Ceva said a major U.S. software and AI platform company chose its NeuPro-M IP for a custom AI silicon program aimed at next-generation intelligent computing devices.
NeuPro-M will power on-device inference for generative, multimodal and agentic AI workloads, with Ceva also tailoring neural-network optimizations to improve efficiency and performance under edge power and thermal limits.
The deal pushes Ceva beyond its traditional semiconductor and device-OEM customer base into software platform companies that are increasingly designing their own chips to optimize performance, power and area.
CEO Amir Panush called it one of Ceva's most strategically significant AI licensing agreements, underscoring a broader shift toward AI-first architectures as workloads spread across cloud and edge devices.
Ceva said more than 2 billion devices using its technologies ship annually, and it expects NeuPro licensing momentum to keep expanding across consumer, industrial, automotive, infrastructure and computing markets.
With hyperscalers building their own AI chips, who is Ceva's mystery US partner, and does this deal challenge the Broadcom-Marvell duopoly?
As agentic AI moves to devices, can custom silicon deliver autonomous intelligence without the high cost and latency of the cloud?
CEVA’s NeuPro-M NPU Chosen for Custom AI Silicon in Major U.S. Partnership, Delivering 350 TOPS/Watt at the Edge
Overview
CEVA, Inc. has announced a major licensing deal with a leading U.S. software and AI platform company, focusing on developing custom AI silicon. This partnership highlights a growing trend where top software companies design their own silicon to boost performance for their AI platforms. CEVA will use its advanced NeuPro-M NPU IP as the core technology, working closely with the customer to optimize neural networks for better inference efficiency and performance. This move marks a significant step for CEVA in the edge AI market, showing its ability to deliver tailored, high-performance AI solutions.