AI/ML Must Migrate to the Edge and Endpoints; Here’s Why

The emergence of the cloud, the development of connected embedded systems, and the expanding reach of smartphones, tablets and PCs have fueled a revolution in the creation and consumption of data. IDC expects that by 2025 the amount of data created globally will grow by a factor of 10 times and reach 163 zettabytes (a trillion gigabytes). During the same time period, IDC expects that endpoint devices will be responsible for creating more than half of all that data, and the fastest growing areas in the endpoint segment will be embedded and IoT devices. Earlier this year, IDC estimated that there will be 25 billion devices connected to the internet by 2022. Organizations have begun to harness artificial intelligence (AI) and machine learning (ML), especially deep-learning methods, to deal with this data deluge. Much of this AI/ML development work has occurred on servers and largely in data centers—a.k.a. the Cloud. As large as the Cloud infrastructure has become over the past 10 years, IDC believes there’s an even larger and broader OT market, which is poised to radically change traditional industries with the adoption of AI. However, there isn't enough computing power in all of today’s cloud data centers to transform the mountain of raw data being created by IoT endpoint devices into useful, valuable and actionable insights. This processing challenge will only grow in the coming years.

Taken collectively, IDC estimates that the annual revenue for these OT industries already exceeds IT market annual revenues by 2.5x and that the OT market also exceeds the IT market in terms of installed unit volume and annual shipments. The majority of these existing OT devices are fixed-function endpoints, which are dedicated to process monitoring and control, rooted in safety, and contained in large established vertical industries. In 2017, IDC estimates that the OT industry achieved $2.6 trillion in revenues and is expected to grow at CAGR of 4%, reaching $3.1 trillion in 2022.

The three key OT Market Segments are Consumer, Smart Factory and Infrastructure. OT systems in these three segments range from industrial tools and PLCs (programmable logic controllers), wearables, home automation, video surveillance, smart meters, voice assistant speakers and earbuds, consumer and industrial robotics, digital signage, automotive ECUs (engine control units), and ADAS subsystems are all increasing in complexity and intelligence. The use of artificial intelligence to derive value from the real-time data being created and aggregated at each endpoint is growing rapidly. That is the vision and the reality which make the OT industry a prime candidate for disruption and potential growth for technology suppliers.

Figure 1: IDC’s revenue estimate for 2017 and forecast for 2022, for the Consumer, Smart Factory and Infrastructure OT market segments.

In terms of annual revenue, the OT market is already 2.5x larger than the IT market, and will grow collectively at a 4%CAGR.
Source: IDC, 2018.

The inevitable introduction of more and more autonomous systems—into factories, infrastructure and homes—will drive AI/ML inferencing directly into endpoint systems. IDC expects that AI/ML algorithms, embedded in a wide range of OT devices, will completely redefine markets, industries and revenue streams as they reshape the way we extract value from data. The penetration of embedded AI/ML technology will be limited only by how quickly AI algorithms can be created and trained and how efficiently systems can deliver the massive amount of computing power needed to address the growingly complex computational requirements at the endpoint. This demand for additional computing power to implement AI/ML algorithms on the edge and in endpoints already drives the development of next-generation semiconductors, hardware, software, and AI algorithms that will create disruptive change across all OT markets.