2026-02-03 · 个股论点

看好AXTI作为光子学InP材料瓶颈，预计2026下半年因需求激增出现短缺。

涉及标的： $AVGO $AXTI $COHR $EWY $FLKR $HY9H $LITE $MU $NVDA $POET $SNDK

中文翻译

我持有 $AXTI 的多头头寸。今天我们正目睹 $SNDK、$MU、SK 海力士和三星带来的 HBM/NAND 超级周期，由于需求极度旺盛，价格持续上涨。下一个是什么？-> 光子学 | 共封装光学(CPO) | 光互连。需求规模呈现抛物线式增长。（Lightcounting 数据）我对 AXT 的投资逻辑是：它是从磷化铟(InP)衬底到前驱体供应链的材料瓶颈。而 LightCounting 在过去几个月将需求预测翻了三倍。这是一个物理/材料问题。未来，西方供应链可以通过硅光子(SiPh)或 $POET 中介层来减少 InP 衬底/原料的使用。但超大规模云服务商无法绕过这一限制。目前，中游瓶颈如 $LITE、$COHR、$AVGO 等依赖住友 | JX（日本）等西方供应链。但这些公司使用的许多前驱体源自中国。在中国最近的出口管制后，日本公司虽有部分产能，但缺乏原料前驱体，他们可能正在消耗库存（约1年，无公开信息）。但随着超大规模云服务商需求激增，库存将耗尽，使 $AXTI 掌握更多供应链控制权。我们已经看到 $NVDA 确保了 EML 产能（使用 InP）。CPO CW 激光器、SOA 和其他组件也使用 InP。 “短缺”瓶颈正是 $AXTI 卡脖子的 InP 晶圆产线。根据预测，我预计超大规模云服务商将在 2026 年开始为 2028 年的扩张储备库存，导致 2026 年下半年供应短缺。 $AXTI 的主要问题在于出口管制 + 60 天行政延迟。中国可能正通过 $AXTI 出牌，迫使美国超大规模云服务商依赖其供应链而非日本。这在财报前的延迟收入确认中体现为双刃剑。但这些出口管制对住友等竞争对手伤害最大。即使上游没有看到 100% 的月度涨价，基准情形下 InP 衬底和前驱体需求的增加也将提升 AXT 的营收底线。 AXT 仍是我对 2027 年及以后即将到来的超大规模光子学扩张最看多的材料瓶颈标的。起初我认为这只是一个短期的风险对冲交易，但在研究了涉及光子学的新数据和架构后，我认为 $AXTI 将是未来 AI 建设的关键参与者。我的逻辑可能过早，或者这恰恰是拐点。

英文原文

I'm long shares on $AXTI. We're seeing the HBM/NAND supercycle with $SNDK, $MU, Sk Hynix, and Samsung today. With constant price hikes due to extreme demand. The next? -> Photonics | CPO | Optical Interconnects. Demand scale up looks parabolic. (Lightcounting) My thesis was on on AXT was that they're the material bottleneck from InP Substrates to the precursor supply chain. And LightCounting's projections tripled the demand forecast in the past few months. This is a physics/material problem. In the future, Western supply chains can engineer away less InP substrates/feedstock usage with SiPh or $POET interposers. But hyperscalers can't get around it. Currently, midstream bottlenecks like $LITE, $COHR, $AVGO, and others rely on Western supply chains like Sumitomo | JX (Japan). But many precursors that companies these companies originated from China. After China's most recent export controls, the Japanese companies have some production, but without raw precursors, they are likely using inventory (maybe 1 year, there is no public information). But as hyperscaler demand ramps up, the inventory will disappear, leaving more control of the supply chain to $AXTI. We've already see $NVDA secure EML capacity (which uses InP). CPO CW lasers, SOA, and other components also use InP. The "shortage" bottleneck is the same InP wafer lines that $AXTI chokepoints. Judging from the projections, I expect hyperscalers to start securing inventory for the 2028 ramp in 2026, leading to supply shortages H2 2026. $AXTI's main issue is export controls + 60D administrative delays. China is likely playing their hand with $AXTI to force American hyperscalers to become dependent on their supply chain over Japan. This has been a double edged sword as seen with their pre-earnings release with delayed revenue recognition. But these export controls hurt their competitors in Sumitomo the most. Even if we don't see 100% monthly price hikes upstream, the base-case is ramped up demand for InP substrates and precursors increasing AXT's bottom line revenue. AXT remains my most bullish material chokepoint play for the impending hyperscale photonics ramp in 2027 -> onward. Before I expected this to be a short few month play for risk management, but after looking new data and architectures involving photonics, I think $AXTI will likely be a critical player for the AI buildout moving forward. My thesis might be early, or perhaps this may very well be the inflection point.

在 X 上查看原推 ↗