One of the biggest new demand drivers for AI hardware is Edge AI - essential for robotics and autonomous driving, where split-second responses are vital. Unlike cloud AI, it runs inferences locally near devices and sensors, cutting reliance on servers. “We are reaching a tipping point where the capabilities previously only available in server-class hardware are now being integrated into small, fanless devices. For instance, we are working on building the world’s smallest audio system-on-module,” says Sameer Sharna, AVP of MediaTek's IoT unit.

That compact power is spurring new types of devices. LG Electronics, for example, is embedding ‘Affectionate Intelligence’ in home appliances, which “empathizes” with users and delivers “hyper-personalized lifestyles where devices and services harmonize to transcend mere physical locations,” according to Head of Global Customer Strategy Division Alexio Rhee. Meanwhile, Arm is putting SLMs in cars: “One of our partners is working on replacing the user manual with Gen AI… you can talk to your car: ‘What does this warning light mean?’” says Suraj Gajendra, VP of Automotive. NVIDIA has launched DGX Spark, a desktop supercomputer priced under $500. “And Sam Altman is reportedly investing $6 billion with Jony Ive to design a secretive, AI-powered personal device,” says Frankwell Jyh-Ming Lin, Founder and CEO, Andes Technology. “Everyone is exploring AI-powered devices that run LLMs locally. If they crack it, the market will explode.”

Khein-Seng Pua, CEO of the Taiwan-based Phison Electronics, says Asia outstrips the world in developing Edge AI: “Taiwan and Malaysia are leading. India is booming… The U.S. is slow. Europe is nearly inactive. Developers there have been conditioned to build on cloud infrastructure. In Asia, we have built for the edge.” 

But in Europe, others would push back on that claim. “Edge AI is the standout innovation for 2025,” says Prof. Dr.-Ing. Axel Sikora, chairman of the Embedded World Conference in Germany. France-based STMicroelectronics has already made it a priority: “We expect that by 2030, close to 10 percent of microcontrollers will use AI from literally zero today. Edge AI will be a huge volume business, as opposed to high-end, power-hungry AI chips,” says Aidoune.