AIチップ市場の展望 (2023-2028年)：需要と供給の視点からの洞察

生成AIのリリースは、急速に世界的な熱狂を巻き起こしています。大規模言語モデル (LLM) の開発も極めて重要ですが、真のゲームチェンジャーは、学習後のエッジデバイス上でのAI推論タスクの実行です。これによって新たな用途や需要が生まれ、将来のAIの事業機会はさらに大きくなるでしょう。このコンピューティングパワーを実現する鍵はAIチップにあり、その開発が単に焦点となるだけでなく、業界にとって極めて重要な必需品となるでしょう。この緊急性から、多くの企業が主にGPUや自社開発のASICアクセラレータに焦点を当ててAIチップ市場に参入し、競争は激化し、半導体業界は新たな章が開かれようとしています。

AIチップ市場の観点から、ハイエンドのデータセンター向けGPUの需要は2024年に218％増加し、ASICアクセラレータは178％成長し、広帯域メモリ (HBM) 市場を139億米ドルに押し上げると予想されています。エッジでは、AIスマートフォンとAI PCの導入が関連チップの出荷を押し上げ、2024年にはそれぞれ2億6,000万ユニット、5,000万ユニットの出荷が予測され、年間成長率はそれぞれ293%、79%となる見通しです。AIアプリケーションの急増は、2024年のAIチップ出荷数の指数関数的な成長を促進し、有望な市場の将来と業界投資の増加を予告しています。

当レポートでは、進化するAIチップ市場を調査し、製造プロセスやパッケージング技術の先進性、さまざまなAIハードウェアデバイスの出現など、川上から川下まで議論を広げています。また、市場シェアや開発ロードマップなどの視点を通じて、AIチップの開発動向に焦点を当て、さまざまなチップセグメントにおける競合情勢を分析します。最後に本レポートでは、AIの開発が半導体市場に与える影響を、IC製造に焦点を当てて評価します。これには、AIの需要によりますます重要になる先進プロセス技術や、ファウンドリによる先進プロセスやパッケージングの計画やキャパシティレイアウトの検討も含まれます。

目次

第1章 調査の背景・範囲

第2章 クラウドAIアクセラレータの出荷予測

• クラウドAIアクセラレータ：GPU
• AIアクセラレータ：ASIC
• HBM市場分析
• サマリー

第3章 コンシューマーエッジAIチップの出荷予測：AIスマートフォンとAI PC

• エッジAIチップの需要：AIスマートフォン
• AI PC向けエッジAIチップの需要
• サマリー

第4章 - ハイエンドAIチップ供給予測：ウエハファウンドリ

• ハイエンドAIチップ製造の技術ロードマップ
• 先進ウエハ製造能力の分析
• サマリー

第5章 総論

List of Figures

• Figure 1: The Number of AI Companies Acquired by Leading Companies from 2017-2023
• Figure 2: Overview of Parameters in Leading Large Language Models and Their Development Direction
• Figure 3: A Comparative Overview of the Ability Test of Edge Small Language Models
• Figure 4: The Development Process of Generative AI Software and Hardware
• Figure 5: The Three Major Bottlenecks in The Supply and Demand of Cloud AI Servers
• Figure 6: The Three Major Bottlenecks in Running LLM Model Applications on Terminal Devices
• Figure 7: The Applicable Range of Model Parameter Sizes for Cloud and Edge AI, and Suitable Types for Generative AI Tasks
• Figure 8: Types of Server AI Accelerators
• Figure 9: High-End Cloud GPU Shipment Forecast 2023-2024
• Figure 10: Forecasts of Market Share of High-End Cloud GPU Shipments (2023~2024)
• Figure 11: Development Roadmaps of High-End Cloud GPUs by Major Makers 2023-2026
• Figure 12: Forecast of High-End Cloud GPU Shipments and Annual Growth Rate from 2023-2028
• Figure 13: Global Cloud-End ASIC Accelerator Shipment Forecast 2023-2024
• Figure 14: Development Roadmap of Major Players for Cloud ASIC Accelerators 2023-2024
• Figure 15: Product Roadmaps of Cloud ASIC Accelerators 2023-2026
• Figure 16: Forecast of Global Cloud ASIC Accelerator Shipments and Annual Growth Rate 2023-2028
• Figure 17: Forecast of the Global HBM Market for AI Chips 2023-2028
• Figure 18: Comparison of HBM Specifications Across Different Generations and Their Mass Production Years
• Figure 19: Production Times for Different HBM Generations by SK Hynix, Samsung and Micron
• Figure 20: Production Times Lines and Process Nodes for HBM3e, HBM4 by Three Major Players
• Figure 21: Korean Memory Makers' Business Strategy and Direction on Customized HBM Market
• Figure 22: Description of the Features of Generative AI on PCs and Smartphones
• Figure 23: Two Major Mainstream CPUs Launch New Products to Lay Out the AI PC Market
• Figure 24: Comparison of AP Specifications of Top Five AP Vendors
• Figure 25: Key Observation of AI Smartphone APs and Generative AI Strategies of Top Five AP Makers
• Figure 26: Development Roadmaps/Trends of the Top 5 Smartphone AP Products
• Figure 27: Global AI Smartphone AP Shipment Forecast 2023~2028
• Figure 28: Forecast of Global AI Smartphone Shipments and Penetration Rate 2023-2028
• Figure 29: Generative AI and AI PC Ecosystem
• Figure 30: Specification Comparison of Major AI NB CPU 2023~2025
• Figure 31: Current Status of PC Operating Systems and Their AI Applications in AI PC Development
• Figure 32: Global AI PC Processor Shipment Trends and Forecast 2023-2028
• Figure 33: AI PC Shipment Forecasts 2024
• Figure 34: Roadmap for Developing Advanced Wafer Foundry Processes
• Figure 35: Roadmaps and Comparison of EUV and Nano Imprint Technology
• Figure 36: Transformation of Transistor Structure
• Figure 37: Comparison of Transistor Power Supply Solutions and the Timing of Industry Adoption
• Figure 38: Changing Proportion of DTCO's Contribution to Chip Performance Improvement
• Figure 39: Pathways for Chip Performance Improvement
• Figure 40: Maps Combine with Generative AI to Improve User Interaction Experience
• Figure 41: Advantages and Disadvantages of FOPLP Compared to FOWLP
• Figure 42: Schematic Diagram of SoC and Chiplet
• Figure 43: Chiplet Technology
• Figure 44: TSMC LIPINCON Interconnect Technology
• Figure 45: 2.5D Packaging Solutions and Characteristics Comparison
• Figure 46: Interposer Size and Component Integration Trends
• Figure 47: Evolution of CoWoS Technology
• Figure 48: Evolution of Chip Packing
• Figure 49: Comparison of Micro-Bump Bonding and Hybrid Bonding
• Figure 50: Features, Advantages and Disadvantages of Glass Substrates
• Figure 51: Co-Packaging Optical and Light Engine Packaging Solutions
• Figure 52: Advanced Packaging Technology Roadmap for AI Server Accelerators
• Figure 53: Summary of Major Foundries' Advanced Process Roadmaps below 5nm
• Figure 54: Capacity Planning for Wafer Foundry Advanced Processes below 5nm
• Figure 55: Proportion of Wafer Foundries with Advanced Processes below 5nm
• Figure 56: Locations for Advanced Packaging Deployments by Wafer Foundries
• Figure 57: Estimate of TSMC's CoWoS Annual Capacity
• Figure 58: Wafer Foundry Advanced Manufacturing Technology Roadmaps 2023-2028
• Figure 59: Wafer Foundry Advanced Packing Technology Roadmaps 2023-2028
• Figure 60: AI Applications and AI Chip Ecosystem

The release of Generative AI has rapidly ignited a global frenzy. While developing large language models (LLMs) is crucial, the real game-changer is executing AI inference tasks on edge devices after training. This will drive the creation of new applications and demands, suggesting that future AI business opportunities will be even more significant. The key to delivering this computing power lies in AI chips, making their development not just a focal point but a critical necessity for the industry. This urgency has led many companies to enter the AI chip market, focusing primarily on GPUs and self-developed ASIC accelerators, intensifying competition and opening a new chapter in the semiconductor industry.

This research report explores the evolving AI chip market, extending the discussion to the upstream and downstream sectors, including advancements in manufacturing processes, packaging technologies, and the emergence of various AI hardware devices. From the perspective of the AI chip market, demand for high-end datacenter GPUs is expected to increase by 218% in 2024, while ASIC accelerators will grow by 178%, driving the high-bandwidth memory (HBM) market to US$13.9 billion. At the edge, the introduction of AI smartphones and AI PCs will boost related chip shipments, with projected shipments of 260 million and 50 million units, respectively, in 2024, reflecting annual growth rates of 293% and 79%. The surge in AI applications is poised to drive exponential growth in AI chip shipments in 2024, heralding a promising market future and increasing industry investment.

This research report will analyze the competitive landscape across various chip segments, focusing on AI chip development trends through perspectives such as market share and development roadmaps. The report will examine not only major chipmakers like Nvidia, AMD, Intel, MediaTek, and Qualcomm but also the chip development strategies of companies like Apple, Amazon, Google, and Microsoft, highlighting the diverse advancements in AI chips and exploring emerging trends in the chip market driven by AI applications.

As AI applications expand and new AI chips are introduced, the demand for improved AI model performance - particularly regarding response speed, accuracy, and reliability - continues to grow. Enhancing AI chip performance across all dimensions will hinge on advancements in chip manufacturing technology. The report will conclude by assessing the impact of AI development on the semiconductor market, focusing on IC manufacturing. This includes examining advanced process technologies that are becoming increasingly important due to AI demands, as well as the planning and capacity layout of advanced processes and packaging by foundries.

Before 2024, the development of generative AI focused on AI model training. However, as large language models begin to enter edge devices, AI's implementation will drive more business opportunities, including new applications, new chips, and new products. This will not only expand existing markets proportionally but also change existing technology and capacity demands under new requirements, altering the competitive landscape of the chip market and leading to unprecedented changes in the semiconductor ecosystem. This is the core focus of the research report.

Table of Contents

Chapter 1 - Research Background and Scope

• 1.1. Factors and Dynamics Driving AI Growth
• 1.2. Demand for Chips in Generative AI Applications
• 1.3. Research Definition and Scope

Chapter 2 - Cloud AI Accelerator Shipment Forecast

• 2.1. Cloud AI Accelerators - GPU, 2023-2028
• 2.2. Could AI accelerators - ASIC, 2023-2028
• 2.3. HBM Market Analysis from 2023 to 2028
• 2.4. Summary

Chapter 3 - Consumer Edge AI Chip Shipment Forecasts - AI Smartphones and AI PCs

• 3.1. Demand for Edge AI Chips: AI Smartphone, 2023-2028
• 3.2. Edge AI Chip Demand for AI PCs, 2023-2028
• 3.3. Summary

Chapter 4 - Forecast of High-End AI Chip Supply - Wafer Foundries

• 4.1. Technology Roadmap for High-End AI Chip Manufacturing
• 4.2. Analysis of Advanced Wafer Fabricating Capacity, 2023-2028
• 4.3. Summary

Chapter 5 - Conclusion