AI 行情见顶?美银新新预测:未来5 年AI 将创造新增 1万亿美元芯片销售额.

报告发布：2026-06-23，BofA Global Research，首席分析师 Vivek Arya（全球半导体首席）

核心主题：AI 资本开支周期延续至 2028，5 年新增 1 万亿芯片销售额，2030 半导体总规模上调至 2.7 万亿美元

一、报告开篇核心原文（英文原版）

It took the semiconductor industry roughly 50 years to hit its first $1tn in annual sales. We expect AI to deliver another incremental $1tn of semiconductor revenue over the next five years, lifting our 2030 global semi TAM forecast from $2.3tn to $2.7tn, representing a 28% CAGR from 2025–2030.

We believe the market is prematurely pricing an end to the AI capex cycle, but AI infrastructure buildout remains mid-cycle within an 8–10 year upgrade journey for traditional IT stacks optimized for generative AI and agentic workloads. CY2023 marked the kickoff of the AI semi cycle; we are only at the halfway point of this multi-year investment wave, with clear visibility for AI-related capital spending extending through 2028.

Consensus materially underappreciates the mission-critical, offensive and defensive nature of capex from hyperscalers, sovereign governments and enterprise end users. AI ROI is now materializing via two core levers: 1) higher operational efficiency by replacing legacy CPU infrastructure with GPU/custom ASIC accelerators; 2) data moat defense for cloud, search, e-commerce and social platforms. Enterprise AI deployment remains in early penetration, while sovereign AI spending is accelerating globally as nations build domestic AI compute capacity.

中文直译对照

芯片行业耗时约 50 年才达成首个年度 1 万亿美元销售额。我们预计人工智能将在未来五年为半导体行业额外创造 1 万亿美元增量营收，将 2030 年全球半导体整体市场规模（TAM）预测从 2.3 万亿美元上调至 2.7 万亿美元，2025-2030 年复合增速达 28%。

市场过早定价 AI 资本开支周期见顶，但 AI 基础设施建设仍处于 8-10 年传统 IT 架构 AI 改造周期的中段；2023 年是 AI 半导体周期起点，当前仅处于本轮多年投资浪潮的中点，AI 相关资本开支需求清晰可见，至少持续至 2028 年。

市场一致预期严重低估了超大规模云厂商、各国政府、企业终端客户资本开支的刚需属性、进攻与防御双重价值。

AI 投资回报已通过两条核心路径兑现：

1）用 GPU / 定制专用芯片替代传统 CPU 架构，大幅提升运营效率；

2）为云、搜索、电商、社交平台构筑数据护城河。企业端 AI 落地渗透率仍处于早期，全球各国政府主权算力采购正在加速，各国均在搭建本土 AI 算力底座。

二、五大 AI 增长驱动因素 原文段落

Five structural tailwinds will drive the next $1tn semiconductor revenue expansion from AI:

1. AI data center systems: TAM set to expand from $273bn (2025) to $1.7tn (2030), with AI accelerators alone accounting for $1.2tn of this total, driven by hyperscaler custom ASIC rollouts and LLM training/inference demand.
2. Memory resilience anchored by multi-year long-term supply agreements (LTAs): HBM market to hit $168bn by 2030 (37% 2025–2030 CAGR), as agentic AI pushes token consumption 10–1000x higher than conventional chatbots, exploding KV cache and high-bandwidth memory demand.
3. Wafer fab equipment (WFE) & onshoring: We lifted 2028 WFE forecast to $250bn, 2029 to $268bn, 2030 to $292bn, supported by cleanroom capacity expansion, advanced node buildouts and domestic semiconductor manufacturing incentives across North America, Europe and Asia.
4. Power & analog semi boom: AI datacenter power draw is rising up to 100x legacy racks; power infrastructure analog TAM to grow from $300mn today to $1.8bn by end-decade, with SiC solid-state transformers, solid-state circuit breakers and liquid cooling hardware as key growth vectors.
5. Agentic AI server CPUs: Agent workloads will create a $170bn server CPU market across x86 and ARM architectures, as autonomous AI agents require persistent state storage, multi-step reasoning and continuous real-time orchestration.

中文对照

五大结构性红利将推动 AI 创造下 1 万亿半导体增量收入：

1.AI 数据中心整机：市场规模从2025年2730 亿美元扩张至2030 年1.7 万亿美元，仅 AI 加速器板块就占1.2 万亿美元，由云厂商自研 ASIC 芯片、大模型训练与推理需求拉动；

2.长期供货协议支撑存储行业高景气：2030 年HBM 市场规模达1680 亿美元，2025-2030 年复合增速37%；智能体 AI 的 Token 消耗量较传统对话机器人提升 10-1000 倍，KV 缓存与高带宽内存需求爆发；

3.晶圆设备（WFE）与芯片产业回流：上调2028 年晶圆设备支出至2500 亿美元、2029 年2680 亿、2030 年2920 亿；洁净室扩容、先进产线建设、欧美亚本土芯片制造扶持政策提供支撑；

4.电源与模拟芯片爆发：AI 数据中心功耗较传统机架最高提升100 倍；电力基础设施模拟芯片市场从当前3 亿美元增长至2030 年末18 亿美元，碳化硅固态变压器、固态断路器、液冷硬件为核心增长点；

5.智能体服务器 CPU：AI 智能体多步骤推理、长期状态存储、实时流程调度需求，将在 x86 与 ARM 双架构下创造 1700 亿美元服务器 CPU 市场。

中文对照

AI 行业核心瓶颈正在从单纯算力转向内存带宽与容量。长上下文、多模态模型、智能体流程场景下，系统需要持续加载模型权重、留存 KV 缓存数据；单颗 AI 加速器搭载 HBM 平均容量将从 2025 年 187GB 提升至 2030 年 800GB 以上，DRAM、NAND 存储供需缺口将贯穿整个十年周期。

OpenAI、Anthropic 等头部大模型厂商收入高速增长，验证 AI 算力需求持续扩张；云厂商与存储企业签订多年长期锁量订单，锁定未来数年需求，存储板块不再复刻过往科技周期的剧烈下行风险。

四、市场波动与投资结论原文

Near-term AI stock volatility will persist amid heightened scrutiny of AI ROI and hyperscaler free cash flow, but pullbacks create buying opportunities for semiconductor leaders with exposure to AI compute, memory, WFE and power analog. We raise price targets on Micron, Intel, Arm, maintaining Overweight ratings on NVIDIA, AMD, Broadcom as core AI infrastructure picks.

This AI cycle is not a short-term bubble driven by consumer chatbot hype—it is a multi-decade enterprise infrastructure rebuild comparable to the rollout of cloud computing in the 2010s, with tangible, recurring revenue streams from mission-critical corporate and government AI deployments.

中文对照

短期市场会持续波动，市场将持续审视 AI 投资回报与云厂商自由现金流，但回调即是布局机会，优先配置算力、存储、晶圆设备、电源模拟赛道龙头。

上调美光、英特尔、安谋目标价；维持英伟达、AMD、博通 “增持” 评级，列为 AI 基础设施核心标的。本轮 AI 周期并非由消费端聊天机器人热度催生的短期泡沫，而是堪比 2010 年代云计算普及的数十年企业底层基础设施重构，企业与政府刚需 AI 场景将带来稳定、可持续的长期现金流。

补充：宏观 AI 生产力展望（BofA 2026 年 5 月宏观团队配套原文）

Investors understate AI’s long-run productivity impact. Current measured aggregate AI productivity lift is only 0.1%, but AI will eventually deliver a 10x larger global productivity boom than the internet or electricity. Enterprise AI will lift S&P 500 operating margins by 200 bps over five years, equaling $55bn annualized cost savings. Current weak macro AI data reflects early deployment phase; productivity gains compound exponentially as agentic AI automates end-to-end business workflows.

中文对照

投资者低估 AI 长期生产力价值。当前宏观层面可观测的 AI 整体效率提升仅 0.1%，但长期来看 AI 带来的全球生产力提升将是互联网、电力时代的 10 倍。未来五年企业 AI 落地将推升标普 500 企业营业利润率 2 个百分点，折合每年 550 亿美元成本节约。当前宏观数据偏弱仅因行业处于落地早期；随着智能体 AI 实现全业务流程自动化，效率提升将呈指数级放大。

另外，软银集团董事长孙正义也看好未来AI的发展，他在股东大会上透露：针对物理AI应用场景，“已在某工厂开始机器人量产，即将正式发布，相信大家会感到惊讶”。他同时表示，软银将通过汇聚各垂直领域顶级机器人企业，成为“压倒性世界第一的机器人公司”，并提及2026年计划完成收购瑞士工业巨头ABB旗下机器人业务。