New Heat-Managing Software Makes AI Run 40% Faster on Your Phone

When you ask an AI chatbot on your phone a question, the answer has to be generated by a chip the size of a fingernail. That chip — the processor, the graphics unit, and the memory — all share a single cooling path. Run AI hard for a minute and things get hot. Once a device gets too warm, it deliberately slows itself down to avoid damage, which is why your phone sometimes feels sluggish or the AI seems to stall mid-sentence.

A team of researchers has built software called ThermE that tackles this problem by forecasting heat instead of just reacting to it. Think of it like a smart thermostat for your phone's brain: rather than waiting until the room is uncomfortably hot, it predicts the temperature a few steps ahead and adjusts how the AI works — moment by moment — so the device never gets pushed into slowdown mode. It does this without actually running the AI to check, which keeps it fast and cheap.

In tests across four different AI workloads, ThermE made the AI's first word appear 40.55% faster and sped up each following word by about 12%. It also cut the rate of missed performance promises roughly in half — 5.70% versus 12.30% for the best competing method — while predicting temperatures within about 2°C.

Why should you care? As AI moves off big data centers and onto phones, laptops, smart speakers, and cars, heat becomes the real limit on what these devices can do. Better heat management means faster answers, longer battery-friendly sessions, and AI features that don't quietly degrade after a few minutes of use. The catch is that ThermE currently lives in a research setting, built on top of a tool used mostly by engineers. It isn't a consumer app you can install, and real-world devices vary widely — so don't expect an instant upgrade on your phone.