Everyone is talking about AI compute, but very few are paying attention to the layer that will quietly determine how far this scaling can actually go: energy.
As AI systems become more powerful, the constraint is no longer just GPUs or model performance, but the ability to deliver and manage power efficiently inside data centers. Power density, heat, and conversion losses are becoming structural bottlenecks, not secondary considerations, and solving them is critical if current growth trajectories are meant to hold.
This is where companies like Navitas Semiconductor position themselves. Their focus on GaN technology is not just about incremental innovation, but about enabling more efficient power conversion in systems that are already operating near physical limits, reducing energy loss while improving performance per watt in increasingly demanding environments.
The market, however, seems to be pricing this future in ahead of execution, with high valuation multiples and profitability still projected years out. Which raises a more interesting question: are we looking at a genuinely underappreciated layer of AI infrastructure, or simply assigning value to a narrative that still needs to prove itself?
