Generative AI is changing the way enterprises think about technology infrastructure. Global market intelligence firm International Data Corporation’s (IDC) report titled ‘The Edge Evolution: Powering Success from Core to Edge’, developed in collaboration with cloud service provider Akamai, highlights how legacy systems are falling short as AI transitions from pilots to production. Enterprises are now looking beyond centralised data centres, choosing to move workloads closer to where data is created.
The report forecasts that public cloud-based services at the edge will grow at a 17% CAGR in the Asia-Pacific region, excluding Japan, reaching $29 billion by 2028. At the heart of this shift is generative AI’s demand for scalability and performance, which is pushing enterprises to invest in infrastructure that spans from core to edge.
Cloud to Edge: A Distributed Fabric
The report underlines that 96% of enterprises in the region adopting generative AI will rely on public cloud infrastructure-as-a-service (IaaS) for training and inferencing.
Sanchit Vir Gogia, chief analyst and CEO at Greyhound Research, told AIM, “The old arguments about choosing between cloud or edge simply do not hold anymore. Models may still be trained in vast cloud data centres, but once they are put to work, the story changes.”
He explained that for daily use, speed, privacy and sovereignty are crucial, necessitating inference closer to data creation or service delivery.
Speaking to AIM, Mitesh Jain, regional VP at Akamai India, explained, “Training workloads, which are resource-intensive and require massive computing power, are best suited for public cloud IaaS.”
However, he pointed out that keeping all workloads in the cloud can become costly due to continuous data transfer and storage needs. “This is where the edge becomes critical.”
Edge deployments, he believes, are ideal for inferencing workloads and real-time GenAI applications like IoT monitoring, fraud detection and personalised customer engagement. This is due to their need for low latency, localised compliance and accelerated decision-making.
Rushikesh Jadhav, CTO of ESDS Software Solution Limited, agreed, noting that organisations must weigh “latency, data sovereignty, scalability and compliance” before deciding workload placement.
In his view, public cloud will remain the default for compute-intensive training.
“Conversely, inference workload that demands real-time decision-making jobs like financial services’ fraud prevention, manufacturing’s predictive maintenance or video analytics for smart cities are best to be executed at the edge,” he said. “This provides low-latency performance, lower bandwidth cost, and local adherence to data privacy regulation.”
Industry Use Cases Driving Adoption
Examples of AI adoption are already visible across industries. Banks are experimenting with AI-enabled mainframes to facilitate real-time transactions. Factories are embedding intelligence into production lines to identify defects, while hospitals are running assistants on-site to ensure patient data remains private. Gogia described these not as mere trials on the sidelines but as “fundamental shifts in design”.
Jain from Akamai India highlighted that predictive AI is leading adoption across the Asia-Pacific region.
“Enterprises are scaling predictive workloads to power real-time insights, fraud detection and operational optimisation. This momentum is being driven by the need to process data closer to its source, reducing latency, lowering connectivity costs and ensuring compliance,” he said.
He further pointed out that in India, 38% of enterprises prioritise interpretive AI, highlighting the country’s specific requirement to process large amounts of edge data.
Meanwhile, Jadhav noted predictive AI’s strong traction in manufacturing and utilities, thanks to its tangible returns in reducing downtime and improving efficiency.
Brijesh Patel, founder and CTO of SNDK Corp, reinforced the business case and said to AIM, “Workloads that demand immediate insights, local decision-making or processing of sensitive data should be deployed closer to the edge, where latency is minimised, data privacy is better managed and operational continuity is ensured.”
Lessons and Challenges Ahead
The move to edge AI is not without pitfalls. “We have seen organisations invest heavily in cloud-based inference engines, only to abandon them when response times proved too slow for industrial control. Others discovered that sprinkling GPUs across remote sites created more cost than value when traffic was sparse,” Gogia explained.
He argued that success depends on calibrating workloads carefully across cloud and edge.
Enterprises must prepare for growing challenges around costs and infrastructure, Jain cautioned. “Rising compute costs, energy demands and hardware availability are among the most pressing concerns,” he said. He further stressed that “to overcome these challenges, enterprises must modernise their digital backbone with edge-optimised architectures, embrace interoperable multicloud strategies and adopt cost management practices that balance performance with scalability”.
Patel noted that high-performance GPUs at the edge can be costly, energy-intensive, and necessitate effective cooling and upkeep. Jadhav echoed the warning, highlighting that managing a distributed fleet of edge devices necessitates a new operational model. Most IT teams, accustomed to centralised cloud management, are currently unprepared for this shift.
Despite the hurdles, momentum is clear. Gogia summed it up, saying that generative AI’s future lies in distributed intelligence, not cloud centralisation. While the cloud remains vital, edge computing will be key to performance and trust.
Enterprises that orchestrate cloud and edge as one fabric, rather than choosing one over the other, are most likely to lead in an AI-first world.
The post GenAI Infrastructure Will Drive Public Cloud Models at the Edge appeared first on Analytics India Magazine.
