A base station emits radio waves. Without a special nozzle, the antenna acts like a garden hose opened full blast: water flies everywhere, but only a little reaches the flower beds. Researchers created a system with two neural networks. The first picks the optimal antenna size: its aperture (basically, the radiating surface area). The second learns to rapidly calculate how to steer narrow beams toward all users, even if they move chaotically. It’s like watering where the hose itself instantly finds each plant and gives just the right amount of water, without losing a drop.

This approach nears ideal efficiency but works tens of times faster than classical algorithms. As a result, future 5G and 6G networks will serve more devices while consuming less electricity. And today, cellular communications already gobble up about 1% of the world’s electricity — with up to 80% of that energy turning into heat that must be removed by massive cooling systems.