Learn from history to enhance your understanding, review it daily

This is for my own eyes, to help me stay confident and hold on to long-termism. $Sandisk(SNDK.US) $Lumentum(LITE.US) $Micron Tech(MU.US) $Vertiv(VRT.US) $AXT(AXTI.US)

I. Learning from History: The "Infrastructure Law" of Every Tech Wave
Looking back over the past 30 years, the capital flow of all disruptive technology waves has followed the same iron law:
Before applications become widespread, infrastructure must come first. The shovel sellers always make the first pot of gold, and the valuation premium will last throughout the entire "capacity ramp-up period".
Wave Stage Representative Years
PC/Internet Infrastructure Period 1993-2000
Mobile Internet Infrastructure Period 2007-2015
AI Smart Infrastructure Period 2023-2028+
History repeatedly proves: during the infrastructure explosion phase, "expensive" is the norm, while "cheap" is actually a signal of logical falsification. The market is always pricing future demand with past profits, and the real main upward wave precisely occurs in the 18-36 months where "valuations seem like a bubble, but performance is delivered at a 50%+ CAGR".

🌍 II. The Space for AI Infrastructure: Where Exactly Are We Standing?
You mentioned an extremely crucial perspective: "The current supply and demand we see is only the demand from already-transformed industries; the infrastructure demand from untransformed industries has just begun to awaken."
This is precisely the biggest difference between AI and the internet/mobile internet:

1. Penetration rate is still in single digits: Global enterprise AI adoption rate <10%, the AI adoption rate in traditional manufacturing, healthcare, energy, logistics, agriculture, and finance is generally <5%. This means 90% of the computing power/storage/network/electricity demand is still on the way.
2. From "Training" to "Inference" to "Edge":
   • Stage 1 (Now): Cloud-based large model training → consumes HBM + optical interconnect + liquid cooling + gigawatt power grids
   • Stage 2 (2026-2028): Enterprise-level inference + Agent workflows → consumes enterprise SSDs (SNDK) + memory expansion (MU) + proprietary networks (LITE/AXTI)
   • Stage 3 (2028+): Edge AI + robots + autonomous driving → consumes edge computing chips + low-power optical communication + distributed microgrids
     The switch between each stage is not a replacement, but an addition. Infrastructure demand is exponentially additive.
3. Capacity Barrier = Pricing Power = Profit Accelerator:
   • HBM capacity is limited by CoWoS and yield rates → Micron/SK Hynix continue to raise prices by 15-20%
   • 1.6T/3.2T optical modules are limited by InP substrates (AXTI/LITE) → supply falls short of demand, ASP remains firm
   • Data center power/cooling is limited by transformer/liquid cooling chains (VRT/BE/ETN) → orders are backlogged to 2029, gross margins structurally shift upward
     This is not a cycle; this is a structural shortage. As long as the shortage exists, the valuation premium will not disappear; it will only be quickly digested by earnings growth.

📈 III. The Reconstruction of Valuation Logic: Looking Forward 1-2 Years
The method you mentioned, "forecasting future space based on earnings report growth rates," is the core method for institutional pricing. Let's now use this perspective to remap our holdings:
Target Current Market Valuation (Static)
MU PE ~18x
LITE PE ~40x
SNDK PE ~50x (cycle peak)
AXTI PS ~8x
VRT PE ~90x
Core Conclusion: If you look at today's profits, they are all "expensive"; but if you look at 2027 profits, their current prices are only "reasonable". The valuation of AI infrastructure stocks has never been "killed" down; it has been caught up by "earnings". Every earnings season's Beat & Raise will cause the static P/E ratio to collapse instantly, and then the stock price will start a new round of upward movement. This is the compound interest code of trend trading.