diff --git a/esphome/components/esp8266/hal.h b/esphome/components/esp8266/hal.h
index effa9c9371..f3b33da692 100644
--- a/esphome/components/esp8266/hal.h
+++ b/esphome/components/esp8266/hal.h
@@ -58,6 +58,12 @@ __attribute__((always_inline)) inline const char *progmem_read_ptr(const char *c
 __attribute__((always_inline)) inline uint16_t progmem_read_uint16(const uint16_t *addr) {
   return pgm_read_word(addr);  // NOLINT
 }
+// Bulk PROGMEM copy: routes to the SDK's aligned-flash `memcpy_P` so callers
+// don't have to drop to a byte-by-byte `progmem_read_byte` loop, which on
+// ESP8266 is ~4x as many flash accesses as the bulk path.
+__attribute__((always_inline)) inline void progmem_memcpy(void *dst, const void *src, size_t len) {
+  memcpy_P(dst, src, len);  // NOLINT
+}
 
 // NOLINTNEXTLINE(readability-identifier-naming)
 __attribute__((always_inline)) inline void delayMicroseconds(uint32_t us) { delay_microseconds_safe(us); }
diff --git a/esphome/core/hal.h b/esphome/core/hal.h
index 4babda807d..b44a422836 100644
--- a/esphome/core/hal.h
+++ b/esphome/core/hal.h
@@ -1,6 +1,7 @@
 #pragma once
-#include <string>
 #include <cstdint>
+#include <cstring>
+#include <string>
 #include "gpio.h"
 #include "esphome/core/defines.h"
 #include "esphome/core/time_64.h"
@@ -42,6 +43,9 @@ void __attribute__((noreturn)) arch_restart();
 inline uint8_t progmem_read_byte(const uint8_t *addr) { return *addr; }
 inline const char *progmem_read_ptr(const char *const *addr) { return *addr; }
 inline uint16_t progmem_read_uint16(const uint16_t *addr) { return *addr; }
+// Bulk copy out of PROGMEM. PROGMEM is a no-op everywhere except ESP8266, so a
+// plain `std::memcpy` is correct and the fast path here.
+inline void progmem_memcpy(void *dst, const void *src, size_t len) { std::memcpy(dst, src, len); }
 #endif
 
 }  // namespace esphome