5 Things Failing in AI — And the Technology That Fixes Them All
AI正在失败的五件事——以及修复它们的技术
Copper hits a physics wall at 1.6 Tbps. The fix is silicon photonics — and $7B moved into it in a single week.
铜线在1.6Tbps撞上物理墙。修复方案是硅光子——一周内70亿美元涌入这个领域。
Base Layer — The Giants
基石层——巨头
The physics wall
物理墙
At 800 Gbps, copper barely works. At 1.6 Tbps, it falls apart — too much heat, too much power, signal degrades exponentially. Silicon photonics replaces electrical signals through copper with light through glass pathways etched into silicon chips. Same manufacturing process the semiconductor industry already knows.
800Gbps时铜线勉强可用。1.6Tbps时它崩溃——热量太大、功耗太高、信号指数级衰减。硅光子用刻在硅芯片上的玻璃通道中的光替代铜线中的电信号。半导体行业已经掌握同一种制造工艺。
Middle Layer — The Foundries
中间层——代工厂
TSMC just validated the entire space
台积电刚刚为整个赛道盖章
By announcing a dedicated photonics production line for 2026. When the world's largest foundry builds an entire line for a technology, that's the ultimate confirmation signal.
宣布2026年专用光子产线。全球最大代工厂为某项技术建整条产线——这就是终极确认信号。
Top Layer — Pure Play Photonics
顶层——纯玩光子
These three live or die by the photonics transition
这三家公司的生死取决于光子转型
Highest upside, highest risk. Coherent makes the full stack. Lumentum makes the lasers everyone needs. Marvell makes the DSP chips that interpret the light.
最高上行空间，最高风险。Coherent做全栈。Lumentum做所有人都需要的激光器。Marvell做解读光信号的DSP芯片。
Lumentum is the cheapest stock in this universe
Lumentum是这组中最便宜的股票
With a PEG of 0.61. Its lasers go inside everyone's products — including competitors'. It wins regardless of who takes transceiver market share. Marvell at PEG 0.64 is the second-cheapest, but carries GAAP losses from the Celestial AI acquisition amortization.
PEG仅0.61。它的激光器装进所有公司的产品——包括竞争对手的。谁赢得收发器市场份额它都赢。Marvell PEG 0.64排第二，但GAAP亏损来自Celestial AI收购摊销。
$100 Allocation
$100 配置
Brian's allocation skews asymmetrically toward the mega-caps
Brian的配置不对称地偏向大市值公司
With less in the higher-risk pure plays. The base layer gets the most capital because these companies win even if the photonics transition is slower than expected.
纯玩高风险公司占比较少。基石层获得最多资金，因为即使光子转型慢于预期，这些公司也会赢。
The Bottom Line
底线
Copper is hitting a physics wall. Every major hyperscaler knows it. Over $7B moved into photonics in a single week. Silicon photonics isn't speculative — it's the next foundation layer of AI infrastructure. The only question is which companies capture the most value as this transition accelerates.
铜线正在撞上物理墙。每个大型超算中心都知道。一周内超过70亿美元涌入光子领域。硅光子不是投机——它是AI基础设施的下一个基础层。唯一的问题是：转型加速时，哪些公司捕获最多价值。