diff --git a/src/pystencils/backend/platforms/cuda.py b/src/pystencils/backend/platforms/cuda.py
index 60571db9469c7504c98caf1df1d626a99c547412..6e6488ee1b461aac3694573835488d88760af361 100644
--- a/src/pystencils/backend/platforms/cuda.py
+++ b/src/pystencils/backend/platforms/cuda.py
@@ -33,7 +33,7 @@ class CudaPlatform(GenericGpu):
 
     @property
     def required_headers(self) -> set[str]:
-        return super().required_headers | {'"pystencils_runtime/cuda.cuh"', '"gpu_atomics.h"'}
+        return super().required_headers | {'"pystencils_runtime/cuda.cuh"'}
 
     def resolve_reduction(
         self,
diff --git a/src/pystencils/include/gpu_atomics.h b/src/pystencils/include/pystencils_runtime/bits/gpu_atomics.h
similarity index 100%
rename from src/pystencils/include/gpu_atomics.h
rename to src/pystencils/include/pystencils_runtime/bits/gpu_atomics.h
diff --git a/src/pystencils/include/pystencils_runtime/cuda.cuh b/src/pystencils/include/pystencils_runtime/cuda.cuh
index 6a22e0b9034d224a4fda52233faab73cabd8a01d..691519a0144ac1a50094fac73d165d8dc31adba3 100644
--- a/src/pystencils/include/pystencils_runtime/cuda.cuh
+++ b/src/pystencils/include/pystencils_runtime/cuda.cuh
@@ -3,3 +3,4 @@
 #include <cuda_fp16.h>
 
 #include "./bits/gpu_infinities.h"
+#include "./bits/gpu_atomics.h"