“Encharge AI's on-device chips are designed to support specialized models in the 10 to 20 billion parameter range.”