生成式人工智慧與基於代理的人工智慧市場：2026-2032年

這份 283 頁的報告對 2032 年前的生成式人工智慧和基於代理的人工智慧市場進行了定量分析，包括供應商市場佔有率、企業用例、採用情況和關鍵趨勢。

關鍵數據摘要：

• 2025年市場規模：3,080億美元
• 2032年市場預測：4.4兆美元
• 主要公司：AMD、AWS、埃森哲、阿里雲、Anthropic、凱捷、Google、英特爾、微軟、Mistral、英偉達、OpenAI

生成式人工智慧和基於代理的人工智慧市場：市場規模、趨勢和主要參與者（2026-2032 年）

生成式人工智慧和基於代理的人工智慧市場概述

根據物聯網分析公司（IoT Analytics）的研究，2025年全球生成式人工智慧和基於代理的人工智慧市場規模達3,080億美元。基於企業採用指標和技術投資分配，預計到2032年，該市場規模將達到4.4兆美元，2025年至2032年的複合年成長率（CAGR）為46%。企業採用生成式人工智慧技術在2025年已十分普遍，88%的受訪企業都在使用這項技術。推動市場擴張的因素之一是營運模式的轉變，即從人工指導轉向在人工監督下執行多階段流程的自主代理工作流程。

本報告概述了構成人工智慧技術棧的五個主要部分。

• AI加速器：用於模型訓練和推理的GPU、TPU、LPU和客製化ASIC。
• AI雲端平台：租賃運算基礎設施、模型開發環境、運行時運維、代理編配層
• 人工智慧服務：策略諮詢、系統整合、客製化代理設計、託管服務
• 人工智慧應用：終端用戶軟體，例如聊天機器人、編碼工具和特定領域的企業應用程式。
• 基礎模型：API 利用、微調、私有模型部署

主要趨勢和市場領導

在多階段基於代理的工作負載中，由於需要重複調整模型、整合工具和進行記憶體搜尋，因此推理是整個人工智慧基礎設施中最大的成本因素。資料顯示，多階段推理工作流程每單位訓練計算消耗的推理計算資源高達訓練計算的20倍。同時，模型上下文協定 (MCP) 和代理到代理 (A2A) 協定等開放標準正在開發中，作為連接人工智慧應用、企業資料來源和通用軟體代理的通用介面。此外，企業也擴大採用專用小規模語言模型 (SLM) 來降低延遲並控制代幣消耗成本。

就市場佔有率而言，基礎設施支出主要集中在英偉達的AI加速器和微軟的AI雲端平台。在軟體和基礎模型領域，截至2025年，OpenAI在基礎模型和AI應用方面處於市場領先地位，而埃森哲則在AI服務方面處於市場領先地位。 2026年上半年的數據顯示，在編碼和推理工作負載中API採用率不斷提高的推動下，Anthropic在基礎模型市場的佔有率正在成長。

本報告內容

這份於 2026 年 9 月發布的 283 頁分析報告詳細介紹了五個追蹤細分市場和 15 個行業領域的市場規模估算、對 1,533 個企業 AI 項目的實證分析、所有堆七個行業的競爭對手市場佔有率、對 1,533 個企業 AI 項目的實證分析、所有堆七層的競爭對手市場佔有率、對企業採用案例研究、採用障礙、安全問題和泡沫風險評估。

提及的公司

• AMD
• AWS
• Accenture
• Alibaba Cloud
• Anthropic
• Broadcom
• Capgemini
• Coreweave
• Google
• Huawei
• IBM
• Intel
• Microsoft
• Mistral
• NVIDIA
• OpenAI
• Oracle

目錄

生成式人工智慧與基於代理的人工智慧市場：2026-2032 年（PDF）

第1章：執行摘要

• 生成式人工智慧和基於代理的人工智慧市場：自上次報告以來發生了哪些變化？
• 執行摘要

第2章：引言

• 本報告提出的關鍵行動事項
• 章節概述：引言
• 出發點：生成式人工智慧的定義及其與人工智慧、機器學習和深度學習的關係
• 一種新範式：基於代理的人工智慧
• 導致目前局面的過程
• Anthropic的崛起：基於代理的人工智慧商業化的轉折點
• 生成式人工智慧和基於代理的人工智慧公司的資金籌措創歷史新高。
• 另一方面，（生成式）人工智慧在許多任務中超越了人類的能力。
• 企業應用正進入廣泛普及階段。
• 預計人工智慧領域的投資將大幅成長。
• 預計基於代理的人工智慧的使用將顯著增加。
• 生成式人工智慧和基於代理的人工智慧有可能對現有軟體和服務構成威脅，同時…
• 這將為許多下游產業帶來益處。
• 這些公司都實現了顯著的銷售成長。

第3章：基於代理的人工智慧基礎

• 章節概述：基於代理的人工智慧基礎知識
• 基於代理的人工智慧：將生成式人工智慧從「輸出生成」擴展到「目標實現」。
• 人工智慧代理的工作原理
• 基於代理的人工智慧的演進：從基於工具的工作流程到自主的多代理系統。
• 詳細分析：多智慧體系統
• 軟體開發範例：從傳統開發到多智慧體軟體開發
• 營運模式範例：人類和專家代理在整個供應鏈工作流程中協作。
• 對 AI 硬體/推理的影響：基於代理的 AI 可以擴展其能夠執行的任務範圍，但也可能增加每個已完成工作流程的推理消耗。
• 建構代理所需的條件：（1）框架，（2）建構器環境，（3）編配平台

第4章：市場與競爭格局概述

• 章節概要：市場與競爭格局概覽
• 生成式人工智慧和基於代理的人工智慧的市場環境
• 生成式人工智慧/基於代理的人工智慧市場範圍：包含/排除哪些內容
• 生成式人工智慧和基於代理的人工智慧市場：市場概覽
• 生成式人工智慧和基於代理的人工智慧市場：細分結構的變化
• 生成式人工智慧和基於代理的人工智慧市場：季度趨勢
• 生成式人工智慧和基於代理的人工智慧市場的主要促進因素和阻礙因素
• 競爭格局：市佔率概覽
• 主要公司在各細分市場的定位

第5章：人工智慧加速器

• 章節概述：人工智慧加速器
• 人工智慧加速器概述
• 人工智慧加速器的類型和功能
• 主要人工智慧加速器的功耗
• 訓練晶片和推理晶片（兩部分結構）
• 用於人工智慧加速的資料中心基礎設施
• 人工智慧加速器市場（5 部分系列）
• 領先的人工智慧加速器公司
• AI加速器：NVIDIA
• AI加速器：AMD
• AI加速器：博通
• AI加速器：英特爾
• 人工智慧加速器：Cerebras

第6章：基本模型

• 章節概述：基本模型
• 基本模型原理：Transformer架構
• Transformer模型中的輸入處理
• 透過推理生成輸出的機制
• 模型最佳化方法
• 模型發展方向：推理模型與世界模型
• 開放模型：開放原始碼與 OpenWait
• 小規模語言模型的重要性
• 模型開發不再只是規模擴大的競爭。
• 基礎模型市場
• 領先的基礎模型公司
• 大型公開級比賽通常不直接獲利
• 基礎模型：OpenAI
• 基礎模型：人為因素

第7章：人工智慧雲端和平台

• 章節概述：人工智慧雲端和平台
• 雲端人工智慧與平台：人工智慧平台堆疊
• 資料庫和資料平台
• 向量資料庫與圖資料庫的比較
• IaaS 和 GPU 雲端平台
• 基礎模型開發平台（5個部分）
• AI 運行時與營運平台：概述
• 代理開發與編配平台（4 部分結構）
• 人工智慧雲端平台市場（共6部分）
• 人工智慧雲端平台領域的領先公司
• 主要公司的市場佔有率
• AI雲端平台：微軟（共4部分）
• AI雲端平台：AWS（4部分）
• AI雲端平台：GCP（4個部分）

第8章：人工智慧應用

• 章節概述：人工智慧應用
• 人工智慧應用市場
• 人工智慧應用競爭格局：主要應用
• 人工智慧應用：跨類別和行業特定分類
• 人工智慧應用類別：市場規模與成長率
• 人工智慧應用競爭格局：分析
• 主要參與者擴大了市場佔有率
• 目前尚未實現獲利但被廣泛使用的AI應用

第9章：人工智慧服務

• 章節概述：人工智慧服務
• 人工智慧服務市場
• 公司在人工智慧服務的支出
• 主要公司的市場佔有率
• 人工智慧服務：埃森哲
• 人工智慧服務：IBM
• 人工智慧服務：德勤

第10章 最終使用者實現

• 生成式人工智慧和基於代理的人工智慧專案分析總結：與往年相比發生了哪些變化？
• 生成式人工智慧和基於代理的人工智慧專案分析：調查方法
• 生成式人工智慧/基於代理的人工智慧專案分析：定義
• 生成式人工智慧和基於代理的人工智慧專案分析：分析概述
• 部門特定的生成式人工智慧和基於代理的人工智慧項目
• 基於活動的生成式人工智慧和基於代理的人工智慧項目
• 從資料庫中提取的生成式人工智慧和基於代理的人工智慧專案範例
• 針對特定活動的生成式人工智慧和基於代理的人工智慧專案：在關鍵業務運營中全面實施
• 行業特定的生成式人工智慧和基於代理的人工智慧項目
• 行業特定和領域特定的生成式人工智慧和基於代理的人工智慧項目
• 專案類型：生成式人工智慧和基於代理的人工智慧
• 自主性別生成人工智慧/代理型人工智慧項目
• 業界對生成式人工智慧的採用情況的看法：採用率已經跨越了鴻溝，正在進入廣泛採用階段。
• 關於產業採用基於代理的人工智慧的觀點：科技公司正在領先潮流，但鴻溝尚未跨越。
• 案例研究：本章概述
• 從生成式人工智慧和基於代理的人工智慧實現的案例研究中獲得的全面見解。
• 案例研究1：湯森路透的人工智慧
• 案例二：伊維爾德羅拉
• 案例研究 3：貝克休斯
• 案例研究 4：TotalEnergies Italia
• 案例研究 5：CASMT（常州）自動化
• 案例研究 6：默克
• 案例研究 7：歐特克

第11章 市場與科技趨勢

• 生成式人工智慧和基於代理的人工智慧的關鍵趨勢
• 趨勢

第12章：實施挑戰與人工智慧泡沫風險

• 章節概要：實施挑戰與人工智慧泡沫風險
• 任務
• 人工智慧泡沫風險概述
• 企業投資報酬率泡沫：企業人工智慧真的能帶來投資報酬率嗎？
• 企業投資報酬率泡沫：以企業人工智慧合約續約率作為領先指標進行分析
• 人工智慧能力預期過高：人工智慧能力將以多快的速度提升？
• 資本投資復甦泡沫：銷售收入與資本投資相符嗎？
• 資本投資復甦泡沫：看漲情景
• 資本投資復甦泡沫：悲觀情景
• 資本投資復甦泡沫：悲觀情境－週期性資金籌措的詳細分析
• 資本投資復甦泡沫：模型分析結果
• 資本投資復甦泡沫：我們可以從過去的案例中學到什麼
• 股票估價泡沫：目前處於哪個階段？
• 整體評估：人工智慧是泡沫嗎？ ——物聯網分析的視角

第13章：調查方法與市場定義

第14章：關於物聯網分析

A 283-page report sizing the generative & agentic AI market to 2032, incl. vendor market shares, enterprise use cases, adoption, and trends

Key Data Snapshot:

• 2025 Market Size: $308 billion
• 2032 Forecast: $4.4 trillion
• Key Players: AMD,AWS,Accenture,Alibaba Cloud,Anthropic,Capgemini,Google,Intel,Microsoft,Mistral,Nvidia,OpenAI

Generative AI & Agentic AI Market Size, Trends, and Leader 2026-2032

Generative AI & Agentic AI Market Overview

IoT Analytics research indicates that the global Generative AI & Agentic AI market reached $308 billion in 2025. Based on enterprise adoption metrics and technology investment allocations, the market is forecast to reach $4.4 trillion by 2032, representing a compound annual growth rate (CAGR) of 46% between 2025 and 2032. Enterprise adoption reached mainstream levels in 2025, with 88% of surveyed organizations utilizing generative AI technologies. Market expansion correlates with an operational transition from human-prompted assistance toward autonomous agentic workflows that execute multi-step processes under human oversight.

This report outlines the five core segments comprising the technology stack:

• AI Accelerators: Comprises GPUs, TPUs, LPUs, and custom ASICs used for model training and inference.
• AI Cloud & Platforms: Comprises rented compute infrastructure, model development environments, runtime operations, and agent orchestration layers.
• AI Services: Comprises strategy consulting, system integration, custom agent design, and managed services.
• AI Applications: Comprises end-user software including chatbots, coding tools, and domain-specific enterprise applications.
• Foundation Models: Comprises API usage, fine-tuning, and private model deployments.

Key Trends & Market Leaders

Multi-step agentic workloads require repeated model calls, tool interactions, and memory retrieval, making inference the dominant cost driver across AI infrastructure. Data indicates that multi-step reasoning workflows consume up to 20x the amount of inference compute for every unit of training compute. Concurrently, open standards such as the Model Context Protocol (MCP) and Agent-to-Agent (A2A) protocols are establishing common interfaces to connect AI applications, enterprise data sources, and independent software agents. Parallel to this, enterprises are increasing the deployment of specialized small language models (SLMs) to reduce latency and contain token consumption costs.

Regarding market share, infrastructure spending flows primarily to NVIDIA for AI accelerators and Microsoft for AI cloud and platforms. Within software and foundation models, leadership in 2025 was held by OpenAI for foundation models and AI applications, alongside Accenture for AI services. Data for H1 2026 indicates that Anthropic expanded its foundation model market share, driven by API adoption across coding and reasoning workloads.

What is in this report?

This 283-page analysis published in September 2026 details granular market sizing across 5 tracked segments and 15 industry verticals, an empirical analysis of 1,533 enterprise AI projects, competitive market shares across all stack layers, 7 enterprise case studies, and an evaluation of implementation blockers, security concerns, and financial bubble risks.

Key Questions Addressed

• What is GenAI and agentic AI, how are they connected, and what are the technological components?
• Which use cases are enterprises prioritizing, and how autonomous are deployments becoming?
• What is the current market size, and how is the market expected to grow to 2032?
• How large are the 5 tracked segments (AI accelerators, foundation models, AI cloud & platforms, AI applications, AI services), and how is the segment mix shift expected to develop?
• What are the market shares of key players, and who leads in each of the 5 segments?
• Which are the leading AI applications, and which application categories grow fastest?
• How are leading enterprises deploying GenAI and agentic AI?
• What are the important implementation considerations and adoption blockers?
• What are the current and next trends - from inference costs to MCP/A2A and AI regulation?
• Are we in an AI bubble? What do ROI, CapEx payback, and valuations say?

Frequently Asked Questions

• What is the Generative AI & Agentic AI market size?
  • Data indicates that the global Generative AI & Agentic AI market reached $308 billion in 2025 and is forecast to grow to $4.4 trillion by 2032. The full report provides granular market sizing and forecasts broken down across 5 technology stack segments and 15 industry verticals.
• Who are the leading companies in the Generative AI & Agentic AI market?
  • In 2025, market leadership was held by NVIDIA in AI accelerators (80.9% share), Microsoft in AI cloud and platforms (26.7% share), OpenAI in foundation models (41.9% share) and AI applications (33.7% share), and Accenture in AI services (11.4% share). In H1 2026, Anthropic captured 54.4% of the foundation model market. The full report provides complete market share breakdowns for vendors across all five stack layers.
• What is the top trend in the Generative AI & Agentic AI market?
  • The market trend is the operational shift from single-turn AI assistants to multi-step autonomous agentic workflows, which is making inference the dominant compute cost across AI infrastructure. This transition is supported by open connectivity protocols such as the Model Context Protocol (MCP) and Agent-to-Agent (A2A) standards.

Companies mentioned

• AMD
• AWS
• Accenture
• Alibaba Cloud
• Anthropic
• Broadcom
• Capgemini
• Coreweave
• Google
• Huawei
• IBM
• Intel
• Microsoft
• Mistral
• NVIDIA
• OpenAI
• Oracle

Table of Contents

GenAI & Agentic AI Market Report 2026-2032 (PDF)

1. Executive summary

• The GenAI & agentic AI market: What has changed since our last report?
• Executive summary (6 parts)

2. Introduction

• Key action items from this report
• Chapter overview: Introduction
• Starting point: Defining GenAI and its relationship with AI, ML, and DL
• The new paradigm: Agentic AI
• How we got here (2 parts)
• Anthropic’s rise marked the commercial breakthrough of agentic AI
• Funding for GenAI & agentic AI companies reached record values
• In the meantime, (Gen) AI has surpassed human capabilities in many tasks
• Enterprise usage is now mainstream
• AI investments are expected to increase strongly
• Agentic AI usage is expected to increase strongly
• GenAI & agentic AI may put existing software and services at risk…
• … but it also benefits a number of downstream industries
• … and those companies report strong top-line revenue growth

3. Fundamentals of agentic AI

• Chapter overview: Fundamentals of agentic AI
• Agentic AI extends GenAI from producing outputs to pursuing goals
• How AI agents work
• The agentic AI evolution: From tool-using workflows toward autonomous and multi-agent systems
• Deep-dive: Multi-agent systems
• Example software development: From traditional development to multi-agent software development
• Illustrative operating model: Humans and specialized agents collaborate across supply-chain workflows
• Impact on AI hardware/inference: Agentic AI extends task horizons but can increase inference consumption per completed workflow
• In order to build agents, it requires: (1) a framework, (2) a builder environment, and (3) an orchestration platform

4. Market and competitive landscape overview

• Chapter overview: Market and competitive landscape overview
• The GenAI & agentic AI landscape
• GenAI & agentic AI market scope: What is included and what is not (3 parts)
• GenAI & agentic AI market: Total market overview
• GenAI & agentic AI market: Segment mix shift
• GenAI & agentic AI market: Quarterly view
• General drivers and inhibitors for the generative & agentic AI market
• Competitive landscape: Market share overview
• How the main players are positioned across different market segments

5. AI accelerators

• Chapter overview: AI accelerators
• AI accelerators overview
• Types of AI accelerators and their capabilities
• Selected AI accelerator power consumption
• Training vs. inference chips (2 parts)
• Data center infrastructure for AI acceleration (4 parts)
• AI Accelerator market (5 parts)
• Leading AI accelerator companies
• AI accelerators: NVIDIA (4 parts)
• AI accelerators: AMD (4 parts)
• AI accelerators: Broadcom (4 parts)
• AI accelerators: Intel (4 parts)
• AI accelerators: Cerebras (4 parts)

6. Foundation models

• Chapter overview: Foundation models
• The basis of foundation models: The transformer architecture
• How input is processed in transformer models
• How inference generates outputs
• How models are optimized
• Where model development is heading: Reasoning and world models
• Open models: Open source vs. open weights
• Importance of small language models
• Model development is no longer a one-way race to scale
• Foundation models market (5 parts)
• Leading foundation model companies
• Popular open-weight models that are often not directly monetized
• Foundation models: OpenAI (4 parts)
• Foundation models: Anthropic (4 parts)

7. AI cloud & platforms

• Chapter overview: AI cloud & platforms
• Cloud AI & platforms: The AI platform stack (2 parts)
• Databases & data platforms (2 parts)
• Vector vs. graph databases
• IaaS & GPU cloud platforms (2 parts)
• FM development platforms (5 parts)
• AI runtime & operations platforms: Overview
• Agentic dev. & orch. platforms (4 parts)
• AI cloud & platforms market (6 parts)
• AI cloud & platforms key players
• Key players market share
• AI cloud & platforms: Microsoft (4 parts)
• AI cloud & platforms: AWS (4 parts)
• AI cloud & platforms: GCP (4 parts)

8. AI applications

• Chapter overview: AI applications
• AI applications market (3 parts)
• AI applications competitive landscape: Top applications
• AI applications: Horizontal categories and industry verticals
• AI application categories: Size and growth
• AI applications competitive landscape: Breakdown
• Top movers took market share
• Widely used AI apps that are currently not monetized

9. AI services

• Chapter overview: AI services
• AI services market (5 parts)
• Enterprise spending on AI services
• Key players market share
• AI services: Accenture (3 parts)
• AI services: IBM (3 parts)
• AI services: Deloitte (3 parts)

10. End-user adoption

• Overarching takeaways of the GenAI & agentic AI project analysis: What has changed versus prior years
• Analysis of GenAI & agentic AI projects: Methodology
• Analysis of GenAI & agentic AI projects: Definitions (2 parts)
• Analysis of GenAI & agentic AI projects: Analysis overview
• GenAI & agentic AI projects by department
• GenAI & agentic AI projects by activity
• Set of GenAI & agentic AI sample projects from the database
• GenAI & agentic AI projects by activity: Adoption now strong in primary activities
• GenAI & agentic AI projects by industry
• GenAI & agentic AI projects by industry & department
• GenAI & agentic AI projects by type of GenAI and agentic AI
• GenAI & agentic AI projects by autonomy
• Our view on GenAI adoption by industry: Adoption has crossed the chasm into mainstream
• Our view on agentic AI adoption by industry: Technology companies are leading the way, but the chasm has not been crossed yet
• Adoption examples: Chapter overview
• Overarching insights from the GenAI & agentic AI adoption examples
• Example 1: AI at Thomson Reuters (8 parts)
• Example 2: Iberdrola (3 parts)
• Example 3: Baker Hughes (3 parts)
• Example 4: Total Energies Italia (2 parts)
• Example 5: CASMT (Changzhou) Automation (2 parts)
• Example 6: Merck
• Example 7: Autodesk (3 parts)

11. Market & technology trends

• Key trends in the GenAI & agentic AI landscape
• Trend 1 (2 parts)
• Trend 2
• Trend 3
• Trend 4 (2 parts)
• Trend 5
• Trend 6
• Trend 7 (2 parts)
• Trend 8 (2 parts)

12. Adoption challenges & AI bubble risk

• Chapter overview: Adoption challenges & AI bubble risk
• Challenge 1
• Challenge 2 (2 parts)
• Challenge 3 (2 parts)
• Challenge 4 (2 parts)
• AI bubble risk overview
• Enterprise ROI bubble: Is there ROI in enterprise AI?
• Enterprise ROI bubble: Enterprise AI renewals as a leading indicator
• AI capability expectation bubble: How fast will capability improve?
• CapEx payback bubble: Does the revenue justify the CapEx?
• CapEx payback bubble: The bull case
• CapEx payback bubble: The bear case
• CapEx payback bubble: The bear case – Circular financing deep dive
• CapEx payback bubble: What our model says
• CapEx payback bubble: What history tells us
• Equity valuation bubble: Where do we stand?
• Overall verdict: Are we in a bubble? The IoT Analytics view

13. Methodology & market definitions

• Research Methodology
• Definition of technology stack
• Definition of industries for the market model & outlook
• Definition of industries used to classify projects in the end user adoption chapter
• Countries in APAC
• Countries in EMEA
• Countries in the Americas

14. About IoT Analytics