Avnet and The University of Hong Kong (HKU) have opened a new research and commercialization facility focused on the hardware side of the AI boom. The Emerging Microelectronics and Ubiquitous Systems Lab (EMUS Lab) will bring researchers, startups and industry partners together around edge AI, physical AI, robotics, high-performance computing and emerging microelectronics, with an emphasis on moving prototypes toward manufacturable products.
The AI industry has spent much of the past few years focused on increasingly capable models and the computing infrastructure required to train them. The next challenge is more physical: putting those models into robots, industrial machines, autonomous systems and other devices that must operate outside centralized cloud environments.
That is the problem Avnet and The University of Hong Kong (HKU) are targeting with the newly opened Emerging Microelectronics and Ubiquitous Systems Lab (EMUS Lab).
Located at the Data Technology Hub in Tseung Kwan O InnoPark, Hong Kong, the laboratory is designed as a bridge between academic research and commercial hardware development. Its focus areas include edge AI, physical AI, robotics, high-performance computing (HPC) and emerging microelectronics.
The distinction is important. Building an AI model is only one part of creating a deployable intelligent system. Developers also need suitable processors and accelerators, embedded software, thermal and power management, hardware-software integration, testing and manufacturing processes.
EMUS Lab is intended to provide some of those capabilities in one environment. According to Avnet and HKU, participating innovators can access GPU computing resources, engineering consultation, prototyping support, manufacturability assessments and supply-chain expertise.
That connects academic experimentation with a stage that can often be difficult for startups: converting a working prototype into hardware that can be manufactured consistently and at commercial scale.
Avnet is contributing its engineering, design-chain and global distribution capabilities, while its element14 business can support proof-of-concept development and prototyping. The model gives startups and researchers a potential route from early experimentation to component selection, engineering validation and eventual production.
The laboratory’s timing also reflects the industry’s changing AI architecture. Cloud infrastructure remains central to large language models and generative AI, but applications such as robotics, industrial automation and autonomous machines often require computation closer to the point where data is generated.
Edge AI can reduce latency and dependence on network connectivity, while physical AI introduces additional engineering requirements because AI systems must interact with sensors, actuators and unpredictable physical environments.
EMUS Lab is therefore positioned at the intersection of AI software and semiconductor engineering. Its work will depend on a broader ecosystem that includes chipmakers such as NVIDIA, AMD, Intel and Qualcomm, embedded computing companies, robotics developers and manufacturers.
The lab had already supported research using Avnet-sponsored GPU infrastructure before its official opening. Avnet and HKU said that research contributed to publications in Nature family journals and presentations at AI conferences including ICLR and ICML.
Startup commercialization is another major component. One of EMUS Lab’s first industry initiatives is the DfMA Launchpad for AI MMP Programme, a 12-month co-incubation program involving EMUS Lab, Avnet and ecosystem partners.
The program has selected 27 startup teams and is designed to provide technical mentorship, hardware-software co-design, system validation, manufacturability guidance and industry connections.
For AI startups, that support can address a less visible constraint in the current AI market. A promising model or algorithm does not necessarily translate into a viable physical product. Hardware costs, component availability, production tolerances, certification and supply-chain resilience can all determine whether an AI prototype becomes a commercial system.
Hong Kong’s location also gives the initiative access to the wider Greater Bay Area manufacturing ecosystem. If EMUS Lab can consistently connect research and startups with engineering and manufacturing capabilities, it could become a useful commercialization pathway for emerging AI hardware.
The broader significance is that AI infrastructure is expanding beyond data centers. As intelligence moves into machines and devices, semiconductor engineering, embedded computing and manufacturing expertise are becoming increasingly important parts of the AI development stack.
Market Landscape
The AI infrastructure market is expanding from cloud-based model training toward edge AI and physical AI, where computation increasingly happens inside or close to intelligent devices.
Robotics, industrial automation, healthcare equipment, autonomous systems and smart-city infrastructure all require combinations of AI accelerators, embedded processors, sensors, connectivity and specialized software.
Companies such as NVIDIA, AMD, Intel and Qualcomm are developing hardware platforms for these workloads, while distributors and engineering organizations are increasingly involved in helping customers move from component selection to complete systems.
The commercialization gap remains significant for startups. Research can demonstrate technical feasibility, but productization requires design-for-manufacturing expertise, reliable component sourcing, validation and scalable production.
EMUS Lab’s model addresses this gap by combining university research, startup incubation and industry engineering capabilities. Its success will ultimately depend on how effectively those resources translate into commercially viable AI hardware products.
Top Insights
- Avnet and HKU have opened EMUS Lab to connect AI hardware research with engineering, prototyping, manufacturing and global supply-chain capabilities.
- The facility focuses on edge AI, physical AI, robotics, HPC and emerging microelectronics as intelligence increasingly moves into physical systems.
- Avnet’s engineering and distribution capabilities complement HKU’s research infrastructure, creating a pathway from experimental technology toward commercial hardware.
- A 12-month DfMA Launchpad program has selected 27 AI startup teams for hardware-software co-design, validation and manufacturability support.
- The initiative highlights a growing AI infrastructure challenge: transforming promising algorithms and models into reliable, scalable and manufacturable intelligent devices.
Power Tomorrow’s Intelligence — Build It with TechEdgeAI
