In an era of ever-growing digitalization, the absorbed power of processing units is becoming an actual challenge for cooling systems. The effectiveness is imperative, but compactness and passiveness are driving factors in the design as well. The goal of this paper is twofold: (1) to present a detailed experimental campaign on a thermosyphon system for high-heat-load electronics and (2) to propose a model of the thermosyphon system using a Machine Learning approach. The thermosyphon system is composed of a microchannel evaporator plate directly attached to the heat-generating device and an air-cooled multiport condenser. The height between the evaporator and condenser inlets is 12 cm. The condenser is also proposed in two solutions: the first one has a footprint heat exchange area of 180 × 120 mm2, which allows a single fan's placement; the second one has a footprint area of 240 × 120 mm2, allowing the placement of two fans. The working fluid used in the system is R1234ze(E) with different charges. The experimental results show that the single-fan condenser reached a maximum heat rejection of 330 W, corresponding to a heat flux of 21.9 W/cm2. The double-fan condenser bore a maximum heat rejection of 570 W (37.7 W/cm2). The model, constructed purely via a machine learning tool, shows a satisfactory agreement between experimental and predicted data.