diff --git a/CHANGELOG.md b/CHANGELOG.md
index fff4578b101..eec4a7b0577 100644
--- a/CHANGELOG.md
+++ b/CHANGELOG.md
@@ -63,7 +63,7 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 - Fixed multi device aggregation in `PearsonCorrCoef` ([#998](https://github.com/PyTorchLightning/metrics/pull/998))
 
 
--
+- Fixed compatibility with future Pytorch 1.12 in `pairwise_cosine_similarity` ([#1011](https://github.com/PyTorchLightning/metrics/pull/1011))
 
 
 ## [0.8.1] - 2022-04-27
diff --git a/torchmetrics/functional/pairwise/cosine.py b/torchmetrics/functional/pairwise/cosine.py
index 74df87352bb..29103a52dd9 100644
--- a/torchmetrics/functional/pairwise/cosine.py
+++ b/torchmetrics/functional/pairwise/cosine.py
@@ -20,6 +20,16 @@
 from torchmetrics.functional.pairwise.helpers import _check_input, _reduce_distance_matrix
 
 
+def _safe_matmul(x: Tensor, y: Tensor) -> Tensor:
+    """Safe calculation of matrix multiplication.
+
+    If input is float16, will cast to float32 for computation and back again.
+    """
+    if x.dtype == torch.float16 or y.dtype == torch.float16:
+        return (x.float() @ y.T.float()).half()
+    return x @ y.T
+
+
 def _pairwise_cosine_similarity_update(
     x: Tensor, y: Optional[Tensor] = None, zero_diagonal: Optional[bool] = None
 ) -> Tensor:
@@ -37,7 +47,7 @@ def _pairwise_cosine_similarity_update(
     norm = torch.norm(y, p=2, dim=1)
     y /= norm.unsqueeze(1)
 
-    distance = x @ y.T
+    distance = _safe_matmul(x, y)
     if zero_diagonal:
         distance.fill_diagonal_(0)
     return distance