diff --git a/esphome/components/esphome/ota/ota_esphome_noise.cpp b/esphome/components/esphome/ota/ota_esphome_noise.cpp
index a9ed552211..c33f97cdff 100644
--- a/esphome/components/esphome/ota/ota_esphome_noise.cpp
+++ b/esphome/components/esphome/ota/ota_esphome_noise.cpp
@@ -41,7 +41,9 @@ ESPHomeOTAComponent::NoiseSession::~NoiseSession() {
  */
 bool ESPHomeOTAComponent::noise_start_session_(uint8_t server_feature_flags) {
   // NOLINTNEXTLINE(clang-analyzer-cplusplus.NewDeleteLeaks)
-  this->noise_ = std::unique_ptr<NoiseSession>(new (std::nothrow) NoiseSession());
+  // Default-init: the frame buffer is always written before it is read, so
+  // skip zeroing its 132 bytes
+  this->noise_ = std::unique_ptr<NoiseSession>(new (std::nothrow) NoiseSession);
   static constexpr size_t PROLOGUE_ACK_LEN = 2;  // OTA_RESPONSE_OK + version
   static constexpr size_t PROLOGUE_CLIENT_FEATURES_LEN = 1;
   static constexpr size_t PROLOGUE_FEATURE_ACK_LEN = 2;  // OTA_RESPONSE_FEATURE_FLAGS + server flags
@@ -165,14 +167,20 @@ size_t ESPHomeOTAComponent::noise_frame_payload_len_(const uint8_t *header, size
 /// Non-blocking read of one handshake frame into the session buffer.
 bool ESPHomeOTAComponent::noise_try_read_frame_() {
   NoiseSession &s = *this->noise_;
-  while (s.frame_pos < noise::FRAME_HEADER_SIZE) {
-    ssize_t read = this->client_->read(s.frame_buf + s.frame_pos, noise::FRAME_HEADER_SIZE - s.frame_pos);
-    if (!this->handle_read_error_(read, LOG_STR("read noise"))) {
-      return false;
+  while (true) {
+    // The header first, then the body once the header says how long it is
+    const uint16_t want = s.frame_len == 0 ? noise::FRAME_HEADER_SIZE : s.frame_len;
+    if (s.frame_pos < want) {
+      ssize_t read = this->client_->read(s.frame_buf + s.frame_pos, want - s.frame_pos);
+      if (!this->handle_read_error_(read, LOG_STR("read noise"))) {
+        return false;
+      }
+      s.frame_pos += read;
+      continue;
+    }
+    if (s.frame_len != 0) {
+      return true;
     }
-    s.frame_pos += read;
-  }
-  if (s.frame_len == 0) {
     const size_t payload_len = this->noise_frame_payload_len_(s.frame_buf, 1, 1 + noise::MAX_HANDSHAKE_SIZE);
     if (payload_len == 0) {
       this->cleanup_connection_();
@@ -180,14 +188,6 @@ bool ESPHomeOTAComponent::noise_try_read_frame_() {
     }
     s.frame_len = noise::FRAME_HEADER_SIZE + payload_len;
   }
-  while (s.frame_pos < s.frame_len) {
-    ssize_t read = this->client_->read(s.frame_buf + s.frame_pos, s.frame_len - s.frame_pos);
-    if (!this->handle_read_error_(read, LOG_STR("read noise"))) {
-      return false;
-    }
-    s.frame_pos += read;
-  }
-  return true;
 }
 
 /// Non-blocking write of the pending session-buffer frame.