Next-generation spacecraft developers are earnestly investigating the application of artificial intelligence (AI) algorithms onboard to enable new mission concepts for space exploration and science. However, the current generation of radiation-hardened processors are inferior compared to commercial-off-the-shelf alternatives in terms of the computational performance required by modern AI applications. To address this disparity, space-system designers have started employing novel radiation-tolerant architectures combining both commercial and radiation-hardened components to mitigate radiation effects at a system level. Unfortunately, developing single-board computers with radiation-tolerant, high-performance processors is challenging because designers must balance the sparce selection of radiation-hardened power converters and high-reliability decoupling capacitors with SmallSat/CubeSat area constraints and limited thermal conduction. Consequently, the next-generation of space processors, including the AMD-Xilinx Versal Adaptive Compute Acceleration Platform require demanding power solutions capable of supplying core rails with 0.8V ± 18mV and currents up to 150 A. In this paper, we present a multifaceted analysis of the power system and decoupling network for a future Versal-based design. We developed preliminary power estimates based on expected processor and FPGA resource utilization for common AI processing applications. These estimates drive the power system requirements and a comparative analysis of single-phase integrated converters and multi-phase discrete converters for high-current FPGA supplies. We develop a tradespace between the number of phases, input voltage, load current, switching frequency, power efficiency, and printed circuit board area. Finally, we design and simulate four power delivery networks based on commercial, high-reliability, and flight-qualified capacitors to compare the efficacy of 0201 decoupling capacitors in flight missions.