Your $NVDA isn’t the bottleneck anymore. And that should scare you.
AI is about to hit a wall.
Not compute.
Bandwidth.
Data centers are quietly breaking → copper can’t handle the load
The shift has already started: electrical → optical (photonics)
Here’s the part no one is talking about:
You can have the best GPUs in the world… But if data can’t move fast enough? They sit idle.
Burning cash. Doing nothing.
Everyone crowded into: $NVDA, $SMCI
But the next leg of the trade is forming underneath them:
- Optical networking
- High-speed interconnects
- Photonics
This is where AI either scales… or stalls.
New AI stack (that most people are late to):
- Brain: $NVDA
- Body: $SMCI
- Nervous system: Photonics
Miss the third layer → you miss the next move.
The money flow is shifting to:
- $AVGO, $MRVL backbone players
- $LITE, $COHR, $AAOI pure optical
- $CRDO, $ALAB hidden winners
- $TSEM, $TSM, $AEHR picks & shovels
What people still don’t get:
- AI isn’t just a chip story anymore.
- It’s a data movement + energy problem.
- And both are becoming bottlenecks… at the same time.
Translation:
- The biggest winners of the next cycle
aren’t the ones you already own.
- They’re the ones fixing what’s about to break.
$GOOGL, $MSFT, $META & $AMZN just confirmed $600B+ in 2026 CapEx.
Here’s how that money flows through the AI stack:
GPUs, Accelerators
• $AMD provides second source AI GPUs with MI300 & MI400 series
• $NVDA powers the AI accelerator layer with Blackwell shipping & Rubin in production
ASICs, TPUs, Design Services
• $MRVL provides interconnect & custom silicon for AI racks
• $ARM supplies CPU architecture across hyperscaler server fleets
• $CDNS enables hyperscaler AI chip development through EDA tools
• $AVGO builds custom AI silicon & networking ASICs for hyperscalers
• $SNPS powers hyperscaler silicon roadmaps with chip design software
Memory
• $MU supplies HBM & advanced DRAM for AI accelerators
• $SNDK supplies NAND flash for hyperscaler storage layers
• $WDC provides high capacity storage for AI data pipelines
Networking
• $CRDO supplies high speed SerDes for AI interconnect
• $COHR provides lasers & photonics for optical networking
• $ALAB delivers connectivity silicon for rack scale AI systems
• $ANET powers high speed switching inside hyperscaler AI clusters
• $GLW supplies fiber & optical materials for data center connectivity
Foundry & Advanced Packaging
• $KLAC process control for yield & reliability
• $LRCX etch & deposition tools for chip scaling
• $TSM advanced logic manufacturing for AI chips
• $AMAT wafer fab equipment for advanced nodes
• $ASML EUV lithography enabling leading edge nodes
• $AMKR advanced packaging & test for AI accelerators
• $INTC domestic foundry & advanced logic manufacturing for AI chips
AI Utility Layer
• $IREN builds down into power + compute
• $NBIS builds across the software + developer layer
• $CIFR rents out the physical layer through long-term leases
• $CRWV provides on demand GPU cloud capacity for AI workloads
Energy & Power
• $FSLR adds incremental utility scale solar capacity
• $CEG delivers nuclear baseload for large data center clusters
• $NEE builds large scale renewable capacity for grid expansion
• $VRT controls cooling & power systems inside AI data centers
• $VST supplies dispatchable power for hyperscaler grid demand
• $BE provides on site power that bypasses the grid for AI clusters
• $EOSE enables long duration storage for 24/7 data center power
• $SMR develops small modular reactors for future nuclear buildouts
• $OKLO provides “local nuclear plant” that the AI economy will require
• $GEV supplies turbines, transformers & grid hardware for grid rebuild