2026-01-07 · 供应链分析

铟价分化及日本出口管制使AXTI成AI光子学关键瓶颈，估值被低估。

涉及标的： $AXTI $COHR $GOOGL $LITE $MSFT $NVDA

中文翻译

这只是我个人的观点。自上次财报电话会议以来，许多情况已发生变化： 1. 7纳米铟(Indium)原料价格在SMM开始飙升，我们看到亚洲供应链与西方供应链（$AXTI向西方出口，推高其尚未积压的未来利润率）之间出现分化。 12月19日：Asian Metal报道“铟锭供应商提高美国报价”，标志着西方定价特定分化的开始。 2025年12月24日：随后报道“铟锭供应商提高中国报价”，证实截至周末，定价压力已在国内和出口渠道固化。这影响磷化铟(InP)衬底生产商的底线，但$AXTI拥有整个成本结构/供应链，因此任何价格上涨仅有利于其利润率。 2. 超大规模数据中心/西方供应链已为衬底支付溢价。在$NVDA从$LITE、$COHR等公司锁定外延生长(EML)产能后，他们面临产能紧张。 -> 我们已看到$MSFT和其他超大规模数据中心部署放缓，可能源于产能问题。 -> 他们将向上游锁定材料（此前作为电信大宗商品极其便宜），然后这变成类似高带宽内存(HBM)的博弈论分配战。但除了$AXTI，你几乎找不到其他能大规模生产的公司。例如，$GOOGL尤其依赖于此，仅$MSFT的Maia项目若全面铺开预计将占用全球两位数产能，但实际根本不够。 3. 最大变化是日本InP衬底供应链受到出口管制，当住友/JX方面材料耗尽时，$AXTI实际上成为最大的InP衬底供应商。这纯粹是关于在光子学AI扩张中，对单一公司故障点价值的评估。我们可以争论语义，但如果它倒下，整个供应链也会崩溃。例如，CEO称在出口管制（及其1亿美元融资扩产）之前，他们占InP供应链的40%。这个整个AI供应链的巨大瓶颈市值仅为13亿美元，而一些无实质影响的量子公司市值却超100亿美元。很难对瓶颈进行建模，因为数亿总可寻址市场(TAM)突然用于每个AI部署，且无人知晓衬底或原料价格去向。

英文原文

So this is just my perspective on this. Lot of things have changed since the last earnings call, 1. Price of 7n Indium feedstock has started to spike on SMM and we're seeing bifurcation between asian supply chains + western ( $AXTI exports to western, which inflates their future margins that aren't already backlogged) Dec 19th: Asian Metal reported that "Indium ingot suppliers raise offering prices in United States," marking the beginning of the specific divergence in Western pricing. December 24, 2025: They subsequently reported that "Suppliers of indium ingot raise quotes in China," confirming that the pricing pressure had solidified across both domestic and export channels by the end of the week. This affects the bottom line of InP substrate producers, but $AXTI owns the entire cost structure/supply chain so any price increases just benefit their margins. 2. Hyperscalers/Western supply chains pay premiums for substrates already. They are capacity strained after $NVDA secured EML capacity from $LITE, $COHR, and others. -> We're already seeing slowdown for $MSFT and other hyperscaler deployments likely because of capacity. -> They will go upstream to secure materials (which are extremely cheap, bc earlier they were commodities for telecom), then this just becomes game theory allocation wars like HBM. But instead of sk hynix, mu, samsung, and the others. You basically only have $AXTI left that can do it at scale. EG. $GOOGL especially is heavily dependent on this, and $MSFT alone from maia was est. to take up double digits of the world's capacity if they wanted to ramp up, but there's literally not enough. 3. The biggest thing was that the Japanese InP substrate supply chain got export controlled, effectively making $AXTI the largest InP substrate provider when materials run out on the Sumitomo/JX side. This is just pure what value would you place on single company point of failure on the photonics AI ramp. We can argue semantics but if it goes down the entire supply chain does too. eg. CEO was saying they were 40% of inp supply chain before export controls (and before their $100M fundraise to increase capacity). And this huge bottleneck of the entire AI supply chain is valued at $1.3B, while random quantum companies with no material impact are worth $10B+. It's hard to model bottlenecks where the few hundred million TAM is suddenly used for every single AI deployment because nobody knows where the substrate or feedstock pricing goes.

在 X 上查看原推 ↗