpc::CompletionQueue(grpc_completion_queue_attributes{
         GRPC_CQ_CURRENT_VERSION, GRPC_CQ_CALLBACK, GRPC_CQ_DEFAULT_POLLING,
         shutdown_callback});
 
     // Transfer ownership of the new cq to its own shutdown callback
-    shutdown_callback->TakeCQ(callback_cq_);
+    shutdown_callback->TakeCQ(callback_cq);
   } else {
     // Otherwise we need to use the alternative CQ variant
-    callback_cq_ = CompletionQueue::CallbackAlternativeCQ();
+    callback_cq = CompletionQueue::CallbackAlternativeCQ();
   }
 
-  return callback_cq_;
+  callback_cq_.store(callback_cq, std::memory_order_release);
+  return callback_cq;
 }
 
 }  // namespace grpc