Edge Computing: Accelerating AI Deployment at the Network Perimeter

Interest in edge computing grows, yet many still grapple with its architecture. The same ambiguity exists around artificial intelligence, and the idea of migrating AI to the edge can seem even more perplexing.

Martin Davis, managing partner at DUNELM Associates, notes that edge AI is often dismissed as ‘just theory quoted in articles.’

Nevertheless, industrial and enterprise leaders can no longer ignore edge AI. Traditionally, compute‑intensive tasks like deep learning and computer vision resided in centralized data centers. However, the proliferation of high‑performance networking and on‑premises hardware now enables a shift from a purely cloud‑centric model to the edge, says consultant Chaitan Sharma. ‘It won’t happen overnight, but it is inevitable,’ he adds. Gartner forecasts that by 2025, 75% of enterprise data will be processed at the edge, while Grand View Research projects a 54% annual growth rate for the edge computing market through 2025.

Defining the precise location of edge computing can be ambiguous. The Open Glossary of Edge Computing describes it as the delivery of computing resources to the logical extremities of a network. Situated beyond conventional data centers and the cloud, the edge focuses on the network’s ‘last mile,’ bringing computation physically nearer to the devices and individuals generating data.

Industrial settings, such as factories and mines, face challenges when relying solely on cloud computing. A typical factory demands near‑perfect network reliability—99.9999% uptime—and millisecond‑level latency, while often restricting data transmission beyond its premises. Historically, such facilities have leaned on extensive cabling and vendor‑specific wired protocols, leading to fragmented technology ecosystems. Edge computing offers a pathway to unify these disparate components, as highlighted by Ovum's Market Radar: CSPs' Industrial IoT Strategies and Propositions.

Edge deployments that run independently of the cloud differ from purely on‑device compute models, where every data packet is processed locally. While on‑board computing can enable instant, critical decision‑making, Harald Remmert, senior director of research and innovation at Digi International, notes that the required hardware is expensive and often falls short when executing complex machine‑learning workloads.

In contrast, an AI‑enhanced edge solution within a factory can aggregate and contextualize data from numerous machines, pinpointing and forecasting issues that would otherwise cause downtime. Gal Ben‑Haim, head of architecture at Augury—a firm specializing in process‑industry machine learning—emphasizes that edge‑based inference unlocks large‑scale applications, even when ultra‑low latency is not mandatory.

However, deploying machine learning at the edge is not trivial; it demands mature models and novel deployment‑management strategies, Ben‑Haim explains.

Although certain edge use cases operate entirely independent of central cloud infrastructure, most analysts view edge computing as part of a distributed–centralized continuum. Rather than swinging away from traditional data centers, it offers a ‘truce,’ observes Gartner analyst Bob Gill during a 2018 webinar.

Bill Malik, vice president of infrastructure strategies at Trend Micro, cautions that edge computing will not supplant the cloud.

Daniel Newman, principal analyst at Futurum Research, agrees: only a handful of scenarios justify a fully self‑contained edge.

Typically, data flows bidirectionally between the edge and the cloud. The cloud excels at capturing macro trends—such as shifts in energy use or air quality—while edge computing delivers immediate, context‑specific insights, Malik notes.

Accenture regards edge computing as a natural extension of the cloud. ‘Edge is employed by many clients alongside cloud analytics and machine learning to unlock innovative business services,’ says Charles Nebolsky, managing director and network practice lead at Accenture Technology. A case in point is Accenture's Connected Mine program, which optimizes in‑pit operations for mining firms. Nebolsky explains that they have integrated edge computing into a mining client’s workflow, using high‑resolution video from drilling rigs to assess rock density in real time. This data enables the drill to auto‑adjust angle and speed, while also facilitating predictive maintenance. ‘The bandwidth needed for high‑density video streams cannot be feasibly transmitted to the cloud at the required frame rate without prohibitive costs,’ Nebolsky adds.

Volvo Trucks illustrates this bidirectional model through its telematics and remote diagnostics in newer models. The onboard computer flags abnormal parameters and generates trouble codes, which the telematics system then streams to Volvo’s Uptime Center. The center coordinates responses among repair shops, dealers, and customer‑service teams. While onboard diagnostics aid real‑time troubleshooting, the centralized cloud component empowers external stakeholders to ready for maintenance visits.

Bill Roberts, IoT director at SAS, observes that Volvo is moving toward a prevailing maturity model that integrates edge analytics, AI, and machine learning. ‘The next logical step is to empower the edge on trucks to sift actionable fault data,’ Roberts suggests. ‘This would liberate bandwidth, allowing more telematics to be collected for richer cloud‑based analytics. Those insights can then be deployed either at the edge or in the cloud, depending on the scenario.’

The Distributed Energy Resources Integration test bed exemplifies hybrid distributed‑cloud computing. It offers an alternative to conventional centralized AC grids that struggle to optimally harness distributed DC sources like solar or wind. The test bed deploys real‑time edge analytics across strategically placed hardware, unifying legacy equipment with centralized oversight while enabling autonomous, responsive operations. ‘The platform supports autonomous operation and edge analytics while feeding data and control signals to one or more control centers,’ explains Erik Felt, market development director of future grid at RTI, and Neil Puthuff, software integration engineer at RTI.

The advent of 5G has further stimulated interest in edge architectures that operate beyond conventional data centers. Although only a handful of companies have piloted 5G‑enabled edge projects, this landscape is poised to evolve as the network matures. Harald Remmert highlights that the benefits mirror those of cloud computing—just with reduced latency—and that the architecture is increasingly favored for machine‑learning workloads.