华泰证券研报指出，2026年AI产业正在从大模型预训练阶段切换至AI Agent商业化落地阶段，这使得推理算力需求进入了加速增长的通道。看好三条主线：AI链方向，国产算力闭环加速形成，超节点互联与存储升级形成共振，推理需求拐点得以明确，AI端侧上折叠机、AI眼镜等新品周期将至，结构性创新机遇值得重视；功率与被动元件方向，AI功耗驱动MLCC、电感、电容、功率半导体量价齐升，涨价周期与国产替代形成共振；自主可控方向，上游制造、设备以及零部件国产化进程加速，同时先进封装价值量得以系统性提升。

华泰证券2026中期展望 | 电子 ：AI Agent 驱动算力新周期与自主可控再加速

核心观点

AI Agents are accelerating the expansion of inference compute and storage capacity, while the localization process in the autonomous control chain is accelerating once again.

We believe that in 2026 the AI industry is shifting from the pre-training phase of large models to the commercialization and deployment of AI Agents, and that the demand for inference compute is entering an accelerated growth channel. We see three main lines ahead: in the AI chain direction, the acceleration of the formation of a domestic compute closed loop, the resonance between super-node interconnection and storage upgrades, the clarification of the inflection point of inference demand, and the upcoming cycle of new products such as foldable phones and AI glasses on the AI endpoint side, which deserves attention for its structural innovation opportunities; in the power and passive component direction, the AI power consumption drives the increase in volume and price of MLCC, inductors, capacitors, and power semiconductors, and the resonance between the price increase cycle and domestic substitution; and in the autonomous and controllable direction, the localization process of upstream manufacturing, equipment, and components is accelerating, and at the same time the systematic increase of the value of advanced packaging is taking place.

AI chain: AI Agents drive inference compute for high growth, while domestic compute, transportation, and storage forces accelerate development.

我们看好AI Agent商业化落地驱动推理算力进入加速增长阶段。根据国家数据局统计2026年3月国内Token日均调用量已超140万亿次推理端算力需求正迎来快速增长阶段国产大模型与国产算力形成Day 0适配闭环国产 AI芯片 替代持续加速。互联侧大厂超节点方案陆续落地Scale Up网络驱动交换芯片用量实现显著增长PCB 方面AI架构迭代驱动高端产品量价齐升。存储侧我们预计2H26合约价延续上行AI强劲需求下供给缺口短期难以补足Agent长记忆场景推动高容量内存及企业级SSD需求迈入新量级。端侧方面我们看好AI驱动 消费电子 结构性创新 苹果 折叠屏AI/AR眼镜OpenAI硬件等新品密集涌现2027年iPhone 20周年有望迎来硬件创新集中释放周期。

Power and passive components: the AI power consumption drives the upgrade of components, with MLCC and power generating multidimensional resonance

华泰证券研报指出，2026年AI产业正在从大模型预训练阶段切换至AI Agent商业化落地阶段，这使得推理算力需求进入了加速增长的通道。我们看好AI服务器功耗持续攀升驱动被动元件与功率半导体量价齐升。MLCC方面，AI服务器用量指数级跃升，全产业链进入涨价通道，海外高端产能挤占效应凸显，我们看好本轮超级周期延续至2027年，国内龙头有望受益于国产替代机遇。电感方面，垂直供电有望于2H26上量，TLVR耦合电感单颗价值量从0.5美元向1-2.5美元跃升。铝电解电容方面，AI牛角电容ASP从传统10元级别大幅提升，供给端产能切换存在刚性缺口，供需紧张具备持续性。功率半导体受益AI挤占效应扩大，叠加传统市场经历三年去库存后进入补库周期，我们看好功率公司从收入修复迈入盈利修复。

Autonomous controllable: manufacturing and packaging and testing accelerate catch up, upstream equipment and components localization accelerates.

We think upstream manufacturing and equipment are the most critical bottlenecks in the semiconductor industry chain and that China is accelerating its catch-up. On the manufacturing side, TSMC’s 2nm process has entered mass production and China remains three to four years behind the global leader, but its mature process stands to benefit from the inference compute demand created by the AI boom, and ASP is expected to continue rising. On the advanced packaging side, OSAT 2.5D capacity is scheduled to enter large-scale deployment starting in 2026, 3D packaging is already deployed in mobile SoC, and the far-term trends for glass substrates and CoPoS are clear. On the equipment side, surging expansion demand from downstream storage and logic drives global WFE higher, local vendors are seeing sharply rising orders, and their delivery advantages position them to capture overseas spillover orders. Component demand keeps rising, and we expect vacuum valves, electrostatic chucks, and EFEMs in low domestic-content areas to sustain breakthroughs under the dual drivers of capacity shortages and localization.

风险提示：中美贸易摩擦升级风险、电子产品渗透率不及预期风险、AI技术发展不及预期风险以及半导体行业周期波动风险。本研报涉及的未上市或未覆盖个股内容，均系对其客观信息的整理，并不代表团队对该公司、该股票的推荐或覆盖。

（文章来源：财联社）

来源：华泰证券：AI Agent加速推理算力与存力扩容 自主可控链国产化再加速 | 东方财富网