diff --git a/ray-on-gke/TPU_guide.md b/ray-on-gke/TPU_guide.md
index e004dd206..3cce5b3a8 100644
--- a/ray-on-gke/TPU_guide.md
+++ b/ray-on-gke/TPU_guide.md
@@ -109,6 +109,6 @@ init_jax_from_ray(num_workers=2)
 
 ### TPU Multi-Host Workloads
 
-When initializing multi-host TPUs, the environment variables can be set using a mutating admission webhook. The webhook can be deployed following the instructions in the [README](https://github.com/GoogleCloudPlatform/ai-on-gke/tree/kuberay-tpu-env-injector/ray-on-gke/user/kuberay-tpu-webhook#readme).
+When initializing multi-host TPUs, the environment variables can be set using a mutating admission webhook. The webhook can be deployed following the instructions in the [README](https://github.com/GoogleCloudPlatform/ai-on-gke/blob/main/ray-on-gke/kuberay-tpu-webhook#readme).
 
 A caveat when running multiple TPU pod slices of the same topology and type with Ray is that a single Ray worker group may be scheduled across multiple pod slices. This goes against the assumptions of the webhook and would lead to  pod-to-pod communication occuring over DCN rather than the high bandwidth ICI mesh.
\ No newline at end of file