The chip, designed for inference, will be deployed by end of 2026 and aims to improve performance per watt — Klaro News