We just announced support for TensorFlow on Windows with the 0.12 release candidate. However, due to the vagaries of compiler versions on Windows, we only support Python 3.5 (see below for a fuller explanation).
At present there is no conda package for TensorFlow on Windows, but there is a pre-built PIP package in PyPI. If you install the 64-bit version of Python 3.5, either from Python.org or Anaconda, you can install TensorFlow with the following command (for the CPU-only version):
C:\> pip install tensorflow
If you have a GPU that supports CUDA 8.0, you can use the following command to install the GPU-accelerated version:
C:\> pip install tensorflow-gpu
Why does TensorFlow only support Python 3.5 on Windows? Python on Windows requires that you compile extensions using the same compiler as the one that was used to build the Python interpreter. The official distributions of Python 2.7 from Python.org and Anaconda were built with MSVC 9.0. We can't build TensorFlow with MSVC 9.0 (2008), because it relies on too many features from C++11, so we need to use MSVC 14.0 (2015). Only Python 3.5 on Windows was compiled with MSVC 14.0, so we are limited to supporting that version only.