In this paper, a novel power efficient architecture for detecting the spectral holes over a very high bandwidth is proposed. Wideband spectrum sensing poses serious challenges for low-power spectrum sharing radios (for e.g., cognitive radio) which cannot afford to use high-rate Analog-to-Digital Converters (ADCs) to sample the signals at Nyquist rate and process them digitally thereby spending a lot of power. Here we propose an analog/mixed signal topology for wideband spectrum sensing that replaces the conventional Nyquist ADCs and digital Fast Fourier Transform (FFT) core with a bank of Sample and Hold (S/H) circuits, each operating at sub-Nyquist rate, and an all-analog FFT processor. The results show that even though sub-optimal analog processing leads to worse spectral reconstruction compared to conventional techniques, good detection performances can be achieved along with a substantial reduction in the power consumption.