Brain-Like Chips Could Make AI 37x Faster and Use 16x Less Power

AI assistants like ChatGPT are powered by massive language models that run on energy-hungry computer servers. A new paper from an international research team shows a smarter way to run these models: on neuromorphic hardware, which mimics the human brain. Instead of constantly processing every piece of data, brain-like chips only activate when they need to, like a light switch that turns on only when someone enters the room.

Today's AI models are also limited by a bottleneck called the KV cache, which is like a short-term memory that fills up quickly. The researchers replaced this with a more efficient design called a state-space model, which uses a fixed-size memory. Then they added a clever trick: they train the model to ignore unimportant data points and only focus on the critical ones. This "sparse" approach means the AI does up to four times fewer calculations, with almost no drop in accuracy.

The real magic is the hardware. Neuromorphic chips process events as they happen, not on a fixed schedule. This lets them turn the sparse, selective data into even bigger speed gains. Compared to a standard GPU on a phone or drone, the new approach could deliver up to 37 times higher throughput and 16 times lower power consumption. That's the difference between a battery that lasts hours and one that lasts days.

What does this mean for you? Future phones, smartwatches, and other devices could run powerful AI locally, without sending your data to the cloud. That means faster responses, better privacy, and far less battery drain. The catch? This research is still early. Brain-inspired chips aren't in your phone yet, and real-world results may vary. But it's a strong sign that the next wave of AI won't just be smarter — it'll be more efficient and more personal.