Ceva’s NeuPro-M neural processing unit will serve as the core accelerator for the program. By moving beyond traditional semiconductor clients to partner directly with software platform owners, Ceva is positioning its hardware to handle the specific demands of generative and multimodal AI workloads. This collaboration allows for fine-tuned OS-to-silicon optimization, a critical requirement for devices where thermal management and battery life are strictly constrained.

Amir Panush, CEO of Ceva, characterized the agreement as one of the most strategically significant in the company’s history. As AI tasks distribute between cloud and edge environments, the ability to sense and reason locally has become a competitive differentiator. The NeuPro-M architecture provides the scalability required for these tasks, allowing the customer to co-design hardware that balances raw performance with the power efficiency necessary for advanced intelligent computing.