Power limits may require preprocessing or signal classification to happen on-device, using lightweight models or thresholds. Full AI processing may be deferred to the cloud, so the device sends only essential features, reducing energy use while still enabling intelligent insights.