Industrial sites facing a shrinking workforce are being cited as the field where distributed AI delivers the greatest value. As ageing reduces the number of people available to work, robots and equipment that can decide and move on their own must fill the gap on the ground. Distributed AI links multiple computing environments, including cloud, edge, on-device and on-premises, and processes each AI task at the most suitable location depending on latency, privacy protection, energy efficiency and cost.

Global technology policy consulting firm Access Partnership said in a report on Tuesday that autonomous industrial operations could generate more than $23 billion a year in economic value in 2035. It said about 40 percent of South Korea's population is expected to be elderly in 2050. A population cliff and structural labour shortages are also items the report cited as challenges for South Korea. Workforce gaps are already materialising at industrial sites. The semiconductor industry alone is estimated to face a shortfall of about 18 percent of the workers it will need by 2031.

For machines to take over jobs that companies cannot fill, they must be able to make decisions immediately on site without waiting for human instructions. That requirement has become more demanding as AI expands beyond data centres into physical AI, such as robots, cars and manufacturing equipment that make decisions and operate in the real world.

Industrial robots and equipment directly tied to safety must make real-time decisions and respond within a short time, and the approach of waiting for a central data centre response each time has limits, the analysis said. It also said production data could be exposed to cybersecurity risks in the process of passing through external servers.

Access Partnership said these problems can be addressed by splitting computing locations through distributed AI. It links cloud, edge, on-device and on-premises environments and processes each task where it best fits based on latency, privacy protection, energy efficiency and cost. Large-scale computation is handled by the cloud, while tasks that require immediate decisions are processed at an edge site close to the field or on the device itself.

Devices that use AI decision-making on site to fill workforce gaps are concentrated in industrial settings. Examples include industrial and construction robots, humanoid robots for hazardous work, autonomous inspection robots and predictive maintenance. The structure is for machines to take over work that is dangerous for people to enter or difficult to staff. Applying distributed AI to these devices can help address labour shortages while reducing equipment downtime and workplace safety accidents. The report calculated economic value based on additional revenue and cost savings companies gain from adopting AI.

Care settings also face a lack of staff. Access Partnership highlighted distributed AI as a way to respond to the population cliff. Using distributed AI for mobility support, fall detection, safety monitoring and conversational companion robots can help older adults live independently and ease shortages of care workers and cost burdens. It also cited the point that health conditions or biometric information do not need to be continuously sent to external servers. The report said the companion and assistive robot field is also expected to generate more than $5.9 billion a year in value in 2035.

Distributed AI that fills workforce gaps in the field is also built into devices South Korean companies sell overseas. South Korea's trade dependence is about 91 percent of GDP, meaning changes in export competitiveness have a big impact on the overall economy. The report said expanding on-device and distributed AI functions in smartphones, smart home appliances, home and service robots and extended reality devices could increase product value added and global market share and also create new product categories. The value in this field is more than $21 billion a year in 2035. It cited finished-goods companies such as Samsung Electronics, LG Electronics and Hyundai Motor, and foundry, memory and on-device neural processing unit capabilities as the foundation supporting this. It added that whether the current structure can handle the computing these devices will use is a separate issue.

The report said a cloud-centred structure for running AI, as it is now, would make it difficult to support that value. If the current trend in data centre investment continues, about 86 percent of the AI inference processing capacity needed in 2030 could be lacking, it forecast. The value of distributed AI across the three areas of industrial sites, care and export devices was estimated at more than $50 billion a year in 2035.

The report presented additional investment, standards cooperation with major countries, supplier diversity to reduce dependence on specific providers, and the establishment of sector-by-sector safety and performance standards as tasks to address this.