diff --git a/docs/release-notes/0.16.0.md b/docs/release-notes/0.16.0.md
index 4e92893b..d409138b 100644
--- a/docs/release-notes/0.16.0.md
+++ b/docs/release-notes/0.16.0.md
@@ -15,6 +15,7 @@
 ```
 * {func}`~rapids_singlecell.pp.neighbors` with ``method="jaccard"`` no longer crashes with ``CUDA_ERROR_ILLEGAL_ADDRESS`` on large datasets where ``n_obs * (n_neighbors - 1)**2`` exceeds ``2**31`` {pr}`700` {smaller}`S Dicks`
 * {func}`~rapids_singlecell.pp.pca` no longer imports a removed cuML internal, fixing ``ImportError`` on recent cuML builds {pr}`708` {smaller}`S Dicks`
+* Promote dense integer inputs in {func}`~rapids_singlecell.pp.normalize_total` and {func}`~rapids_singlecell.pp.scale` to floating point {pr}`718` {smaller}`S Dicks`
 * Fix {func}`~rapids_singlecell.pp.scale` with ``obsm`` and metadata changes from {func}`~rapids_singlecell.pp.log1p` with ``inplace=False`` {pr}`719` {smaller}`S Dicks`
 * Reject Dask arrays chunked along the feature axis {pr}`717` {smaller}`S Dicks`
 
diff --git a/src/rapids_singlecell/preprocessing/_normalize.py b/src/rapids_singlecell/preprocessing/_normalize.py
index 2b18a29f..a551172f 100644
--- a/src/rapids_singlecell/preprocessing/_normalize.py
+++ b/src/rapids_singlecell/preprocessing/_normalize.py
@@ -80,7 +80,9 @@ def normalize_total(
 
     _check_gpu_X(X, allow_dask=True)
 
-    if not inplace:
+    if X.dtype.kind in "iu":
+        X = X.astype(cp.float32)
+    elif not inplace:
         X = X.copy()
 
     if sparse.isspmatrix_csc(X):
diff --git a/src/rapids_singlecell/preprocessing/_scale.py b/src/rapids_singlecell/preprocessing/_scale.py
index e4f12700..35773e66 100644
--- a/src/rapids_singlecell/preprocessing/_scale.py
+++ b/src/rapids_singlecell/preprocessing/_scale.py
@@ -137,6 +137,8 @@ def scale(
 
 
 def _scale_dispatch(X, *, mask_obs, zero_center, inplace, max_value):
+    if X.dtype.kind in "iu":
+        X = X.astype(cp.float64)
     if isinstance(X, DaskArray):
         return _scale_dask(
             X,
diff --git a/tests/test_normalization.py b/tests/test_normalization.py
index ad877686..b0ae7323 100644
--- a/tests/test_normalization.py
+++ b/tests/test_normalization.py
@@ -27,6 +27,18 @@ def test_normalize_total(dtype, sparse):
     )
 
 
+@pytest.mark.parametrize("dtype", [np.int32, np.int64])
+def test_normalize_total_promotes_dense_integers(dtype):
+    cudata = AnnData(cp.array([[1, 1], [2, 4]], dtype=dtype))
+
+    rsc.pp.normalize_total(cudata, target_sum=10)
+
+    assert cudata.X.dtype == cp.float32
+    cp.testing.assert_allclose(
+        cudata.X.sum(axis=1), cp.full(cudata.n_obs, 10, dtype=cp.float32)
+    )
+
+
 @pytest.mark.parametrize("dtype", [np.float32, np.float64])
 def test_normalize_total_layers(dtype):
     cudata = AnnData(csr_matrix(X_total, dtype=dtype))
diff --git a/tests/test_scaling.py b/tests/test_scaling.py
index a099a91c..305c32ea 100644
--- a/tests/test_scaling.py
+++ b/tests/test_scaling.py
@@ -113,6 +113,16 @@ def test_scale_simple(dtype):
     )
 
 
+@pytest.mark.parametrize("dtype", [np.int32, np.int64])
+def test_scale_promotes_dense_integers(dtype):
+    adata = AnnData(cp.array(X_original, dtype=dtype))
+
+    rsc.pp.scale(adata)
+
+    assert adata.X.dtype == cp.float64
+    cp.testing.assert_allclose(adata.X, X_centered_original)
+
+
 def test_scale_obsm_does_not_write_var_statistics():
     adata = AnnData(cp.ones((3, 4), dtype=cp.float32))
     adata.obsm["X_embedding"] = cp.array([[1, 2], [2, 4], [3, 6]], dtype=cp.float32)