GM. This is Milk Road AI, the newsletter trying to stay one bottleneck ahead of the market.
The AI trade has already gone through three chapters. In today's edition, we break down what we think chapter four looks like and the two stocks we're watching.
Speaking of staying ahead of bottlenecks, here's one Plasma already solved: spending your stablecoins in the real world.
Plasma One lets you spend stablecoins anywhere Visa is accepted and Milk Road readers get $10 back on their first $100 of spend. Start spending in minutes on iOS and Android.
CHAPTER FOUR OF THE AI TRADE π
The first three chapters of the AI trade:
- GPUs with Nvidia.
- Memory with Micron and SK Hynix.
- Neoclouds with Nebius and CoreWeave.
Melvin (our lead analyst at Milk Road PRO) thinks chapter four is Photonics.
This is the light that moves data between chips.
Before your eyes glaze over, here's the simple version.
AI models need tens of thousands of GPUs working together. The problem is that copper cables are becoming too slow to connect all of them.
So, the solution is to move data with light with photonics.
Melvin has two names in this space.
1. Applied Optoelectronics (AAOI)
AAOI builds the optical modules and lasers that move data around AI data centers.
Their real edge is that they grow their own lasers in-house.
Most competitors rely on outside suppliers but AAOI controls its own manufacturing. That's becoming a major advantage while the industry faces shortages of high-speed lasers.
Demand is already showing up in the numbers:
- 400G shipments are up 10x YoY.
- One hyperscaler placed $124M of 800G orders in a single month.
- 1.6T products are now entering production.
- Management says demand exceeds capacity through mid-2027.
- AAOI's slice of the market expands every year through 2030. π
2. Marvell Technology (MRVL)
Marvell is a much broader AI infrastructure play.
While AAOI builds optical hardware, Marvell designs many of the chips that make AI networks actually work.
Their business now spans:
- Optical networking.
- Ethernet.
- Connectivity chips.
- Custom AI chips.
- Memory interconnects (CXL).
Wall Street expects Marvell's revenue to almost triple again by 2029.
Whether Amazon, Microsoft or Google buy Nvidia GPUs, they still need Marvell to connect everything together.
Marvell is becoming the "connectivity layer" of AI.
Management believes the industry's bottleneck has shifted from compute to connectivity and that's exactly where Marvell operates.
Jensen Huang has even publicly called Marvell a future trillion-dollar company.
The bottom line
How Melvin ranks the three sectors he's covered recently from safest to riskiest over a 36-month horizon:
- Neoclouds (safest).
- Memory (second).
- Photonics (highest risk, highest reward with AAOI being the most volatile name of the group).
We were early to the memory trade with MU, early to the neocloud trade with NBIS and CRWV and now, we will be early to the connectivity trade with MRVL & AAOI.
If you want to see every position our analysts own, every trade they make, and the full investment thesis behind each one, try Milk Road PRO for a buck.
View the portfolios before the next AI trade takes off.
THE BEST WAY TO SPEND STABLECOINS
If you're already holding stablecoins, Plasma One is the missing piece.
Plasma One is the global money app built for people who earn, save, send, and spend across borders :
- Stablecoin Visa card: spend crypto anywhere in the world
- Get started on iOS or Android: Get your virtual card and start spending in minutes with Apple Pay or Google Pay.
- Earn on your balance: your idle dollars work while you sleep
- Boosted Rewards: Earn up to 4% cashback on every purchase, with boosted rewards on AI tools like ChatGPT, Claude, and Cursor.
Milk Road reader deal: spend $100, get $10 back with code MILK26.
