diff --git a/.github/workflows/build-kernels.yml b/.github/workflows/build-kernels.yml
index cc842dd..d789d38 100644
--- a/.github/workflows/build-kernels.yml
+++ b/.github/workflows/build-kernels.yml
@@ -22,8 +22,9 @@ jobs:
 
       - name: build kernels
         run: |
+          pip install build
           ts=$(date +%s)
-          DS_KERNELS_MAKE_JOBS=10 DS_KERNELS_BUILD_STRING=".dev${ts}" CUDA_ARCH_LIST="80;86;89;90" python setup.py bdist_wheel
+          DS_KERNELS_MAKE_JOBS=10 DS_KERNELS_BUILD_STRING=".dev${ts}" CUDA_ARCH_LIST="80;86;89;90" python -m build --wheel
           fname=$(ls dist)
           nname=$(echo $fname | sed 's/cp[0-9]\+-cp[0-9]\+/py3-none/' | sed 's/linux/manylinux1/')
           mv "dist/$fname" "dist/$nname"
diff --git a/.github/workflows/release.yml b/.github/workflows/release.yml
index 13afa25..fef9951 100644
--- a/.github/workflows/release.yml
+++ b/.github/workflows/release.yml
@@ -32,8 +32,9 @@ jobs:
           python release/check_release_version.py --release_version ${{ env.RELEASE_VERSION }}
       - name: Build DeepSpeed-Kernels
         run: |
+          pip install build
           ts=$(date +%s)
-          DS_KERNELS_MAKE_JOBS=10 DS_KERNELS_BUILD_STRING=".dev${ts}" CUDA_ARCH_LIST="80;86;89;90" python setup.py bdist_wheel
+          DS_KERNELS_MAKE_JOBS=10 DS_KERNELS_BUILD_STRING=".dev${ts}" CUDA_ARCH_LIST="80;86;89;90" python -m build --wheel --no-build-isolation
           fname=$(ls dist)
           nname=$(echo $fname | sed 's/cp[0-9]\+-cp[0-9]\+/py3-none/' | sed 's/linux/manylinux1/')
           mv "dist/$fname" "dist/$nname"
diff --git a/README.md b/README.md
index a5b695d..5fdfe65 100644
--- a/README.md
+++ b/README.md
@@ -32,5 +32,5 @@ pip install -v .
 
 You can create a pre-compiled portable wheel that supports different CUDA architectures via the `CUDA_ARCH_LIST` environment variable. By default the kernels will be compiled using the `native` compute capability. If you want to compile for more than one you can set the `CUDA_ARCH_LIST` environment variable. We currently only support Ampere and above architectures (i.e., 8.0+). See example below to build for GPUs like A100 and A6000:
 ```bash
-CUDA_ARCH_LIST="80;86;89;90" python setup.py bdist_wheel
+CUDA_ARCH_LIST="80;86;89;90" python -m build --wheel
 ```
diff --git a/build.sh b/build.sh
index 295c02a..6c647e4 100644
--- a/build.sh
+++ b/build.sh
@@ -1,4 +1,4 @@
 set -ex
 
 # set CUDA_ARCH_LIST if desired, otherwise falls back to native
-CUDA_ARCH_LIST="80;86" python setup.py bdist_wheel
+CUDA_ARCH_LIST="80;86" python -m build --wheel
diff --git a/setup.py b/setup.py
index b62d7a8..294eb46 100644
--- a/setup.py
+++ b/setup.py
@@ -55,11 +55,11 @@ def command_exists(cmd):
     except subprocess.CalledProcessError:
         pass
 
-# Parse the MII version string from version.txt
+# Parse the ds-kernels version string from version.txt
 version_str = open('version.txt', 'r').read().strip()
 
 # Build specifiers like .devX can be added at install time. Otherwise, add the git hash.
-# example: BUILD_STR=".dev20201022" python setup.py sdist bdist_wheel
+# example: BUILD_STR=".dev20201022" python -m build --sdist --wheel
 
 BUILD_STRING = 'DS_KERNELS_BUILD_STRING'
 BUILD_FILE = 'build.txt'