NOTE: This document is a free excerpt of a larger research report.

If you are interested in purchasing the full report, please contact

Omdia at insights@omdia.com.

EXECUTIVE SUMMARY
Market Report:
Artificial Intelligence for Edge
Devices
Edge-Based AI Chipsets and Accelerators for
Mobile Phones, Smart Speakers, HMDs,
Automotive, PCs/Tablets, Drones, Security
Cameras, Robots, Edge Servers, and Machine
Vision: Global Market Analysis and Forecasts

Publication Date: 2Q20

Anand Joshi

Aditya Kaul
Artificial Intelligence for Edge Devices

Table of contents
Executive summary................................................................................................................................................. 6
Introduction ........................................................................................................................................................ 6
Key findings......................................................................................................................................................... 7
Market forecasts................................................................................................................................................. 8
Market issues.......................................................................................................................................................... 9
Market ecosystem .............................................................................................................................................. 9
Market drivers .................................................................................................................................................. 10
Privacy........................................................................................................................................................... 10
Application-specific latency requirements ................................................................................................... 11
Network bandwidth ...................................................................................................................................... 11
Data security ................................................................................................................................................. 11
Industry regulations ...................................................................................................................................... 12

Cost ............................................................................................................................................................... 12
Market barriers ................................................................................................................................................. 13

Availability of power on edge devices .......................................................................................................... 13

Increasing complexity and variety of AI models ........................................................................................... 13
Neural network optimization tools on edge devices .................................................................................... 13
Model lifecycle management ....................................................................................................................... 13
Key markets and applications ........................................................................................................................... 14

Automotive ................................................................................................................................................... 14
Consumer and enterprise robots.................................................................................................................. 15
Drones .......................................................................................................................................................... 15
Head-mounted displays ................................................................................................................................ 16
Smart speakers ............................................................................................................................................. 17
Mobile phones .............................................................................................................................................. 17
PCs/tablets .................................................................................................................................................... 18
Security cameras........................................................................................................................................... 18
Machine vision .............................................................................................................................................. 19
Edge servers.................................................................................................................................................. 19
Others ........................................................................................................................................................... 20
Technology issues ................................................................................................................................................. 20
Wide range of edge devices ............................................................................................................................. 20
AI edge hardware: Key devices ......................................................................................................................... 21

© Omdia. All rights reserved. Unauthorized reproduction prohibited. Page 2

Artificial Intelligence for Edge Devices

Edge chipsets .................................................................................................................................................... 22

Power and performance ............................................................................................................................... 22
Emerging architectures................................................................................................................................. 23
AI edge chipset architectures ........................................................................................................................... 24
Central processing units ............................................................................................................................... 24
Graphics processing units ............................................................................................................................. 25
Field-programmable gate arrays .................................................................................................................. 26
Application-specific integrated circuits ........................................................................................................ 26
System-on-chip accelerators ........................................................................................................................ 27
Architecture comparison .............................................................................................................................. 28
Edge training ..................................................................................................................................................... 28
Edge inference and model optimization........................................................................................................... 29
Key industry players ............................................................................................................................................. 29
Alibaba .............................................................................................................................................................. 29
Apple................................................................................................................................................................. 31
Arm ................................................................................................................................................................... 32
Baidu ................................................................................................................................................................. 33
CEVA ................................................................................................................................................................. 34
Cognex .............................................................................................................................................................. 34
DJI ..................................................................................................................................................................... 35
Google .............................................................................................................................................................. 36
Horizon Robotics............................................................................................................................................... 36
Intel................................................................................................................................................................... 37
MediaTek .......................................................................................................................................................... 38
Mobileye (an Intel company)............................................................................................................................ 39
Movidius (an Intel company) ............................................................................................................................ 40
Mythic ............................................................................................................................................................... 40
NVIDIA .............................................................................................................................................................. 41
Qualcomm ........................................................................................................................................................ 42
Tesla .................................................................................................................................................................. 43
Xilinx ................................................................................................................................................................. 43
Other industry players ...................................................................................................................................... 44
Market forecasts................................................................................................................................................... 45
Forecast methodology ...................................................................................................................................... 45
Two different models ................................................................................................................................... 46
AI edge chipset forecasts by device category ................................................................................................... 48
AI edge chipset forecasts by chipset type ........................................................................................................ 53

© Omdia. All rights reserved. Unauthorized reproduction prohibited. Page 3

Artificial Intelligence for Edge Devices

AI edge device forecasts by application ........................................................................................................... 54

Additional breakdowns..................................................................................................................................... 55
AI edge chipset forecasts by power consumption ........................................................................................ 55
AI edge chipset forecasts by compute capacity............................................................................................ 56
AI edge chipset forecasts for training vs. inference ..................................................................................... 57
Conclusions ....................................................................................................................................................... 57

© Omdia. All rights reserved. Unauthorized reproduction prohibited. Page 4

Artificial Intelligence for Edge Devices

List of tables
Table 1: Power needs of chipsets in popular edge devices .................................................................................. 22
Table 2: Correlation between power and performance in AI chipsets ................................................................. 23
Table 3: Emerging architectures in AI edge chipsets ............................................................................................ 24
Table 4: Comparison of AI chipsets for edge applications .................................................................................... 28
Table 5: Other industry players ............................................................................................................................ 44
Table 6: AI edge device attach rates by application, world markets: 2019–25 .................................................... 54

List of figures
Figure 1: AI edge chipset revenue, world markets: 2019–25 ................................................................................. 8
Figure 2: AI edge device attach rates, world markets: 2019–25 ............................................................................ 8
Figure 3: Market ecosystem for edge chipset companies ...................................................................................... 9
Figure 4: Large spread of edge devices in terms of power and performance needs ........................................... 21
Figure 5: AI edge chipset revenue by device category, world markets: 2019–25 ................................................ 48
Figure 6: AI edge chipset shipments by device type, world markets: 2019–25 ................................................... 52
Figure 7: AI edge chipset year-over-year revenue growth rates, world markets: 2019–25 ................................. 52
Figure 8: AI edge chipset revenue by chipset type, world markets: 2019–25 ...................................................... 53
Figure 9: AI edge chipset shipments by chipset type, world markets: 2019–25 .................................................. 54
Figure 10: AI edge revenue by power consumption, world markets: 2019–25 ................................................... 55
Figure 11: AI edge revenue by compute capacity, world markets: 2019–25 ....................................................... 56
Figure 12: AI edge revenue for inference vs. training, world markets: 2019–25 ................................................. 57

© Omdia. All rights reserved. Unauthorized reproduction prohibited. Page 5

Artificial Intelligence for Edge Devices

Executive summary
Introduction
AI chipsets appeared on the horizon in 2016 after the industry began to grasp the vast computing
needs of neural network (NN) processing. Early artificial intelligence (AI) chipsets, led by general-
purpose graphics processing units, focused on the enterprise market and training workloads. They
provided high compute capacity and the ability to run state-of-the-art networks at the time.
However, the need for AI on edge devices was realized soon after and the race to design edge-
optimized chipsets began. Edge inference has emerged as a key workload in 2019–20, and many
companies have introduced their chipsets.
While AI edge processing today is primarily focused on inference on the device, the training is still
taking place within the data center. Many enhancements in hardware and software architectures
have been implemented to run compute-intensive NNs on the edge chipsets. New training
techniques such as federated learning have emerged that are poised to drive some part of training
to the edge device in the future.
Several different factors are driving AI processing to the edge device. Privacy, security, cost, latency,
and bandwidth are all being considered when evaluating data center versus edge processing needs.
Applications like autonomous driving and navigation have sub-millisecond latency requirements that
make edge processing mandatory. Other applications such as speech recognition on smart speakers
generate privacy concerns. Keeping AI processing on the edge device circumvents privacy concerns
while avoiding the bandwidth, latency, and cost concerns of cloud computing.
Depending on the AI application and device category, there are several hardware options available
for performing AI edge processing. The options include central processing units (CPUs), graphics
processing units (GPUs), application-specific integrated circuits (ASICs), field-programmable gate
arrays (FPGAs), and system-on-chip (SoC) accelerators. This report analyzes the technical and market
aspects of these chipsets in edge devices and outlines the advantages and limitations. It provides
forecasts of these categories through 2025 and a summary of market drivers and challenges.
Omdia focuses on a particular set of device categories in this report that are believed to be the
primary drivers for volume shipments and revenue. The device categories include automotive,
consumer and enterprise robots, drones, head-mounted displays (HMDs), smart speakers, mobile
phones, PCs/tablets, security cameras, machine vision, and edge servers. This report also details the
applications that will be run on AI edge devices, in turn driving hardware requirements. Several
software platforms are also emerging to help embedded application developers leverage AI in their
applications. This report does not cover the AI edge software stack in detail but does recognize its
importance in driving the development of AI edge processing.

© Omdia. All rights reserved. Unauthorized reproduction prohibited. Page 6

Artificial Intelligence for Edge Devices

Key findings
 The edge chipset market continues to expand rapidly. Shipments of devices needing
AI are increasing rapidly and so is the volume of AI processing. The need for AI on
edge devices has been recognized widely by the world, and the lack of compute
capacity is the key constraint to implementing many of the AI features on edge
devices. Despite the ramp-up, application developers have a lengthy wish list of
features that promises to keep the innovation going for a long time to come.
 This reports adds two new categories of devices to the forecast in addition to the
eight covered in the last Artificial Intelligence for Edge Devices report published in
2018: machine vision and edge servers. Both categories have received growing
interest in recent months that will continue due to the applicability of AI. Many
more devices such as digital televisions (DTVs) and home appliances are showing
signs of an embedded AI chipset usage, yet no consistent theme has appeared that
will drive their adoption in 2020.
 Innovation is rampant in the AI chipset market, with new architectures appearing
that compete with mainstream and established chipset architectures
(CPU/GPU/FPGA). These new architectures focus on a niche application area and
provide maximum benefits, unlike the general applicability of mainstream
architectures. At the time of writing this report in 2020, processing in memory (PIM)
is the only architecture that has started shipping.
 There is no clear winner in the chipset industry just yet, and it is unlikely that one
winner will take all. Different architectures provide different benefits for AI edge
applications, and it is common to mix one with another to get the best for AI
algorithm processing. Omdia believes that this trend will continue, with multiple
architectures merging on a chipset in the long run. The integrated chipsets will
become more widespread at the edge in comparison with the discrete chipsets.
 Power continues to be the key restriction for the edge chipset market and that
limits the amount of compute capacity available on the device. The ASIC chipset
vendors are exploiting this fact and providing the highest compute capacity within
the given power envelope for edge applications. As of 2020, several ASIC vendors
have started shipping ASICs for the edge market.
 Software infrastructure continues to be a bottleneck and much innovation is needed
on the toolchain, NN compression, and overall AI application lifecycle management
on the edge. Every chip company has chosen to stick with one or the other standard
for its NN data path optimization and that is leading to fragmentation. A lack of
standard optimized libraries at the chip level for NN acceleration also continues to
be an issue.
 Overall, Omdia remains bullish on the prospect of AI edge chipsets and expects the
upward trend in revenue to continue—despite the slowdown in the market from
the coronavirus (COVID-19) pandemic.

© Omdia. All rights reserved. Unauthorized reproduction prohibited. Page 7

Artificial Intelligence for Edge Devices

Market forecasts
Omdia forecasts that global AI edge chipset revenue will grow from $7.7bn in 2019 to $51.9bn by
2025 at a CAGR of 37.5%.
Figure 1: AI edge chipset revenue, world markets: 2019–25

Source: Omdia

The largest device market for AI edge chipsets in 2019 was mobile phones, which made up more
than 50% of revenue. Two other categories with billion-plus markets in 2019 were PCs/tablets and
automotive. Smart speakers were the fourth largest category in 2019 with $560m in AI edge chipset
sales. The mobile phones market segment is expected to continue leading in 2025; it is forecast to
account for over 50% of the global AI edge chipset market and reach a 100% attach rate.
Figure 2: AI edge device attach rates, world markets: 2019–25

Source: Omdia

© Omdia. All rights reserved. Unauthorized reproduction prohibited. Page 8

Artificial Intelligence for Edge Devices

Appendix
Methodology
This Omdia report provides a quantitative and qualitative assessment of the global opportunity for
AI edge processing across several consumer and enterprise device markets. The device categories
include automotive, consumer and enterprise robots, drones, HMDs, mobile phones, PCs/tablets,
security cameras, smart speakers, machine vision, and edge servers. Breakdown by chipset
architecture, power consumption, compute capacity, training versus inference, and application
attach rate are provided for each device category, with shipments and revenue forecast for the
period 2019–25. The given device categories are expected to be the primary drivers for AI edge
processing chipset volume and revenue in the near to medium term. As the global AI market evolves
and Omdia gains more clarity on other device categories and AI use cases, we expect to add to the
list of device categories in future report iterations.

Further reading
Artificial Intelligence Market Forecasts (4Q19)
Deep Learning Chipsets (1Q19)
Artificial Intelligence for Edge Devices (3Q18)

Author
Anand Joshi, Principal Analyst, Omdia
Aditya Kaul, Research Director, AI and Intelligent Automation, Omdia
askananalyst@omdia.com

Citation Policy
Request external citation and usage of Omdia research and data via citations@omdia.com.

Omdia Consulting
We hope that this analysis will help you make informed and imaginative business decisions. If you
have further requirements, Omdia’s consulting team may be able to help you. For more information
about Omdia’s consulting capabilities, please contact us directly at consulting@omdia.com.

Copyright notice and disclaimer

The Omdia research, data and information referenced herein (the “Omdia Materials”) are the
copyrighted property of Informa Tech and its subsidiaries or affiliates (together “Informa Tech”) and
represent data, research, opinions or viewpoints published by Informa Tech, and are not
representations of fact.

© Omdia. All rights reserved. Unauthorized reproduction prohibited. Page 58

Artificial Intelligence for Edge Devices

The Omdia Materials reflect information and opinions from the original publication date and not
from the date of this document. The information and opinions expressed in the Omdia Materials are
subject to change without notice and Informa Tech does not have any duty or responsibility to
update the Omdia Materials or this publication as a result.
Omdia Materials are delivered on an “as-is” and “as-available” basis. No representation or warranty,
express or implied, is made as to the fairness, accuracy, completeness or correctness of the
information, opinions and conclusions contained in Omdia Materials.
To the maximum extent permitted by law, Informa Tech and its affiliates, officers, directors,
employees and agents, disclaim any liability (including, without limitation, any liability arising from
fault or negligence) as to the accuracy or completeness or use of the Omdia Materials. Informa Tech
will not, under any circumstance whatsoever, be liable for any trading, investment, commercial or
other decisions based on or made in reliance of the Omdia Materials.

© Omdia. All rights reserved. Unauthorized reproduction prohibited. Page 59

CONTACT US
omdia.com
askananalyst@omdia.com