Heterogeneous computing and the future of AI

Heterogeneous computing is quietly redefining how AI appears in everyday devices, bringing intelligence closer to where people actually use it rather than keeping it locked in remote cloud systems.

As AI expands into physical products, success depends less on raw performance and more on how efficiently CPUs, GPUs, and NPUs work together. Heterogeneous computing manifests itself in faster AI, better battery life and intelligence that responds to context without constantly calling attention to itself, according to Chris Bergey (photo), executive vice president of the Edge AI business unit at Arm.

“Our partnership with Nvidia has been extremely strong and it really comes from, in this world of AI, the blending of compute and acceleration – and you can see that on any of these devices,” Bergey said. “In these cell phones that have an Arm CPU, an Arm GPU and also an NPU, it’s really about heterogeneous computing and how we use that to further accelerate what’s possible in AI.”

Bergey spoke with Savannah Peterson has CES 2026during an exclusive broadcast on theCUBE, SiliconANGLE Media’s live streaming studio. They discussed how heterogeneous computing enables AI to operate effectively at the edge, transforming devices, wearables, security, and the developer experience.

How heterogeneous computing is changing AI at the edge

Heterogeneous computing is becoming essential as AI workloads become increasingly complex and interactive. Rather than forcing each task through a single processor, workloads are distributed among specialized components, each doing what it does best. This approach allows for real-time responsiveness while controlling energy consumption, Bergey explained.

“When these three things work together, you can maximize your battery life in a different way,” he said. “You can really optimize every feature in action that happens on this phone rather than just dumping on the same thing all at once.”

This architectural change is also changing the way developers think about building AI-powered applications. AI is no longer something fixed on a product after the fact; this becomes an essential design consideration. As intelligence moves closer to the user, software must be written to take advantage of multiple computational paths working in concert.

“One of the biggest challenges is actually programming and programming AI,” Bergey said. “How do you program GPUs? How do you program NPUs? That’s something we’re focusing on a lot.”

Beyond smartphones and consumer electronics, heterogeneous computing opens the door to entirely new classes of devices. Wearable devices, in particular, are emerging as a testing ground for what happens when AI processing lives directly on the body. These systems collect richer data while maintaining privacy and responsiveness by processing information locally, Bergey noted.

“AI is enabling a sort of renaissance of wearable devices,” he said. “We’re now starting to interact with these devices. It gives you capabilities in these wearable devices that weren’t possible.”

The long-term impact extends beyond convenience and touches areas such as health, safety and quality of life. As devices gain the ability to process biometric and contextual data in real time, early detection and personalized insights become realistic outcomes rather than distant promises. The main challenge is to ensure that this intelligence is combined with trust and security at the architectural level.

“It’s super powerful, but it also has to meet all the safety requirements,” Bergey added. “We look at security just at the level of its distinction and all kinds of things that you can do to really be able to give that capability at the base level of really complete trust. »

Here’s the full video interview, part of SiliconANGLE and theCUBE’s coverage on CES 2026:

Photo: SiliconANGLE