Taking some time to share about my conviction in making $LITE my second-largest position, and in maintaining a heavy exposure to the photonics ecosystem. Pls do yr own DD

Spent the past two days down the optics rabbit hole, and the problem that it solves is fascinating. Side note: did you know that different colors of light travel at different speeds through fiber?

The thesis is simple: next-gen AI models are currently limited by the bandwidth and latency ceilings of the switches and cables connecting GPUs—perhaps even more than by the raw compute power of the GPUs themselves. Today’s infrastructure relies heavily on copper cables and electrical switches, which are: 1) power-hungry, 2) heat-intensive, and 3) limited in reach.

In fact, the reason you can only fit ~72 GPUs into a single rack (like the NVL72) is largely due to the physical limits of copper; electrical signals degrade too much if they travel any further.

Given the potential market share gains from replacing copper-heavy infrastructure, there is plenty of growth ahead for optical solutions.

Scale-out is the immediate oppy; scale-up is the mid-term oppy but which remains the 'holy grail' given the sheer volume of copper currently inside each rack. Scale-across will be the long-term driver.

Regarding $LITE, they are the dominant manufacturer of EML chips and a first-mover in next-gen optical tech like CPO (Co-Packaged Optics) and OCS (Optical Circuit Switching). To me (and my untrained eye), these look like the defining technologies of the next decade.

Producing these high-power, high-reliability laser chips is incredibly difficult; hence, these chips are the defensive moat of an optical transceiver. Lumentum and Coherent effectively operate as a duopoly, holding ~90% of the EML market. $LITE also holds a dominant share of the External Laser Source (ELS) market, which is a binary requirement for CPO— you simply cannot execute CPO without an external laser source.

I view CPO as existential for scaling GPU density and bandwidth. Reliability is a massive moat here. Customers are reluctant to design $LITE out because their products are battle-tested, stemming from decades of experience producing lasers for subsea cables — basically the harshest environment imaginable.

A mentor once told me: 'You can be contrarian on direction, but you can also be contrarian on magnitude.' I believe the adoption of optics will be far larger than what the market perceives today; it is the key to unlocking future model advancement.

Hence, this will be the incentive for hyperscalers, NVIDIA, and Broadcom to underwrite the risk to shift away from the copper-dominated links of today. And If you look at NVDA/AVGO pushing CPO and Google’s use of OCS, signs of that adoption are already here.

https://x.com/trendforce/status/1998060191946875298?s=46

Also added $AXTI long @ $12.11 to the optical basket

Pure-play producer of InP substrates (key ingredient for EML chips) to LITE. In very short supply now as EML chips are pretty much sold out till 2027 -- see link above.

Low cap play so inherently risky. A key risk here is that the company is susceptible to China’s export controls as it produces its raw materials through Chinese subsidiaries (Indium, Gallium, Germanium). AXTI has secured some export licenses for Indium for now, but we can never know if/when this will change.

Cut $CRDO long @ $178 for -1%

Just slipped short on it @ $165.94. Given the rich multiples and the inevitability of copper -> optics, r/r for short makes sense to me.

Added $TSEM long @ $121.72 to optical basket — currently includes:

$LITE $CIEN $COHR $TSEM $AXTI

Added a small long of $AMKR @ $46.54, think it will continue to benefit from packaging demand overflow. Bit late on this though I think

Rebuilt my short positions over the course of the week — unplanned but mostly at better prices than when I closed it all.

Largest size below, smallest above. Avg price = cost basis, last price = current price.

$ORCL commentary dragging down the AI capex plays today.

$ALAB taught me that when a fundamental cornerstone of your long thesis changes, you should not only exit your longs, but also flip short.

That lesson came in handy with the $CRDO flip from long to short this week, with rumours of NVDA bringing forward their CPO adoption timeline.

So much of investing is making mistakes and learning from them — and that’s precisely what makes it so fun and dynamic.

https://x.com/arronwei3n/status/1998570417624322397?s=46

Nice beat and guide from $CIEN

Degen-ed some $ORCL short-dated calls for Dec 26 $197.5 Strike for $7.6 premium. Very small position that I already wrote off mentally.

Feels oversold, and seems like a good hedge if market turns bullish on OpenAI again.

Closed most of my shorts the past 2 days. I think the bear doom regarding AI is getting too much — it’s *probably* not over (yet).

Now riding unfettered into the long (pun intended) night

$VRT @ $157-$162
$CRWV @ $72-$77
$CRDO @ $143-$146
$NVDA @ $176
$CVNA @ $457-$460

Have a feeling that $MU print tonight will make or break the negative AI sentiment; expectations are very high though so it’s a tough call but I’m staying long a medium-sized position into the print.

I’m hopeful (faith is inversely correlated with the # of times that OAI/ORCL appears in headlines) that the $MU print can be the catalyst that shifts market sentiment about AI.

MS and many others are already drawing parallels between Micron’s surprise beat-and-raise and $NVDA’s "big bang" guidance raise in May 2023 —the event that effectively ignited the multi-year bull run for Nvidia and the wider AI complex. The similarities in the headlines are striking:

NVIDIA (May 2023): "NVIDIA SEES 2Q REV. $11.00B +/- 2%, EST. $7.18B"
Micron (Now): "Micron sees Q2 Adj Rev $18.3B–$19.1B (est. $14.38B)"

Nvidia's inflection point fundamentally expanded investors' vision of what the GPU TAM could become, fueled by insatiable AI compute demand.

If this is indeed the "big bang" quarter for Micron, it forcefully brings a pivotal question into the spotlight: should Micron— and the broader ecosystem of historically cyclical and commoditised industries like memory and optics —continue to be penalized due to their extreme cyclicality? If so, then $MU near-single-digit fwd P/E while growing EPS at near triple digits sequentially is justified.

If not, then the alternative is that these companies should be re-rated as critical AI enablers that allow the AI cycle to continue scaling non-linearly. After all, we may only be in the first innings of multi-modal models’ capabilities; it is impossible to predict what their memory and storage requirements will look like five years— or even twelve months—from now. Pre-training scaling law is also still intact, while test-time compute and chain of thought reinforcement learning have simultaneously expanded the boundaries of model training — in essence, models are still getting smarter.

Ultimately, this depends on one’s long-term view of AI’s utility and how long we take to get there. How much will SamA/Elon/the hyperscalers chase? How motivated are they? Because right now, it still remains that the more $$$ that you throw at the models, the better of a LLM you will get — basically, we haven’t found the ceiling that money can’t fix yet.

So to borrow a thought from my $VST/$SNDK post above: is this time different? If it is, then we must be careful not to apply "pre-AI" frameworks to a post-AI world.

I guess we will have our answer soon, dictated by how Micron trades in the wake of this print.

Ofc if the market fades this print, then that’s 30 mins of my life spent typing this that I’m never getting back, and I will delete this name from my watchlist

Screw it, longed small position $CRWV @ $67.24

Stock is trading like default is certain? But hyperscalers will want to keep neoclouds around as they don’t want to onboard the GPU obsolescence risk themselves and hence seek to offload it to the neoclouds. I think some of the top-tier neoclouds will hence be kept alive by hook or by crook. Long r/r makes sense to me here

Per above, shorted more:

$DELL @ $128
$HPQ @ $23.6-$23.87
$QCOM @ $175.2

Sold covered calls for $93 strike on the $CRWV position — if it gets called it gets called!

Multi-modal models with the ability to create entire hours-long engaging movies from the ground up, supported by personal assistants with perfect, infinite memory.

A proof point of structural re-rating will occur when the market no longer punishes, but instead rewards, memory capex expansion. TBC

https://x.com/jonhernandezia/status/2001967472077918255?s=46

$MU $SNDK