嵌入式和边缘 AI 硬体：迈向量产

本报告的内容

嵌入式和边缘 AI 市场正在快速发展，这得益于处理技术、软体工具、模型可用性和开源计划的广泛进步。边缘 AI 机会涵盖整个嵌入式硬体领域，从超低功耗感测器到高性能近边缘基础设施，核心技术供应商不断寻求将其定位为新收入来源的核心，而这些创新也使其受益匪浅。目标边缘系统类型、覆盖范围、部署参数和最终用户产业的碎片化，使得任何一家公司都无法完全主导庞大的边缘 AI 运算市场。市场不断发展，新的解决方案和参与者不断涌入，从小型新创公司到已经在嵌入式领域深耕数十亿美元的大型企业，不一而足。

本报告探讨了商用嵌入式 AI 硬体市场中最具影响力的厂商和关键参与者，这些硬体产品销往第三方/外部 OEM 和 ODM、系统整合商、服务供应商以及企业/工业终端用户。市场正在不断扩张，边缘 AI 解决方案已部署到各个领域，包括航空航太和国防、汽车、通讯和网路、消费性电子、数位安全和监控、能源和电力、工业自动化和控制、医疗设备和医疗保健、零售自动化和数位看板、机器人以及运输和物流。本报告也详细分析了各种形态的商用嵌入式和边缘 AI 硬体解决方案的竞争格局。

本报告也研究了嵌入式和边缘 AI 硬体的总面积市场 (TAM)，提供了独立处理器、运算模组、主机板、单板电脑、整合系统和伺服器的市场规模估算和趋势分析。该报告还深入探讨了嵌入式 AI 硬体领域的关键应用驱动因素、生产规模扩大和转型。

本报告探讨的问题：

• 全球嵌入式与边缘 AI 硬体市场规模？哪些因素推动了其成长？
• 哪些产业在边缘 AI 的全面量产部署方面处于领先地位？
• 哪些区域因素和法规将影响边缘 AI 解决方案的发展？
• 嵌入式和边缘 AI 处理器、模组、主机板、单板电脑 (SBC)、整合系统/伺服器及其所包含的硬体加速器的领先供应商有哪些？
• 哪些形态的 AI 硬体加速器将达到最快的成长？
• 在您目前的工程专案中，哪些嵌入式堆迭组件和特性/功能与边缘 AI 的协同作用最强？
• 边缘系统现在和三年后需要多少边缘 AI 效能？

本报告中涉及的技术提供者

主要发现：

• 预计全球嵌入式和边缘 AI 硬体市场规模将从 2024 年的约 178 亿美元增长到 2029 年的 464 亿美元以上，复合年增长率为 21.1%。
• 在嵌入式分立 MPU、MCU 和 SoC 中，整合 NPU 是整个预测期内成长最快的 AI 加速器类型。
• 在嵌入式运算模组市场（CoM 和 SoM），整合 GPU 加速将占边缘 AI 出货量收入的近一半，当用作主机板、SBC 和整合系统/伺服器上的附加加速器时，采用率甚至更高。
• 嵌入式和边缘 AI 硬体的最大终端用户产业包括汽车、通讯和网路、消费性电子、工业自动化和控制、零售自动化和数位看板，医疗保健和机器人技术也在迅速崛起。
• 美洲地区在嵌入式/OEM AI 系统和伺服器市场中占有最大份额，而亚太地区由于其製造基地的集中，在组件处理器和计算模组方面处于领先地位。
• VDC Research 的 "工程师之声" 调查数据显示，嵌入式 AI 专案/产品与行动或电池供电的外形尺寸、增强的安全性（硬体和软体）、远端应用程式部署和管理以及机器视觉/物体检测等功能之间存在很强的相关性。

报告摘录

目录

该报告的内容

摘要整理

• 主要调查结果

全球市场:概要

• AI 功能的全系统堆迭考量
• 支援硬体所需的 AI 软体
• 整合与附加加速技术
  • 最近的M&A，策略性联盟

各地区的趋势·预测

各终端用户产业趋势·预测

• 汽车
• 通讯·网路
• 工业自动化·控制
• 机器人技术
• 新兴市场

AI加速器趋势·预测

• NPU
• 图形处理器
• 其他

终端用户的洞察

• 边缘 AI 与其他嵌入式功能和特性紧密相关。
• 嵌入式 AI 专案容易出现技术障碍和整合问题。
• 边缘 AI 需要一个全面的技术平台和强大的软体支援。
• AI 工作负载和加速器需求正在不断发展。

竞争情形

• 竞争情形概要
• 供应商洞察
  • 新兴企业

调查范围·调查手法

• AI，机器学习，神经网

关于作者

关于VDC Research

Inside this Report

The embedded and edge AI marketplace is rapidly evolving behind widespread advances in processing technology, software tooling, model availability, and open source initiatives. The market opportunity for edge AI spans the gamut of embedded hardware footprints from ultra-low-power sensors to high-performance near edge infrastructure, which all stand to benefit from the continuous innovation of core technology providers trying to center themselves within this new revenue stream. As with such fragmentation in target edge system types, footprints, deployment parameters, and vertical end users, no single company can completely take over the expansive edge AI computing market. It is under near-constant change with new solutions and players entering the market, including small upstarts as well as multi-billion-dollar organizations that are firmly entrenched in the embedded sector already.

Within this report is VDC's research-driven commentary about the largest influences and players in the burgeoning commercial market for embedded AI hardware sold to third-party/external OEMs/ODMs, system integrators, service providers, and enterprise/industrial end users. The market is increasingly broad with varying regional adoption and real-world deployments of edge AI solutions into domains such as aerospace and defense, automotive, communications and networking, consumer electronics, digital security and surveillance, energy and power, industrial automation and control, medical devices and healthcare, retail automation and digital signage, robotics, transportation and logistics, and other verticals. This research also includes an in-depth analysis of the competitive landscape for commercial embedded and edge AI hardware solutions of various form factors.

This market research report investigates the global merchant TAM for embedded and edge AI hardware, including dedicated sizing and trends analysis across the markets for discrete processors, computing modules, motherboards, single-board computers, and integrated systems and servers. It features a deep investigation into the key drivers for adoption, production scale up, and transformation in the embedded AI hardware space supported by highly granular market sizing, research interviews and ongoing conversations/projects with several dozen key ecosystem members and leading device suppliers, as well as interviews with and large-scale surveys of decision makers at product development organizations.

What Questions are Addressed?

• How large is the global market for embedded and edge AI hardware, and what is driving its growth?
• Which vertical markets are leading the charge for full-scale production rollouts of edge AI?
• What regional factors and regulations will impact the development of edge AI solutions?
• Who are the leading providers of embedded and edge AI processors, modules, boards, SBCs, and integrated systems/servers and the hardware acceleration within their offerings?
• Which types of AI hardware accelerators are expected to grow the fastest for different form factors?
• Which embedded stack components and capabilities/features have the most synergy with edge AI in engineering projects today?
• How much edge AI performance is needed in today's edge systems and three years from now?

Who Should Read this Report?

This report was written for those making critical decisions regarding product, market, channel, and competitive strategy and tactics. This report is intended for senior decision-makers who are developing, or are a part of the ecosystem of, IoT, embedded, and/or edge AI computing solutions, including:

Technology Providers in this Report:

Demand-side Research Overview

VDC launches numerous surveys of the IoT and embedded engineering ecosystem every year using an online survey platform. To support this research, VDC leverages its in-house panel of more than 30,000 individuals from various roles and industries across the world. Our global Voice of the Engineer survey recently captured insights from a total of 600 qualified respondents. This survey was used to inform our insight into key trends, preferences, and predictions within the engineering community.

Executive Summary

Edge AI is sweeping the embedded hardware market faster and with more impact than any individual technology revolution before - whether multicore processing, hardware-based security, M2M communications, IoT connectivity platforms, or local intelligent analytics. The edge AI revolution has transformed decision-making in product development and reshaped the very core of hardware architecture choice and design. It is a blanket trend that is driving rapid advances in many established and new embedded/OEM use cases, both to internal systems and services optimization as well as for external/customer applications. Edge AI is creating a rare inflection point in the embedded sector, driving transformation in a market long characterized by slow adoption and resistance to major architectural change due to entrenched design dependencies, stringent cost constraints, regulatory compliance demands, security requirements, and other barriers.

Mounting data volumes in the field, despite great advances in connectivity and embedded storage technologies through the past several years, continues to challenge data utilization strategies. The only way to achieve maximum autonomy and extract the maximum value from far edge/endpoint and operational data is to employ edge AI at several levels of end-to-end IoT and industrial solutions. However, each edge AI deployment, even within the same industries or environments, has its own unique considerations for power availability, network access, connected devices (i.e., cameras, sensors) and infrastructure, data types, etc. As a result, a wide variety of AI hardware solutions and integrated or attached workload acceleration technologies are in high demand within the IoT, embedded, and edge computing market to cater to the vast needs of manufacturers, integrators, and end users.

Key Findings:

• The global embedded and edge AI hardware market is projected to grow from ~$17.8B in 2024 to over $46.4B by 2029, reflecting a CAGR of 21.1%.
• Integrated NPUs are the fastest-growing AI accelerator type throughout the forecast period for discrete embedded MPUs, MCUs, and SoCs.
• In the embedded computing modules market (CoMs and SoMs), integrated GPU-driven acceleration accounts for nearly half of edge AI revenue shipments, with adoption rates even higher when used as attached accelerators for motherboards, SBCs, and integrated systems/servers.
• The largest vertical markets for embedded and edge AI hardware are automotive, communications and networking, consumer electronics, industrial automation and control, and retail automation and digital signage, with other segments such as medical/healthcare and robotics rapidly emerging.
• The Americas holds the largest share of the embedded/OEM AI systems and servers market, while Asia-Pacific's manufacturing concentration makes it the leading region for component processors and computing modules.
• VDC Research's Voice of the Engineer survey data shows a strong correlation between embedded AI projects/products and features such as mobile or battery-powered form factors, security enhancements (hardware and software), remote application deployment and management, and machine vision/object detection.

Report Excerpt

The embedded and edge AI hardware market is expanding rapidly, increasing its share within the broader markets for embedded processors, computing modules, motherboards, single-board computers, and integrated systems/servers. Through its implementation in both general-purpose hardware as well as via the growing portfolios of dedicated product families, embedded and edge AI capabilities are becoming table stakes for engineering organizations and end users worldwide and across industries. AI and machine learning technologies are unlocking significant value in the field, enabling the next data revolution with advancements in vision and imaging systems, sensor platforms, power and connectivity infrastructure, user interaction, and other domains.

Although AI and machine learning are not new to the embedded market, the surrounding ecosystem of hardware accelerators, software development platforms, open-source models, and AI frameworks has radically expanded their potential, from MCU-based sensors to many-core systems and servers. At the same time, the market opportunity for embedded and edge AI isn't confined to only greenfield deployments or new infrastructure rollouts. Given the long deployment lifetimes typical in many of the leading embedded hardware markets, there is strong demand for computing solutions to augment existing deployments or designs with attached or in-line edge AI capabilities - transforming "dumb" sensors, cameras, client devices, machinery, and other operational systems into intelligent platforms to achieve new efficiencies, controls, or levels of automation.

1 Note: Exhibit has been modified for the Executive Brief. Granular market forecast data is available within the report.

Scope & Methodology

AI, Machine Learning & Neural Networks

For the purposes of this report, VDC defines artificial intelligence as the ability of machines to mimic the cognitive functions of humans and carry out complex tasks in an autonomous or semi-autonomous fashion.

Machine learning (ML) is an application of AI that allows a system to improve its operation through the analysis and reimplementation of collected data without the need for explicit reprogramming. Production ML systems generally consist of a large bundle of code and tools surrounding the core ML decision-making code. Data collection, feature extraction, process management tools, data verification, machine resource management, configuration, analysis tools, monitoring, and server infrastructure are all vital components for production ML systems. The objective of ML is to build a tool and compute-based framework that allows machines to combine inputs to produce useful predictions on never-before-seen data. Machine learning models can be static or dynamic. A static model is trained offline, exactly once, then run on this system for a period of time before being updated. A dynamic model is trained on an ongoing basis, with data continuously entering the system and training the model as it operates.

A neural network (NN) further optimizes an ML system by adding layers of nodes to a model. A node consists of one or more weighted input connections, a transfer function that combines the inputs, and an output connection to the larger model. These layers of nodes transform the data in useful ways before it is passed to the final output node of the ML algorithm to use in its decision or classification. NNs allow for much more complex actions to be performed on incoming data and are useful for nonlinear systems where an ML algorithm must distinguish between many different possibilities (e.g., "Is this image an apple, a bear, an egg, or a duck?") rather than binary yes or no decisions. NNs are classified as deep learning methods or algorithms that attempt to learn broader data representations and patterns, as opposed to inflexible, task-specific algorithms.

Methodology

VDC produces its forecasts from a combination of primary and secondary data, including its own surveys and in- depth interviews with OEMs and industry executives; internal historical data sets and market models; public and privately disclosed numbers from vendors and customers, industry organizations, trade publications, and conferences; and our analysts' extensive base of industry experience and knowledge.

In this report, we size the market for merchant embedded and edge hardware supporting AI applications that are built for, and deployed on, OEM and embedded systems. It does not include the captive portion of the market, which includes hardware designed and built internally at OEMs solely for internal use or integration into their own larger systems or end products.

The embedded and edge AI hardware market sizing estimates and forecasts are based on a combination of:

• (1) VDC's longstanding expertise and ongoing coverage of the total market opportunity for merchant processors, boards, modules, and integrated systems;
• (2) frequently updated engineering survey data from product development organizations designing, building, deploying, and managing edge AI solutions;
• (3) in-depth interviews with decision-makers at OEMs and other embedded technology customers; and
• (4) regular discussions with embedded and edge technology providers and broader ecosystem participants.

Table of Contents

Inside this Report

Executive Summary

• Key Findings

Global Market Overview

• Full System Stack Considerations for AI Capabilities
• AI Software Needed for Hardware Enablement
• Integrated & Attached Acceleration Technology
  • Recent Acquisitions, Mergers & Strategic Partnerships

Regional Trends & Forecast

Vertical Market Trends & Forecast

• Automotive
• Communications & Networking
• Industrial Automation & Control
• Robotics
• Emerging Markets

AI Accelerator Trends & Forecast

• NPU
• GPU
• Others

End-user Insights

• Edge AI has Strong Connections with other Embedded Capabilities/Features
• Embedded AI Projects are Prone to Technical Obstacles & Integration Issues
• Edge AI Demands a Comprehensive Technology Platform Including Rich Software Support
• AI Workloads & Accelerator Needs are Evolving

Competitive Landscape

• Competitive Landscape Overview
• Vendor Insights
  • Emerging Players

Scope & Methodology

• AI, Machine Learning & Neural Networks

About the Authors

About VDC Research

List of Report Exhibits

• Exhibit 1: Global Shipments of Embedded & Edge AI Hardware Segmented by Form Factor, 2024-2029
• Exhibit 2: Typical AI Accelerator Performance vs. Thermal Design Power Segmented by Accelerator Type
• Exhibit 3: Typical Performance Needs for Common Edge AI Workloads
• Exhibit 4: Level of AI/ML Performance Expected for Target Systems to Run Typical End User/Customer Workloads
• Exhibit 5: Notable Acquisitions by Embedded & Edge AI Hardware Leaders Since Mid-2023
• Exhibit 6: Global Shipments of Embedded & Edge AI Hardware Segmented by Region
• Exhibit 7: Global Shipments of Embedded & Edge AI Processors Segmented by Vertical Market
• Exhibit 8: Global Shipments of Embedded & Edge AI Computing Modules Segmented by Vertical Market
• Exhibit 9: Global Shipments of Embedded & Edge AI Motherboards & SBCs Segmented by Vertical Market
• Exhibit 10 Global Shipments of Embedded & Edge AI Systems & Servers Segmented by Vertical Market
• Exhibit 11 Leading Embedded & Edge AI Applications Segmented by Vertical Market
• Exhibit 12 Global Shipments of Embedded & Edge AI Processors Segmented by AI Accelerator Type
• Exhibit 13 Global Shipments of Embedded & Edge AI Boards, Modules & Systems Segmented by AI Accelerator Type
• Exhibit 14 Capabilities/Features in Current IoT, Embedded & Edge AI and Non-AI Projects
• Exhibit 15 Leading Causes for Project Delays in Edge AI Engineering Projects
• Exhibit 16 Embedded Software Stack Components Required in Current IoT, Embedded & Edge AI and Non-AI Projects
• Exhibit 17 Types of Artificial Intelligence Workloads Used in Current Projects & Expected Three Years From Now
• Exhibit 18 Leading Embedded AI Processors & Motherboards Vendor Shares Segmented by Form Factor, 2024
• Exhibit 19 Leading Embedded AI SBCs, Integrated Systems & Servers Vendor Shares Segmented by Form Factor, 2024

List of Market Data Exhibits

Embedded AI Processors

• Exhibit 1: Embedded AI Processors, Discrete Processor Type, Dollar Volume Shipments, 2024-2029
• Exhibit 2: Embedded AI Processors, Vertical Market, Dollar Volume Shipments, 2024-2029
• Exhibit 3: Embedded AI Processors, Region, Dollar Volume Shipments, 2024-2029
• Exhibit 4: Embedded AI Processors, Customer Class, Dollar Volume Shipments, 2024-2029
• Exhibit 5: Embedded AI MPU Processors, CPU Architecture, Dollar Volume Shipments, 2024-2029
• Exhibit 6: Embedded AI MCU Processors, CPU Architecture, Dollar Volume Shipments, 2024-2029
• Exhibit 7: Embedded AI SoC Processors, CPU Architecture, Dollar Volume Shipments, 2024-2029
• Exhibit 8: Embedded AI MPU Processors, Integrated AI Accelerator Type, Dollar Volume Shipments, 2024-2029
• Exhibit 9: Embedded AI MCU Processors, Integrated AI Accelerator Type, Dollar Volume Shipments, 2024-2029
• Exhibit 10: Embedded AI SoC Processors, Integrated AI Accelerator Type, Dollar Volume Shipments, 2024-2029
• Exhibit 11: Embedded AI MPU Processors, Integrated AI Accelerator Provider, Dollar Volume Shipments, 2024-2029
• Exhibit 12: Embedded AI MCU Processors, Integrated AI Accelerator Provider, Dollar Volume Shipments, 2024-2029
• Exhibit 13: Embedded AI SoC Processors, Integrated AI Accelerator Provider, Dollar Volume Shipments, 2024-2029
• Exhibit 14: Embedded AI Processors, Global Supplier Shares, Dollar Volume Shipments, 2024

Embedded AI Computing Modules

• Exhibit 1: Embedded AI Computing Modules, Form Factor, Dollar Volume Shipments, 2024-2029
• Exhibit 2: Embedded AI Computing Modules, Vertical Market, Dollar Volume Shipments, 2024-2029
• Exhibit 3: Embedded AI Computing Modules, Region, Dollar Volume Shipments, 2024-2029
• Exhibit 4: Embedded AI Computing Modules, Customer Class, Dollar Volume Shipments, 2024-2029
• Exhibit 5: Embedded AI Computing Modules, CPU Architecture, Dollar Volume Shipments, 2024-2029
• Exhibit 6: Embedded AI Computing Modules, Integrated AI Accelerator Type, Dollar Volume Shipments, 2024-2029
• Exhibit 7: Embedded AI Computing Modules, Integrated AI Accelerator Provider, Dollar Volume Shipments, 2024-2029
• Exhibit 8: Embedded AI Computer-on-Modules, Global Supplier Shares, Dollar Volume Shipments, 2024
• Exhibit 9: Embedded AI System-on-Modules, Global Supplier Shares, Dollar Volume Shipments, 2024

Embedded AI Motherboards & SBCs

• Exhibit 1: Embedded AI Motherboards & Single-board Computers, Form Factor, Dollar Volume Shipments, 2024-2029
• Exhibit 2: Embedded AI Motherboards & Single-board Computers, Vertical Market, Dollar Volume Shipments, 2024-2029
• Exhibit 3: Embedded AI Motherboards & Single-board Computers, Region, Dollar Volume Shipments, 2024-2029
• Exhibit 4: Embedded AI Motherboards & Single-board Computers, Customer Class, Dollar Volume Shipments, 2024-2029
• Exhibit 5: Embedded AI Motherboards & Single-board Computers, CPU Architecture, Dollar Volume Shipments, 2024-2029
• Exhibit 6: Embedded AI Motherboards & Single-board Computers, Integrated/Attached AI Accelerator Type, Dollar Volume Shipments, 2024-2029
• Exhibit 7: Embedded AI Motherboards & Single-board Computers, Integrated/Attached AI Accelerator Provider, Dollar Volume Shipments, 2024-2029
• Exhibit 8: Embedded AI Motherboards, Global Supplier Shares, Dollar Volume Shipments, 2024
• Exhibit 9: Embedded AI Single-board Computers, Global Supplier Shares, Dollar Volume Shipments, 2024

Embedded AI Systems & Servers

• Exhibit 1: Embedded AI Systems & Servers, Form Factor, Dollar Volume Shipments, 2024-2029
• Exhibit 2: Embedded AI Systems & Servers, Vertical Market, Dollar Volume Shipments, 2024-2029
• Exhibit 3: Embedded AI Box PCs & Other Active Backplane Systems, Vertical Market, Dollar Volume Shipments, 2024-2029
• Exhibit 4: Embedded AI Rackmount & Blade Servers, Vertical Market, Dollar Volume Shipments, 2024-2029
• Exhibit 5: Embedded AI Standalone Systems & Mobile Platforms, Vertical Market, Dollar Volume Shipments, 2024-2029
• Exhibit 6: Embedded AI Systems & Servers, Region, Dollar Volume Shipments, 2024-2029
• Exhibit 7: Embedded AI Systems & Servers, Customer Class, Dollar Volume Shipments, 2024-2029
• Exhibit 8: Embedded AI Systems & Servers, CPU Architecture, Dollar Volume Shipments, 2024-2029
• Exhibit 9: Embedded AI Systems & Servers, Integrated/Attached AI Accelerator Type, Dollar Volume Shipments, 2024-2029
• Exhibit 10: Embedded AI Box PCs & Other Active Backplane Systems, Integrated/Attached AI Accelerator Type, Dollar Volume Shipments, 2024-2029
• Exhibit 11: Embedded AI Rackmount & Blade Servers, Integrated/Attached AI Accelerator Type, Dollar Volume Shipments, 2024-2029
• Exhibit 12: Embedded AI Standalone Systems & Mobile Platforms, Integrated/Attached AI Accelerator Type, Dollar Volume Shipments, 2024-2029
• Exhibit 13: Embedded AI Systems & Servers, Integrated/Attached AI Accelerator Provider, Dollar Volume Shipments, 2024- 2029
• Exhibit 14: Embedded AI Box PCs & Other Active Backplane Systems, Integrated/Attached AI Accelerator Provider, Dollar Volume Shipments, 2024-2029
• Exhibit 15: Embedded AI Rackmount & Blade Servers, Integrated/Attached AI Accelerator Provider, Dollar Volume Shipments, 2024-2029
• Exhibit 16: Embedded AI Standalone Systems & Mobile Platforms, Integrated/Attached AI Accelerator Provider, Dollar Volume Shipments, 2024-2029
• Exhibit 17: Embedded AI Systems & Servers, Global Supplier Shares, Dollar Volume Shipments, 2024
• Exhibit 18: Embedded AI Box PCs & Other Active Backplane Systems, Global Supplier Shares, Dollar Volume Shipments, 2024
• Exhibit 19: Embedded AI Rackmount & Blade Servers, Global Supplier Shares, Dollar Volume Shipments, 2024
• Exhibit 20: Embedded AI Standalone & Mobile Platforms, Global Supplier Shares, Dollar Volume Shipments, 2024

IoT & Embedded Engineering Survey (Partial)

• Exhibit 30: Capabilities/Features of Current Project
• Exhibit 60: Primary Method for Updating Devices Once Deployed in Field
• Exhibit 70: Discrete/Standalone Processing Unit(s) Used on Current Project
• Exhibit 140: Types of Artificial Intelligence Workloads Used in Current Projects
• Exhibit 142: Types of Machine Learning Used for AI in Current Projects
• Exhibit 147: Target Architecture(s) Used for Inferencing In Current Project
• Exhibit 151: Current Level of AI/ML Performance Expected for Target Systems to Run Typical End User/Customer Workloads