CEVA Enters a New Market: Software Companies Designing Their Own AI Chips

6 July, 2026

CEVA has licensed its NeuPro-M AI processor technology to a major U.S. software company developing its first custom AI chip for intelligent computing devices

Photo above: CEVA CEO Amir Panush. Photo: Netti Levy.

Israeli semiconductor IP company CEVA has signed a strategic AI licensing agreement with a major U.S. software and AI platform company that is developing its own custom AI processor. Under the agreement, CEVA’s NeuPro-M neural processing technology will be integrated into a custom AI chip designed for intelligent computing devices.

CEVA described the deal as strategically significant because it expands the company’s addressable market beyond traditional semiconductor vendors and OEMs to include software platform companies.

The agreement reflects a broader industry trend in which software and AI platform providers—including Google, Meta, Microsoft, OpenAI, and Anthropic—are increasingly designing proprietary silicon to optimize AI performance, improve energy efficiency, and better integrate hardware with their software stacks.

“This is one of the world’s leading software and AI platform companies,” said CEVA CEO Amir Panush. “Its decision to build an AI processor based on NeuPro-M reflects the industry’s shift toward AI-centric computing architectures. As AI workloads become increasingly distributed between the cloud and edge devices, platform companies need to optimize the entire technology stack—from silicon and software to operating systems and user experience. This represents one of the most strategically important AI licensing agreements in CEVA’s history.”

CEVA’s NeuPro family of neural processing units (NPUs) is designed to accelerate AI inference across a wide range of applications, from ultra-low-power embedded systems to high-performance computing platforms. According to the company, the architecture can deliver up to 3,500 tokens per second per watt when running large language models such as Llama 2 and Llama 3.2 with more than one billion parameters.

The customer will use the NeuPro-M architecture to accelerate AI inference directly on edge devices. The jointly developed processor is intended to support generative AI, multimodal AI, and agentic AI applications, enabling advanced AI capabilities to run efficiently on custom silicon. As part of the collaboration, CEVA worked closely with the customer to optimize neural networks specifically for the company’s AI workloads.

The deal also highlights a structural shift taking place across the semiconductor industry. Increasingly, companies whose core business has traditionally been software are investing in custom chip development, seeking tighter integration between hardware and software while reducing dependence on off-the-shelf processors. For IP suppliers such as CEVA, this emerging customer base represents a new growth opportunity beyond the company’s traditional markets.

Share via Whatsapp

Posted in: News

Posted in tags: ceva

Other articles you might like

D-Fend Adds Complementary Drone Technology to Motorola's Portfolio

Avnet ASIC to Manage Production of RAAAM Memory Technology at TSMC

Innoviz Appoints Former Rafael CEO Yoav Har-Even to Its Board of Directors