$NVIDIA(NVDA)$earnings are in the rearview. Where does the AI capital cycle flow next? The diffusion map is clear — money is rotating down the stack across 5 stages. The alpha window is different at each one.
Stage 1: GPU & CPU
$NVIDIA(NVDA)$ + $Advanced Micro Devices(AMD)$. Where the AI wave started, where capital concentrated first. Valuations are high. Most retail investors entered late here.
Stage 2: Memory (actively running — highest alpha right now)
HBM demand surge, severe supply-demand imbalance, price and volume both rising. Per Trumoo: Anthropic's compute demand is growing 5x faster than storage capacity — storage is essentially locked in as the bottleneck. China's CXMT and YMTC are low-end capacity only; high-end HBM requires ASML EUV tools that can't be imported freely. That gap won't close. The three memory giants (Micron, SK Hynix, Samsung) have at least a year of stable runway ahead.
Stage 3: Optical Networking (just starting)
$COHERENT(COHR)$, $Semtech(SMTC)$ $POET Technologies Inc(POET)$ Data volume exploding, bandwidth demand surging, optical modules entering volume shipment phase. Bottleneck is visible but the ramp is early-stage.
Stage 4: Data Centers (spreading)
$Digital Realty Trust Inc(DLR)$. The physical home of AI compute — leasing demand strong, long-term visibility high. Less of a high-alpha play now, more of a stable core position.
Stage 5: Power / Energy (the next stop)
$Vistra Energy Corp.(VST)$, $Constellation Energy Corp(CEG)$. Power gap is widening. At the end of the AI compute chain, there's just electricity demand. Long-term certainty is high. The window hasn't opened yet — but the direction is clear.
Where are you in the stack?
$MU$ as the most underappreciated bottleneck — do you buy the "locked-in supply gap" thesis, or is the memory cycle ceiling close enough to worry about?
Anthropic compute 5x storage growth — is $MU$'s alpha window still wide open? Does optical get the next rotation?
Do you position now ahead of the power crunch, or wait until AI data center power demand becomes undeniable?
Leave your comments to win tiger coins~
Comments
Why?
Micron has triggered a high margin supply constraint across the entire global tech grid. This is because an AI chip requires 3 times the wafer footprint of a standard phone or PC chip.
With capacity 100% spoken for through 2026 and 2027 orders locked in, Micron's 7.7x Forward P/E ratio means that investors are essentially buying an enterprise monopoly at a deep discount.
@Tiger_comments @TigerStars @Tiger_SG
I also believe optical networking could be the next rotation. As AI workloads explode, bandwidth demand will surge, making companies like $COHERENT(COHR)$ and $POET Technologies Inc(POET)$ increasingly interesting. AI is no longer just about chips — it’s about the entire infrastructure stack.
For power, I think the trend is inevitable but slightly later-stage for me. Companies like $Vistra Energy Corp.(VST)$ and $Constellation Energy Corp(CEG)$ could benefit massively once AI-driven electricity demand becomes undeniable.
@Tiger_SG @TigerStars @Tiger_comments @TigerClub
The supply gap is structurally real due to the physics of High Bandwidth Memory (HBM), making this cycle different from historical commodity DRAM swings.
The HBM Wafer Penalty: HBM3E and HBM4 require roughly 3x the wafer capacity of standard DDR5 for the same bit output. This massive consumption of industry cleanroom space structurally suppresses the supply of standard PC and server DRAM, keeping industry-wide inventory tight.
Locked-In CapEx: Major hyperscalers secure HBM capacity via non-refundable down payments up to a year in advance. This severely limits the downside risk of a sudden demand collapse in the near term.