Software and AI platform companies are increasingly designing custom chips to run artificial intelligence directly on devices, rather than sending data to cloud servers. This shift reflects a fundamental change in how the tech industry approaches AI deployment, with companies recognizing that owning both the operating system and hardware allows them to optimize performance, power efficiency, and user experience in ways that off-the-shelf processors cannot match.

What's Driving This Move Away From Cloud AI?

The decision by a major U.S. software and AI platform company to license Ceva's NeuPro-M technology for a custom AI silicon program represents what executives describe as "one of the most strategically significant AI licensing agreements" in recent history. The underlying reason is straightforward: as AI workloads become more distributed across cloud and edge devices, companies that control both the software and hardware stack gain a decisive advantage.

When a company owns the operating system and designs its own chips, it can tightly integrate the two, enabling greater performance and power efficiency than generic processors can deliver. This matters especially for portable devices where battery life and heat management are critical constraints. Amir Panush, Chief Executive Officer of Ceva, explained the strategic importance of this shift.

"The decision by one of the industry's leading software and AI platform companies to build custom AI silicon on NeuPro-M reflects a broader shift toward AI-first computing architectures. Intelligent devices are increasingly expected to sense, reason and act locally, driving demand for AI acceleration that delivers high performance within strict power and thermal constraints," said Panush.

Amir Panush, Chief Executive Officer of Ceva

The shift represents a fundamental change in computing architecture. Just as CPUs (central processing units) defined general-purpose computing and GPUs (graphics processing units) accelerated graphics and parallel workloads, AI acceleration is emerging as a third foundational layer of the computing stack. This means NPUs, or neural processing units, are becoming core architectural elements rather than optional add-ons.

How Are Companies Using Custom AI Chips for On-Device Inference?

The NeuPro-M architecture selected by this major platform company enables efficient execution of several types of AI workloads directly on devices:

• Generative AI: Running large language models and text generation tasks locally without sending data to remote servers.
• Multimodal AI: Processing combinations of text, images, and audio simultaneously on the device itself.
• Agentic AI: Supporting emerging autonomous AI agents that can make decisions and take actions in real time on edge devices.
• Machine Learning Applications: Executing various ML workloads while respecting strict power, area, and thermal constraints of portable devices.

The key advantage is that all this processing happens on the device itself, not in a distant data center. This means faster response times, better privacy (data doesn't leave your device), and reduced dependence on internet connectivity. For companies building the next generation of intelligent computing devices, this represents a fundamental shift in how they architect their products.

Why Does This Matter for the Future of AI?

This licensing deal signals that the era of cloud-dependent AI is giving way to a more distributed model where intelligence lives on the device itself. More than 2 billion devices incorporating Ceva technologies already ship annually across consumer electronics, automotive, industrial IoT, and mobile markets, with momentum continuing to expand. As this trend accelerates, expect to see AI capabilities that were previously only available through cloud services now running directly on your phone, laptop, or smart device.

The practical implications are significant. Devices can respond faster because they don't need to wait for a round trip to a server. Users get better privacy because sensitive data stays local. Companies gain more control over their AI stack, from the silicon up through the software. And as thermal and power constraints become less of a bottleneck, manufacturers can pack more AI capability into thinner, lighter, longer-lasting devices.

For consumers, this means the next generation of intelligent devices will be smarter, faster, and more responsive than anything currently available. For tech companies, it means the competitive advantage increasingly goes to those who can design custom silicon optimized for their specific AI workloads and user experience goals.