A few people asked me what my hypothetical AI infrastructure portfolio would look like over the next 6 to 12 months
Honestly, probably something like this:
$SIVE $SIVEF 14% - The cleanest AI photonics choke point I’ve found. Every major CPO architecture needs CW lasers, and Sivers is the sole Ayar supplier right now. Jabil basically said demand is uncapped. CHIPS Act exposure and a NASDAQ listing are nice bonuses on top
If they can execute this could be a wild ride
$AXTI 9% - InP substrates are one of those nobody talks about it but everybody needs it parts of the stack. Backlog already above $100M and they’re doubling capacity twice before 2028. Hyperscalers are clearly trying to lock supply early
$AAOI 9% - One of the few real American made 1.6T transceiver stories. They control their own InP lasers internally, 800G is ramping hard, and 1.6T qualification starts in Q3. The datacenter revenue target for 2027 is massive if they execute
$SOI 8% - Upstream pick and shovel for silicon photonics. Every silicon photonics chip starts with these wafers. Direct leverage to Tower’s contracted photonics ramp
$LPK $LPKF 6% - Probably one of the more overlooked names here. They dominate the laser processing side for glass substrates and already have most of the important customer qualifications. Production orders should start hitting later in 2026
$COHR 4% - Fully vertically integrated from substrate to optical device. NVIDIA taking a $2B stake changed the perception completely. 6inch InP progress is ahead of schedule and CPO revenue should start showing up in H2 2026
$TSEM 4% - Open foundry exposure to basically every CPO architecture. $1.3B of contracted photonics revenue for 2027 and most capacity already reserved through 2028 tells you demand is very real
$GLW 4% - The actual fiber layer underneath AI infrastructure. NVIDIA warrant deal, hyperscaler LTAs, and a $10B photonics roadmap by 2030. Not flashy, but critical
Fittech (TPE: 6706) 4% - Smaller Taiwan ecosystem play tied directly into the photonics buildout. Exposure to FOCI and Browave ramps without needing to own the obvious names.
$PENG 4% - One of the more interesting enterprise AI infrastructure names. Exposure to sovereign AI, neoclouds, and enterprise AI factories. Their MemoryAI CXL positioning looks unique right now and NVIDIA has openly recommended them
KRX:000660 SK Hynix 3% - Best HBM company in the world in my opinion. Margins are insane, DRAM pricing is exploding, and they’re basically sitting at the center of the memory bottleneck
KRX:402340 SK Square 2% - Simpler thesis here: discounted exposure to SK Hynix through the holding company structure
FOCI (TWO:3363) 3% - Sole FAU supplier for TSMC COUPE Gen 1 and Gen 2. Mass production starts H2 2026 and even Jabil hinted demand could exceed supply.
NCI (TWO:4092) 3% - Hidden materials choke point. Near monopoly on 7N red phosphorus which feeds directly into InP production. Tiny valuation compared to the scale of demand coming
Nextronics 8147 3% - Goldman confirmed their CPO connectors are already inside NVIDIA’s supply chain. Still very underfollowed for how important the positioning is.
Win Semi 3105 3% - Key manufacturing partner for scaling Sivers’ CW laser production. Mid 2026 qualification could remove a major bottleneck
Shunsin 6451 3% - Foxconn’s CPO packaging arm. Already shipping AI racks directly to NVIDIA and there are rumours Broadcom orders are coming too
$AIXA 2% - Pure-play equipment leverage to InP laser demand. If optical scaling keeps accelerating, they sell the picks and shovels needed to expand production
HB Tech 078150 2% - Optical inspection exposure to the glass substrate story. Supplier into SKC Absolics and benefits if AMD qualification gets finalised
SKC 011790 2% - Glass core substrate exposure. AMD qualification is reportedly in final testing and Phase 2 investment plans are getting serious.
MSSCORPS 6830 2% - One of the weirdest but most interesting names here. They handle optical loss detection for silicon photonics and effectively sit inside the CPO yield process.
$ALRIB 1% - MBE equipment angle for quantum dot lasers and silicon photonics. Small position but big upside if the ROSIE Tier 1 order lands.
$IQE 1% - InP epi wafer supplier with strategic review and possible M&A optionality. Feels like the market still isn’t paying attention
$NBIS 1% - Pure neocloud exposure. Inference demand sounds completely parabolic right now and pricing power has shown up almost overnight.
$LITE 1% - Obvious beneficiary of NVIDIA’s multi-year optical agreements. Could quietly become one of the biggest winners if CW laser demand keeps surprising higher
$NOK 1% - AI RAN and optical transport exposure. The NVIDIA partnership is real and telecom AI infrastructure feels underappreciated
$AEHR 1% - Owns a very niche but important layer: wafer-level burn-in for PICs. Exposure to both photonics and power semis
$VIAV 1% - Optical testing and validation. Every serious 1.6T deployment eventually runs through VIAVI gear somewhere in the process
I know it looks crowded with a lot of positions, but I’m trying to be realistic about how I’d actually invest
It’s easy to say just go all in on one name but for most people that’s way too binary and mentally exhausting
So instead I’d spread exposure across the parts of the stack I think matter most ie photonics, optical networking, substrates, packaging, memory, testing, foundries, AI factories etc
Basically the companies helping build the physical backbone AI actually runs on
Not financial advice obviously. Just how I’m personally thinking about the space right now
What would you add ?
I am digging into $POWI and $HLIT this weekend
BREAKING: We did it.
We built a trading companion connected to real-time market data, straight from the source.
Meet Mr. Whale.
He knows what happened to your favorite stock 2 minutes ago.
And how to use our tools to find your next favorite stock.
Ask him anything today.
Everyone's thinks Leopold's latest 13F is bearish AI. But they're wrong.
he just told us where the next major AI bottleneck is: Power and memory infrastructure... and he's placing a massive $3.5B bet on it.
The thesis is simple, but the trade is less obvious:
-> The AI semiconductor trade is too crowded. Nvidia, AMD and Intel are up 200%+ over the last year.
-> The next unlock? Powering the GPUs... his massive bloom energy and corweave prove it
But there's also a genius move he's executing with corweave and bitcoin mining...
We covered it all on todays episode 👇
$PENG is a $2.5 billion market cap stock.
Let me show you who they work with.
Partners: $NVDA. $AMD. $INTC. SK Hynix. SK Telecom. Shell. Dell Technologies.
Customers: NASA. The US Air Force. The US Navy. Lockheed Martin. Boeing. Dolby.
Read that again.
A $2.5 billion company that NASA trusts to build their compute infrastructure. That the US Air Force and US Navy rely on for mission critical systems.
That Lockheed Martin and Boeing have chosen as their systems integrator. That partners directly with the most valuable chip companies on earth.
These are not logos on a website. These are active relationships with the most important institutions and most valuable companies in the world. Organizations at this level do not work with vendors they do not completely trust.
And the market still has not priced in the $MRVL photonic memory partnership. The Photonic Memory Appliance being built right now. Zero dollars of that revenue in any model today.
$2.5 billion market cap. Partners and customers that most public companies would never land in a lifetime.
The market has not figured this one out yet.
@MisterMCAP Take a look at $AXTI everyone thought it was gonna happen if a Cell off after earnings and it did not probably you’re gonna see the same thing
For sake of transparency. Im not gonna hold $SIVE on earnings. Simply because the broad retail owner base has shown extreme short sightedness when it comes to the earnings of the photonic sector.
Basically Is predict earnings day sell-off.