Indoor localization is an important part of the application of the Internet of Things (IoT). Machine learning methods, such as Convolutional Neural Networks (CNNs) are popular research topics in indoor localization tasks. Spiking neural networks (SNNs) utilize the biological features of the neuron, where the values passed through layers are all 0 and 1, making the process sparse and requiring less computation. With specially designed hardware such as Spinnaker2, SNNs are particularly suited for applications that have restrictions on battery power and size. In this work, we implement an SNN model for indoor localization using wireless channel measurements and evaluate the energy consumption on both the NVIDIA chip and the neuromorphic hardware. The dataset is obtained from a 5G system using a USRP X410 platform with one transmitter and four receivers operating at 3.3\,GHz and with a 100\,MHz bandwidth. Channel estimates and range–Doppler maps are obtained through a DSP pipeline and then input to both a CNN and an SNN in the same network structure. The SNN is based on leaky integrate-and-fire neurons and trained with surrogate gradients with SnnTorch. We consider the classification of transmitter positions on a grid with a spacing of 0.4\,m. For 36 location classes, the CNN achieves 95.28\% validation accuracy, while the SNN reaches 66.67\%. When grouping positions into 9 classes (0.8\,m resolution), the SNN accuracy increases to 92.86\%, approaching the CNN performance of 95.19\%. These results show that SNNs can provide competitive localization performance at moderate spatial resolution. The SNN model can be implemented on the SpiNNaker2 neuromorphic platform, which enables highly parallel, event-driven processing. Our previous experiments with related models indicate up to 70\% energy reduction compared to conventional embedded platforms such as NVIDIA Jetson Xavier at comparable processing time. This highlights the potential of SNNs and neuromorphic hardware for energy-efficient real-time indoor localization.