diff --git a/containers/base/Dockerfile b/containers/base/Dockerfile
index 5a2fa5b43..2abd6c4ab 100644
--- a/containers/base/Dockerfile
+++ b/containers/base/Dockerfile
@@ -15,7 +15,7 @@
 # We use different base images for GPU vs CPU Dockerfiles, so we expect
 # that the appropriate image is pulled and tagged locally.
 # CPU should use ubuntu:16.04
-# and GPU uses nvidia/cuda:8.0-cudnn6-devel-ubuntu16.04
+# and GPU uses nvidia/cuda:9.0-cudnn7-devel-ubuntu16.04
 FROM datalab-external-base-image
 MAINTAINER Google Cloud DataLab
 
@@ -89,7 +89,7 @@ RUN echo "deb-src http://ftp.us.debian.org/debian testing main" >> /etc/apt/sour
     pip install -U --upgrade-strategy only-if-needed --no-cache-dir pillow==3.4.1 && \
     pip install -U --upgrade-strategy only-if-needed --no-cache-dir google-cloud-dataflow==2.0.0 && \
     pip install -U --upgrade-strategy only-if-needed --no-cache-dir h5py==2.7.1 && \
-    pip install -U --upgrade-strategy only-if-needed --no-cache-dir tensorflow==1.4.1 && \
+    pip install -U --upgrade-strategy only-if-needed --no-cache-dir tensorflow==1.5.0 && \
     pip install -U --upgrade-strategy only-if-needed --no-cache-dir python-snappy==0.5.1 && \
     pip install -U --upgrade-strategy only-if-needed --no-cache-dir xgboost==0.6a2 && \
     find /usr/local/lib/python2.7 -type d -name tests | xargs rm -rf && \
@@ -141,7 +141,7 @@ RUN echo "deb-src http://ftp.us.debian.org/debian testing main" >> /etc/apt/sour
     pip3 install -U --upgrade-strategy only-if-needed --no-cache-dir bs4==0.0.1 && \
     pip3 install -U --upgrade-strategy only-if-needed --no-cache-dir crcmod==1.7 && \
     pip3 install -U --upgrade-strategy only-if-needed --no-cache-dir pillow==3.4.1 && \
-    pip3 install -U --upgrade-strategy only-if-needed --no-cache-dir tensorflow==1.4.1 && \
+    pip3 install -U --upgrade-strategy only-if-needed --no-cache-dir tensorflow==1.5.0 && \
     pip3 install -U --upgrade-strategy only-if-needed --no-cache-dir python-snappy==0.5.1 && \
     pip3 install -U --upgrade-strategy only-if-needed --no-cache-dir xgboost==0.6a2 && \
 # Install IPython related packages with no-deps, to ensure that we don't
diff --git a/containers/base/Dockerfile.gpu b/containers/base/Dockerfile.gpu
index 591ae883a..13f942cfb 100644
--- a/containers/base/Dockerfile.gpu
+++ b/containers/base/Dockerfile.gpu
@@ -16,7 +16,7 @@ FROM datalab-core-gpu
 MAINTAINER Google Cloud DataLab
 
 # Download and Install GPU specific packages
-RUN pip install -U --upgrade-strategy only-if-needed --no-cache-dir tensorflow-gpu==1.4.1 tflearn h5py && \
-    pip3 install -U --upgrade-strategy only-if-needed --no-cache-dir tensorflow-gpu==1.4.1 tflearn h5py
+RUN pip install -U --upgrade-strategy only-if-needed --no-cache-dir tensorflow-gpu==1.5.0 tflearn h5py && \
+    pip3 install -U --upgrade-strategy only-if-needed --no-cache-dir tensorflow-gpu==1.5.0 tflearn h5py
 
 
diff --git a/containers/base/build.gpu.sh b/containers/base/build.gpu.sh
index 006cdacdf..773da9f81 100755
--- a/containers/base/build.gpu.sh
+++ b/containers/base/build.gpu.sh
@@ -32,11 +32,11 @@ fi
 
 trap 'rm -rf pydatalab' exit
 
-docker pull nvidia/cuda:8.0-cudnn6-devel-ubuntu16.04
+docker pull nvidia/cuda:9.0-cudnn7-devel-ubuntu16.04
 # Docker tag flags changed in an incompatible way between versions.
 # The Datalab Jenkins build still uses the old one, so try it both ways.
-if ! $(docker tag -f nvidia/cuda:8.0-cudnn6-devel-ubuntu16.04 datalab-external-base-image); then
-  docker tag nvidia/cuda:8.0-cudnn6-devel-ubuntu16.04 datalab-external-base-image
+if ! $(docker tag -f nvidia/cuda:9.0-cudnn7-devel-ubuntu16.04 datalab-external-base-image); then
+  docker tag nvidia/cuda:9.0-cudnn7-devel-ubuntu16.04 datalab-external-base-image
 fi
 docker build ${DOCKER_BUILD_ARGS} -t datalab-core-gpu .
 docker build ${DOCKER_BUILD_ARGS} -f Dockerfile.gpu -t datalab-base-gpu .
diff --git a/tools/release/cloudbuild.yaml b/tools/release/cloudbuild.yaml
index 636000ff0..7987f9785 100644
--- a/tools/release/cloudbuild.yaml
+++ b/tools/release/cloudbuild.yaml
@@ -66,10 +66,10 @@ steps:
 
 ## Second, we build the GPU base image
 - name: 'gcr.io/cloud-builders/docker'
-  args: ['pull', 'nvidia/cuda:8.0-cudnn6-devel-ubuntu16.04']
+  args: ['pull', 'nvidia/cuda:9.0-cudnn7-devel-ubuntu16.04']
   id:   'pullNvidiaUbuntu'
 - name: 'gcr.io/cloud-builders/docker'
-  args: ['tag', 'nvidia/cuda:8.0-cudnn6-devel-ubuntu16.04', 'datalab-external-base-image']
+  args: ['tag', 'nvidia/cuda:9.0-cudnn7-devel-ubuntu16.04', 'datalab-external-base-image']
   id:   'tagNvidiaUbuntu'
   waitFor: ['buildBase', 'pullNvidiaUbuntu']
 - name: 'gcr.io/cloud-builders/docker'