With “The Buildout” I’m looking for changes in architecture, and niche companies enabling those changes. Hard-to-qualify component manufacturers, and the like.

Semis are having a rough week. The market attributes the weakness to the reporting on OpenAI’s revenue, which is apparently at $50Bn annually, instead of $70Bn.

I think its more of a function of long-term interest rates and the war in Iran but who is to say? Could also just be a healthy correction from absurd prices, and then the market attaches whatever narrative it wants to that price action.

What is apparent is that semis and the big AI-exposed names, will be rife with volatility. Why wouldn’t they be?

Nobody can reliably predict how long it will take for AI to become embedded into the economy, and nobody knows what the ultimate magnitude of buildout is required- because use cases keep appearing.

Agents like Muse from Meta are in early adoption.

Video models are becoming more practical day by day.

Most enterprises still use co-pilot (which is shit).

Robotics are on the cusp.

The Sci-Fi future we saw in movies is playing out right before our eyes.

Is the buildout over? Hell no- it will just keep rotating to the next use case. Things are changing. Fast. There is opportunity in this volatility.

ASML reported today and the CEO noted:

“Our customers continue to increase their short- and medium-term equipment demand forecasts with us, driven primarily by their views on the size and duration of AI-related investments. In light of their demand visibility, some customers are expressing an interest in placing orders with us for 2029 and even 2030.”

Placing orders for 2030. Incredible.

“With such strong visibility and an order book momentum that continues to increase quarter on quarter this year, we now expect that our supply as currently planned for both 2027 and 2028 will not fully meet the demand of our customers.”

Datacenter liquid cooling

I’m particularly interested in the liquid cooling space. It’s not as followed as memory, or optics, and I would like to become more familiar one some of these names before they pull a NHK Spring and their share prices launch into the stratosphere.

There are a lot of architecture changes happening within the datacenter. And with those changes come opportunity.

Fositek (6805 TW)

Fositek is probably my favourite name, but not at this price. It makes hard-to-qualify QDs which I mentioned in the previous buildout here.

At a forward price to earnings of 26x, its very expensive. As always with these kinds of ‘bottleneck’ plays, you need to be very sure of the duration of the bottleneck. And with capacity coming online from Fositek, and every Fositek competitor, it’s not really something you can be sure of. Best to wait for a real dip.

Another name I am adding to my watchlist:

BTL, Inc (6840 TW)

BTL is a potential bottleneck beneficiary of the transition toward 800V DC power architectures, megawatt-scale AI racks and liquid cooling.

These systems require increasingly complex validation and certification, supported by specialised, capital-intensive testing facilities that cannot be replicated overnight.

BTL has already secured 800V/1MW power-rack testing orders, reportedly has AI-related testing booked into Q1 2027, and is planning a new 2.5MW-capable facility. Testing revenue could grow faster than underlying AI server volumes as power density and system complexity increase.

It’s interesting because the financials look dismal, but they are securing orders and getting positioned for the demand.

In other news:

I have been absoluting scouring small-caps in Taiwan and Korea. As well as going through <$100m companies in the US to see if there’s anything new that has potential. I constantly scan the names there, I know many of them quite well. It’s how I found HRTG…didn’t stay sub $100 for long unfortunately. I am confident there will be another.

We are seeing High-End chips like NVIDIA, AMD, and Google driving the adoption of liquid cooling.

Trendforce reports that:

• Liquid cooling penetration among AI chips is projected to rise to 53% in 2026 and approach 60% in 2027.

• High-end chips from NVIDIA, AMD, and Google are driving the adoption of liquid cooling. NVIDIA’s rack-scale solution shipments are expected to double in 2026, while more than 80% of Google’s AI servers already use liquid cooling.

It’s an interesting space to keep an eye on, because we are also seeing evidence of much higher content per AI rack.

And who benefits in this liquid cooling adoption?

Names to watch

I don’t have positions in any of these.

A QD = quick disconnect. In an AI liquid-cooling system, it is the small connector that lets you join or separate coolant lines quickly without spilling liquid.

A QD costs almost nothing relative to the AI hardware it protects, but failure is catastrophic.

small component + mission critical + hard qualification + low tolerance for failure + rapidly rising unit demand

Fositek looks akin to the NHK Spring in that it manufactures this niche component- and yea, the market knows it. It trades at 26x Forward PE.

Fositek 6805.TW ~26× fwd

QD / UQD

H1 2026 server products = 55.7% of revenue, up from just 19.2% a year earlier. Full-year AI-related exposure estimated ~60%. Company does not separately disclose QD revenue. QD capacity is near full and being expanded ~50% in Q4.

NVIDIA GB200 to GB300: roughly +70% to +125% QD content.

Capacity at full & expanding, AI Racks using more QDs. Cheap component relative to full setup.

Constraints may mean price pressure comes next? (and then margins increase) - I’m postulating. Fositek seems to have the purest torque to this.

Subscribe now

Auras 3324 .TWO ~23×PE

Cold plates, vapor chambers, manifolds, CDUs, server thermal modules

Q2 2026: server products 78% of revenue, of which water-cooling products were 55% of total company revenue.

Lotes 3533.TW ~20×PE

QD was only about 1% of August revenue, with secondary estimates suggesting it could approach ~3% by year-end as shipments ramp.

Lotes is interesting as optionality because they are exposed to many different niches in the AI Datacenter boom. They create many CPU components too for example.

Tsugami Corp. TSE:6101 15X fwd

Makes high-precision CNC automatic/Swiss-type lathes, machining centers, grinders and rolling machines. Its machines mass-produce small precision metal parts including connectors and fittings.

Its Swiss-type lathes are exactly the class of machine used to mass-produce small, tight-tolerance metal parts such as QD bodies, valve bodies, fittings, sleeves, connector components and other turned cooling hardware.

Tsugami Corp. is probably my favourite. Management is guiding flat profits for 2027 and I think they are being conservative- sandbagging. Tsugami China is literally saying the opposite (Tsugami receives royalties from China)

Faster than Cerebras?

If we take Volantis’s disclosed A-1 targets at face value and rank the systems specifically for low-latency frontier-model inference, then A-1 would be #1 on paper.

Inference: Up to 10,000 tok/s/user on >20T models

Memory: 10 TB @ 240 TB/s

Volantis just receieved $88m in funding. Sounds small-fry, but what they are working on is game-changing. And they say they are shipping to customers next year.

What problem has Volantis solved?

In photonics, the “memory wall” refers to the same fundamental problem as in electronic computing: compute capability is improving faster than the system can move data to and from memory.

Let’s take a look at the NVIDIA Rubin for example. The chip is absurdly powerful, so much so, that the engineering challenge has become “how do we move the data from the Rubin to the High-bandwidth-memory (HBM) and back, fast enough to keep the GPU from idling and wasting precious compute?

The problem is visually illustrated in the image below- you simply don’t have space for more memory chiplets next to the GPU because you are limited by the reach of the current (copper) electrical links.

These electrical links are microscopic copper interconnects. Put simply, if you try make the copper links longer, you get worse performance. That forces the memory to sit extremely close to the compute die, limiting how many high-bandwidth memory devices can surround it.. Less than you would like.

Here’s an analogy: Lian Yunzhi is a 6 year old speedcuber from China who recently broke the women’s Rubik’s Cube speed record- at 3.549 seconds.

Her brain is a NVIDIA Rubin, and her brain can actually solve the cube in 2 seconds flat.

The process looks like this:

• Lian’s brain — the Rubin GPU — can solve the cube in 2 seconds.

• But the information it needs is stored in HBM — her working memory.

• Her brain constantly has to pull data from HBM, process it, and send the results back.

• Meanwhile, her hands are waiting for the next instruction.

So where’s the problem?

Her brain can think faster than information can move between her brain and her memory.

That gap is the memory wall.

NVIDIA is trying many different solutions to the memory wall too, by using a lot more copper, but not in the way Volantis is attacking the problem- by using mature VSCEL technology.

Coherent is also developing wide-and-slow VCSEL-array architectures for AI scale-up interconnects.

The Volantis solution: integrated micro-VSCEL arrays and an optical memory fabric. If you can move data between the GPU and memory fast enough, more of Rubin’s enormous compute capacity can actually be used. (I don’t have confirmation for whether Volantis is using a Rubin, just using it as an example here).

The image on the left, below, shows the eletrical-interconnect memory wall problem.

On the right- Volantis' solution- you can attach vastly more memory because optical links relax the short-reach limitation of the electrical memory interface.

Who are the public winners?

Read more

AMS Osram has a few things that look ripe for a speculation.

-It recently announced its strategy to develop complete integrated photonics engines for AI interconnects.

-NVIDIA Hire now VP & GM, Optical Interconnects.

-Much of the underlying micro-emitter, CMOS-integration, wafer-manufacturing and packaging platform was already proven in automotive. (Not NEW Tech).

-Exposure to AR/VR - a fast growing industry on the cusp of a real blowout.

-Trades at forward EV/EBITDA of 8.4x. (adjusted- big on adjusted)

And I think, most importantly for a speculation, it has similar characteristics to Sivers- optics ‘bottleneck’, growing social media presence, analysts changing their assumptions of the technology & the stock… Again, this is speculation on my part, but it feels ripe…

The game just changed for ams OSRAM

Until very recently, the AI case for ams OSRAM was interesting but also fairly easy to dismiss. The company has worked on MicroLEDs and VCSEL for a long time.

As a generalist, I won’t pretend to be an expert on the technology. There are plenty of Optics accounts on X, and Substack, who I rely on to form a view for a technology I can only broadly understand.

What matters mostly here is the VCSEL- its a lazer.

One of the analysts I rely on when dipping my toes into optics is .

Has long HATED MicroLED, calling it a hoax. His opinion on VCSEL was more nuanced, but basically described it as approaching the end of the road in January.

But then Ashkan Seyedi (from NVIDIA) joining AMS OSRAM materially changed his view.

According to Irrational Analysis, Seyedi came from Nvidia's ring-modulator side of the internal photonics debate and had publicly criticised MicroLED approaches. So seeing someone with that background move to AMS OSRAM made him think there must be something interesting happening around VCSEL CPO/NPO rather than MicroLED.

His thesis went from: “VCSEL CPO might technically work but probably isn’t the winner.”

to:

“Wait… AMS OSRAM may possess exactly the integration technology required to make the architecture commercially viable.”

Now, for generalists like me who find it difficult to keep track of all these acronyms…his current position is essentially that ams OSRAM’s VCSEL technology might have a legitimate place in the datacenter. More specifically- a legitimate role, in optical interconnects, especially in short-reach CPO/NPO architectures.

Furthermore, another specialist , is also constructive on microVCSEL for short-reach AI interconnects..

How did we get here? How did Osram get here?

The company had a lot of the right technology. It knew how to manufacture microscopic light emitters at scale. It had spent years developing microLEDs, VCSELs, optical sensors and advanced packaging. Some of that expertise had already been industrialised in automotive products such as EVIYOS, where thousands of individually controlled microscopic emitters are manufactured in volume.

OSRAM benefits from mature manufacturing know-how.

One of the emerging problems inside enormous AI clusters is no longer just producing more compute. It is moving enormous amounts of data between GPUs without burning an equally enormous amount of power.

Copper is beginning to struggle.

Optics is moving closer and closer to the processors.

And ams OSRAM realised that the same ability to manufacture huge arrays of microscopic light emitters could potentially be used to move data instead of illuminate a road.

That was the first breadcrumb.

But the original investment case was still relatively modest:

Maybe ams OSRAM could become an important supplier of the tiny lazers that sit inside these new optical links.

Interesting.

But ultimately still a component supplier.

Then the breadcrumbs started getting bigger.

In May 2026, ams OSRAM announced that it had signed a development agreement with a leading AI data-centre infrastructure partner and had formally begun product development for optical interconnects. Somebody on the customer side was apparently interested enough to start developing the technology with them.

Around the same time, something else happened that deserves more attention. ams OSRAM brought in Ashkan Seyedi to run Optical Interconnects. Seyedi wasn’t coming from lighting.

In 2024 he was Director of LinkX Products at NVIDIA, working specifically on high-bandwidth, energy-efficient optical interconnects for next-generation AI processors. Before NVIDIA he had worked on photonics at Intel and Hewlett Packard Enterprise.

In other words, the guy now running ams OSRAM’s optical-interconnect business had previously been sitting inside NVIDIA thinking about essentially the exact problem ams OSRAM was now trying to solve.

Then came August.

ams OSRAM announced that it wasn’t stopping at the light-emitting side.

It had started developing micro-photodiode arrays which are the components that receive the light at the other end of the fibre and turn it back into an electrical signal.

Management explicitly described this as “expanding BoM (Bill of Materials) coverage.”

That is a deceptively important phrase. Because suddenly the story was moving from:

We make the lazer.

to:

We make the lazer... and perhaps the receiver as well.

And on 1 July the company had already established dedicated Digital Photonics business lines to accelerate development and commercialisation.

At this point the pieces were beginning to form a picture.

Then came 21 September.

You follow the breadcrumbs and eventually end up standing next to the loaf.

ams OSRAM finally told investors explicitly what it is trying to build. At ECOC 2026 it unveiled a new Thin-Film microVCSEL platform designed specifically for AI scale-up networks.

The most important sentence:

The company’s stated goal is to combine what is has into what it calls a “complete and optimized photonics engine.”

ams OSRAM said it is now developing:

And Seyedi described ams OSRAM’s role as:

That changes the investment case.

Selling an integrated photonics engine is very different to selling a component of it. They want to sell the whole pie. If they can sell the whole pie, they can charge a higher margin for it.

We had two recent filings for Worldex. Basically, net-bullish for the governance thesis, but modestly so.

They related to a 17 September 2026 extraordinary shareholder meeting:

• Expand Worldex’s permitted business activities into additional organic/inorganic materials and compounds.

• Remove the separate executive-pay rules and restore annual shareholder approval of director remuneration.

• Approve a ₩4bn director-pay ceiling, down from ₩7bn.

The vote passed, so VIP must be at least aligned with the founders now. Which is grand news as far as the governance thesis goes. Muted reaction from the stock.

The August filings called the meeting and solicited proxies; the September filing confirmed both resolutions passed.

The 2027 AGM still remains the biggest major catalyst.

Now an update on our HK stocks:

Read more

It is a mission-critical, SaaS-like business embedded deep inside the day-to-day operations of more than 110,000 merchants in China (the merchants who sell on Taobao, Tmall, JD, Pinduoduo, Douyin, and other apps).

Once installed, it becomes difficult to replace, because a large part of the growth is increasingly coming from existing customers expanding their usage of the platform, rather than simply from constantly acquiring new customers.

The company reported 115% net revenue retention in 2024.

Its higher-value “synergy” products grew revenue by roughly 65% last year, while the total SaaS customer base still expanded by almost 28% to around 113,000 customers.

So you have both -more customers entering the ecosystem and existing customers spending more once they are using the platform.

You can buy the company for roughly 13x my estimate of 2026 adjusted earnings. But even that substantially understates how cheap the underlying business may be. More than half of the company’s market value is represented by cash. Strip out a reasonable portion of that cash, and you may effectively be paying somewhere around 6x forward earnings for the operating business.

But wait! There’s more! The business is also at an inflection point- financially.

Revenue increased 25.6% last year.

and adjusted profit increased by 380%

Gross margins have expanded from approximately 50% in 2021 to almost 74% today.

And the operating leverage is only now beginning to become visible.

The latest guidance suggests adjusted profit to increase again.

That is because this is fundamentally a highly scalable business.

The software has already been built.

Customers largely pay in advance. There is very little physical capital required to support growth. And as more revenue passes through the existing infrastructure, a larger proportion of every incremental renminbi can fall through to profit. This is the point in a SaaS business where the economics can begin to change very quickly.

But wait! There is even MORE!

This company is already the largest player in its niche in China, and still the market itself remains fragmented and relatively underpenetrated.

The company also has no debt, and tons of cash.

And importantly, management has already begun returning some of it! Sharebuybacks- representing just over 3% of the company, during 2026.

The board is also considering an interim dividend.

So the thesis is not dependent purely on multiple expansion, or the market finally noticing the business, or investors finally starting to appreciate that there may be value on the Hong Kong Stock Exchange.

You have a mission-critical software business.

You have recurring, subscription-like revenue.

You have existing customers expanding their spend.

You have new customers still joining the platform at a rapid rate.

You have gross margins approaching 75%.

You have significant operating leverage.

You have a dominant competitive position.

You have billions in cash.

You have almost no financial debt.

And you may be paying only 5-7x forward earnings for the operating business after adjusting for that cash.

There are risks, of course, there always are, but I believe the market has presented us with an opportunity here, and I have certainly acted on that opportunity…

If you want to become a paid subscriber- do so on the Desktop Website- otherwise you pay 30% extra for Apple’s Walled-Garden tax.

Read more

It is a wonderful restaurant Chain in China (my personal favourite), growing rapidly, highly profitable, and still opening stores at a remarkable pace.

And you can buy it for roughly 6x forward earnings.

Revenue grew more than 20% last year, while adjusted profit grew around 40%.

The company has no meaningful financial debt and carries a substantial cash position.

Management is still expanding the store base aggressively, giving the business a long runway if unit economics hold.

Yet the stock trades at a valuation more appropriate for a business in structural decline.

And you are being paid handsomely while you wait.

Management has formally committed to distributing at least 50% of annual earnings to shareholders.

At today’s price, that implies an ordinary dividend yield of roughly 7-8%, before any special dividends.

Here’s the simple math.

If you buy a growing business on a 15-16% earnings yield, receive 7-8% of your purchase price back each year in dividends, and earnings continue compounding at even a fraction of their recent rate, the return potential becomes difficult to ignore.

But if the company can prove that its new stores remain economically attractive, and that mature-store performance is stabilising, I struggle to see why this should remain a 6x earnings business.

It is profitable.

It is growing.

It has a large runway for expansion.

It has lots of cash. No bank borrowings.

It returns capital.

And it may be trading at less than half the multiple investors would ordinarily pay for those characteristics.

My favourite part about this restaurant is that it’s my personal favourite in China. I have been to multiple branches- conducted on the ground due-diligence, and will always return.

The company I am pitching today is…

Read more

Earnings:

Prime wafers.

Clear thesis confirmation.

Revenue increased 19% QoQ, from ¥6.35bn to ¥7.55bn, while operating profit rose 44%, from ¥1.33bn to ¥1.91bn.

Eight-inch monthly production capacity increased from 250,000 to 270,000 wafers, while shipment volumes reached record levels.

Reclaimed wafers.

Demand is healthy but with temp…

Read more

What followed, was akin to the market’s secondary function “weighing”- as many of these companies posted stellar earnings, only to be met with a tanking share price, from what I like to attribute to investors taking profits on companies trading at valuations that no amount of stellar performance could make sense of.

Investing is simple, but not easy. When you see screenshots of X accounts up 3541% YTD, you start to wonder if your strategy might need an update. When you dive a bit deeper to discover those accounts are essentially the lucky few who took a 1 factor bet on a few hype names it reminds you to stick to your strategy- you’re in this for the long run.

For every X account up a thousand percent year to date, there are plenty who wiped out their accounts. You don’t see them posting, because they’re busy nursing real wounds.

The bull-case for AI is not over, far from it, but it does not mean an AI exposed company trading at 30X sales is a good investment. Good business? Sure. A good investment needs to have a valuation that makes sense.

Soitec (156%)

Earnings:

The earnings picture was stellar. The thesis played out beautifully and quickly, and the market re-rated the company to what I believe is a fair value for now.

The market is keenly focused on the photonics story. The company is delivering: “Photonics-SOI sales doubling year-on-year.”

What I struggle to understand is whether or not Soitec will benefit from true bottleneck-like economics: if they are able to increase prices to take advantage of excess demand. It is possible, but I cannot get a good read on it, it’s beyond my pay-grade, and with every single analyst watching Soitec like a hawk now, the edge of it being “undercover” is lost.

How The News Effects the Thesis:

No matter which way I model it, even under my bullish assumptions, it still looks fairly valued to expensive at 122EUR per share which puts FY27 EV/EBITDA of 20X.

That being said- the real ramp up for photonics is still set for 2027, onwards, but as I said before, I cannot get comfortable modelling that.

I do personally believe that the stock will continue performing well into 2027, 2028, and 2029. I would just like a better entry price after locking in a 100%+ gain.

I was greedily eyeing Soitec as it fell from its high of 192 EUR. I wanted to buy in the 70s, but alas the company reported earnings and the stock flew again.

I believe Soitec will continue to do well into the future, I also believe that management will continue to sandbag earnings because photonics will continue to surprise both them, and the market.

My Position:

I hold no position now.

I cannot buy Soitec at current prices because I cannot properly model the TAM to a degree where I feel comfortable with its current valuation. When I originally pitched Soitec it traded at 51 EUR. It now sits at 120 EUR. I sold out most of my Soitec around 110 EUR before it went to 192 EUR.

Argosy Research (-14%)

Earnings:

Revenue down, the earnings didn’t collapse with it, but the growth I have been expecting has not arrived yet.

A brief reminder of the original thesis: I expected LONG-DIMM, Type-C, and Micro-stamping which all accounted for 30% of the revenue to grow much faster than the lower margin legacy SO-DIMM.

Management explicitly guided for shipments of LONG-DIMM to double, with demand coming from Meta and Google. It did not happen- LONG-DIMM actually fell by 1%.

However, in the May investor call they mentioned that this growth is expected to be concentrated in the second half of the year, and this is why I think it is too soon to say whether or not the thesis has played out.

Instead we had Metal Bar growing 13% QoQ - this is also higher margin business, and growing 47% YoY now at 12% of total revenue.

Inside “Others” we have CAMM2- LPDDR5, and CABB connectors for AI PCs and workstations. “Others” is growing extremely fast, and if we get qualification in CAMM2 and CABB might become its own revenue segment line.

M2 remained stable. SO-DIMM, as expected in our thesis, dropped significantly, but net profit remained stable for the company.

I am still waiting for the Q&A transcript to be published online. Especially interested to see what management says regarding the “double shipments” from LONG-DIMM- whether that’s expected to all be in H2.

How The News Affects the Thesis:

Mixed play on the thesis.

1- The revenue mix is changing towards higher margin, higher quality products. Confirms part of the thesis

2- Long-DIMM growth? Possibly weighted towards H2. This will be validated or invalidated either in the Q&A that I am still waiting for, or from next results.

I will update this writeup as well as the Substack chat regarding the Q&A when it arrives.

My position:

Holding the position for now. It was a full position on entry. No changes made. I want to see the Q&A on Long-DIMM. Would be nice if its just weighted towards H2.

Dividend yield at ~6%.

EV/EBITDA at ~8X with lots of growth optionality in its new products keeps me interested.

GnCenergy (+38%)

How many times do I need to learn this lesson?

News Update:

GnCenergy up 42% since my pitch- but it was not one for weak hands (me) because it first dipped 45% before going on a massive tear.

I obviously don’t have a problem finding winners, I have a temperament problem. I should stop looking at my portfolio. I sold because I got scared, took a small loss.

The six recent deals they got total KRW245.36bn. That equals approximately 93.4% of GnCenergy’s entire 2025 revenue of KRW262.63bn.

- KT / SGC AI Infra Gunsan: KRW75.51bn, through December 2027.

- LG Uplus Paju Center Phase 2: KRW38.16bn, through March 2028.

- National AI Computing Center / Samsung C&T: KRW35.96bn, through October 2028.

- Yongin Deokseong AI data centre / Samsung C&T: KRW33.18bn, through March 2028.

- KINX Siheung gas-turbine generators: KRW32.81bn, through March 2029.

- Samsung SDS Gumi AI data centre: KRW29.75bn, through May 2029.

How The News Effects the Thesis:

GnCenergy is experiencing exactly the kind of demand I was expecting from the datacenter buildout in Korea. What is left to be determined regarding my original thesis, is whether or not these projects convert into higher margin revenue vs previous projects.

My Position:

I chickened out of this position for a small loss.

In future I will be putting price targets on all my pitches- these targets will be based on where I believe the company is fairly valued, and should act as decent targets for where positions should be sold. (Not advice, but my personal targets of fair value).

JCU Corporation (+31.1%)

Earnings:

Recent earnings support the base thesis, not the bull case (AI Ramp).

- Revenue: ¥7.49bn, up 6.5%.

- Operating profit: ¥2.74bn, down 2.3%.

- Net profit: ¥2.08bn, up 2.5%.

- Chemicals: Revenue rose 18.2%, driven by via-filling chemicals (+21.5%) and etching chemicals (+40.1%). (YoY growth)

What bothers me about JCU’s results is that we have not seen a further substantial increase in Via Filling, but at the same time we saw their closest peer, C. Uyemura, eak out a QoQ gain in their chemicals segment.

This latest quarter is historically seasonal, so the Q4- Q1 decline is not new to the business, but Taiwan chemical sales fell ~11% QoQ.

We already know that the majority of JCUs via filling revenue comes from high-end smartphone PCBs, hence the high margin.

The AI-Ramp bull-case thesis assumes that datacenter PCBs become a bigger part of revenue. We are not seeing this with current results, while we are seeing it play out with C. Uyemura.

The apparent margin deterioration was substantially caused by Kumamoto depreciation: D&A rose from ¥223m to ¥572m, so the margin detoriation is not structural in my opinion.

How The Earnings Affects the Thesis:

The core of the bull thesis for JCU was that they would experience a ramp-up in high margin revene by selling more CU-BRITE. This would in turn shift their revenue from majority high-end Smartphone PCBs to high-end PCBs used in AI Chips.

We had confirmation of this when I first pitched the company- Via Fill revenue from Taiwan was increasing substantially.

This could mean a few things:

1. Capacity still qualifying (with my original estimates that we see the ramp in 2027).

2. New generation substrates haven’t entered JCU’s numbers yet.

3. JCU is not selling more chemicals to AI-related customers.

4. Competitors are taking the share.

Peers in the chemical space have all been growing QoQ, MKS, Element Solutions, Qnity. This is uncomfortable because where is JCU’s growth?

I am willing to wait until November results because JCU’s overseas subsidaries report to JCU with a revenue lag. This most recent results from Taiwan is what happened from January to March 2026, not from April to June.

C. Uyemura also reported a slight revenue increase- and they too have a similar accounting lag for overseas subsidiaries.

So if we look at the other competitors, we might be comparing the wrong financials.

My Position:

In my last positions update I mentioned closing most of my position in JCU, the reason was two-fold:

1. JCU had become and uncomfortably big position relative to the rest of my portfolio. The stock hit a high of 8000Yen, 100% up from my pitch, it blew past my target.

2. The market has been shooting first and asking questions later- by buying up any name that had an AI-Exposure label attached. The thesis wasn’t validated yet, but the market had decided to re-rate JCU to my fair value. When the market re-rates a company to your fair value, before your thesis is validated, you want to take profit.

My leftover position has now been reduced to ~5% in the sell off. I intend to hold this until the next results. At that point, if JCUs Chemical business (mainly Taiwan) has not grown in-line with other chemical names, the thesis would look invalidated.

The AI-bull case would be invalidated. But the company currently trading at around ~8X EV/EBITDA is not demanding at all. My base case still has JCU trading at around 6,600yen.

Nippon Chemical (-7%)

Earnings:

Earnings were fine, management is guiding for a 50% increase in Profit, but this is largely due to offloading of cross-holdings rather than a material increase in profitability of the operating business.

The company continues to offload cross-holdings. Using the proceeds to increase capacity for High-Purity Red Phosphorus which are scheduled for production in fiscal 2027- which is 2026- this year, they have a weird fiscal year.

Electronic materials results were muddied again with an automotive decline. The company mentioned that the “Level of demand, particularly related to AI servers, is exceeding expectations.” they are also possibly expanding Barium Titanate Capacity.

How The Earnings Affects the Thesis:

As with JCU Corp, and many other pitches, like Worldex, Shin Etsu, we will only really see the AI-bull case thesis play out from 2027 onwards, as the demand ramps up from their customers who are increasing capacity.

This earnings update has been fruitful because I have realized that my portfolio is far too overweight into different companies that reflect one-factor bet- a bet that demand appears later, in 2027 onwards, for mostly cyclicals.

But do I really need so much exposure to different variants of the same thing? From a portfolio standpoint, the answer is no. I should be more diversified.

While these earnings do not validate or invalidate my thesis, I need to re-assess my entire portfolio, and decide just how much exposure I want to these single factors.

My Position:

I am probably going to sell off my Nippon Chemicals position in its entireity. Not because the stock is expensive, or because I am unhappy with the results, but because I am too concentrated in chemicals, and need to decide what to keep and how much to keep, whether its JCU, Shin Etsu, or other cyclicals.

I do still want exposure to the capacity expansion bet, but it cannot make up such a large portion of my portfolio.

I will continue following Nippon, because I want to catch any signs of when the capacity thesis takes root, so I will keep you updated.

On this note- I will also be making the spreadsheet so that you can follow my changes live. With price targets, and cost-basis.

Update on:

Shin Etsu (+12.6%)

RS Technologies (+84.5%)

Worldex (+4%)

To come out in Earnings Update Part 2.

This lovely screenshot came from my chat with , the guy who found Sivers before anyone else I know found it. Appreciate discussing ideas with you, Moody!

If you need to understand what Worldex does I’ve written about it here.

For those subscribers who have been following the Worldex story, we just got our thesis confirmed. The controlling shareholder revised the value-up plan to substantially satisfy activist demands.

I am already too overweight, at 30%, but if I wasn’t I would add here. The KOSPI is getting murdered, and Worldex is holding up even though the market is trying to drag everything down.

Price discovery will happen next as the market starts to understand the governance story. The company will do its first buyback in 18 years, with a shareholder-return policy. There are some caveats, which I will discuss later.

This was more bullish than the most bullish path I foresaw. The market reacted, but somewhat muted still because KOSPI is liquidating all the degen leveraged retail traders.

Massive opportunity in Worldex at the current price.

For a quick recap, the situation has played out as follows:

June 18th, I outline how deeply undervalued Worldex is. Governance was a major problem BUT activists are involved and I expected something to happen before the March 2027 AGM. The direction was positive.

(One piece of pre-history matters here. At the March 26 AGM, shareholders had already rejected Worldex’s director-remuneration proposal, with 69.2% opposed or abstaining. Under the Supreme Court’s Namyang precedent, Bae could not vote his own shares on a proposal governing his compensation. The defeat left Worldex without an approved 2026 remuneration framework for its registered directors, which is why the company called the June EGM and tried to pass the proposals again.)

June 29th, controlling shareholder Bae pulls out all the dirty tricks, holds an in-person vote in Gumi and tries to increase exec pay to egregious levels. Disables e-voting to make it harder for outside shareholders to participate. VIP Asset Management, the activists, rally the troops and reject ALL THREE of the controlling shareholder’s proposals. On the two items covering Bae’s own pay, where his votes were legally restricted, the rejection ran ~94% against. On item 2-1 — the sons’ pay limit, the one item where Bae could vote his full 34.8% bloc — the item still failed, 51.7% against, because 94.7% of general shareholders (everyone except Bae) voted no.

July 13th, I expected a "gift" from Bae, the CEO, to his son before the value-up plan. But Bae capitulates to the activists' economic demands, for reasons I will explain below. This is more bullish than I expected, but there are caveats!

And the capitulation came with something interesting: a written apology from the founder himself. The company's press release said the revision reflects management's "solemn recognition of the fatigue and disappointment shareholders felt" from delayed shareholder returns. Bae personally called the EGM result "a stern warning letter sent by shareholders and the market, which management accepts heavily and actively" — and added: "I sincerely apologize for failing to fully console shareholders' deep hurt and disappointment."

I think the market needs to re-rate this, probably around 30-50% just on the current value-up plan.

More could happen prior to the 2027 AGM. This is overall bullish, but the fun ain’t over yet…

Read more

The market is pricing this company as if there isn’t a memory capacity expansion and as if the cash on the books will never be seen by shareholders. For good reason too — governance, which I believe is changing.

The asymmetry involved is striking.

The company is already valued as if cash will be trapped forever. The pessimism is fully baked in, and even in my bear case, the company is worth more than it is here. There is a reason for this too: the founder needs to keep the price low for tax reasons, but only for now.

There is a clear signal that will indicate to the market when that time is up.

Valuation snapshot (at the ₩28,700 close, July 3, 2026):

• Market cap: 16.51m shares × ₩28,700 = ~₩474bn

• Less net cash: ~₩220bn (₩230bn cash & financial instruments, minimal debt)

• Enterprise value: ~₩254bn

• EV/EBIT ≈ 4.3x trailing (FY2025 EBIT ₩58.6bn)

• EV/EBITDA ≈ 3.3x trailing (EBITDA ~₩78bn)

• On my FY2026 estimates that falls to ~3.3x EBIT / ~2.6x EBITDA, and on FY2027 to ~2.7x / ~2.3x — the market is paying less for this business every year the capacity build compounds.

While competitors trade at EV/EBIT of 10x+.

If you want to catch up on the business, I analysed it in the post below. This article will focus solely on the governance. And our pathway to alpha.

This situation is COMPLICATED, but it is damn well worth understanding. I think there is money to be made here.

On the 29th of June, shareholders rejected ALL of Bae’s (CEO, Founder, Controller) ordinary resolutions.

These 3 resolutions were all trying to vote on similar things — exec pay. Egregious amounts of it.

The 2-1 ordinary resolution is important to understand (red boxes above). It was a vote on execs’ compensation excluding Bae’s pay — and this means that Bae himself and his full 35% ownership could vote — which is why the vote was so close at 51.7%. This is crucial because even when you include Bae’s full 35% vote, the other shareholders came together and politely told management to go and “F*%$ themselves”, and rejected the vote.

This was his last desperate attempt in my opinion.

In ordinary resolutions 1 and 2-2, Bae’s full 35% was excluded because of a rule where controllers cannot vote on their own pay. Which is why we see it lost by such a large margin.

This is insanely bullish because prior to the meeting, Bae pulled out all the stops for this meeting to try and raise his pay:

• He disabled e-voting.

• He forced a stamp process to introduce friction for proxies.

• He forced an in-person meeting in Gumi.

And he got voted down.

The outcome of this vote is insanely bullish for Worldex minority shareholders. (not because they won anything, but because it means they can win).

It is far more bullish than I expected it would be.

I did not expect the minority shareholders, against the disabled e-voting, against the frictions, to be able to vote Bae down when his full 35% share vote counted.

The share price did a nice 10% pop on the news, but the next day it didn’t go anywhere. I think the market is completely asleep on this name — while Hana Materials trades at a 3x higher valuation on a P/E basis. Especially when I regard Worldex as the better financial operator vs Hana.

I really do believe this is only the beginning for Worldex shareholders.

Worldex is exposed to the biggest HBM capacity buildout in history — which will be a tailwind for more than 5 years (most likely). Memory prices aren’t what matter to them — it’s the capacity buildout.

At a price to earnings of ~10, with nearly half the market cap sitting in cash, and an activist that just beat the founder’s full 35% stake in an open vote — this changes the game completely.

But here’s the part almost nobody understands, and it’s the exact reason the stock is still this cheap:

The man with the most information in this situation is deliberately keeping the share price down. Legally. Methodically. And on a deadline.

Worldex is priced the way it’s always been priced — a company trapped inside a family. Succession, the Korean edition. Except Logan Roy is 75, cornered, and running out of calendar. The suppression that created this price has an expiry date — and it’s written in statute.

Better still: there is a single public filing that starts the clock. It will look like nothing — a routine disclosure most investors have never read and wouldn’t recognize if it printed in front of them. But the day it drops, everything after it is already choreographed: a quiet window I can time almost to the week, then the concession that unlocks ₩230bn of trapped cash, then a March 2027 meeting where the founder must win four separate votes from a register that has already beaten him at full strength.

The market will reprice when the results hit the headlines. The money is made by knowing what to watch before they do.

Behind the paywall:

• The signal — the exact filing, who files it, and why it means the rest of the script is already running. My scraper watches DART for it every day; the moment it prints, subscribers hear it from me first — chat and email.

• The tax mechanic that dictates the entire sequence — once you see it, the founder’s every move for the past 18 months snaps into focus.

• The statutory deadline — why his whole strategy dies on one specific date in March 2027, no matter what he wants.

• My base, bear and bull targets — including why my bear case is roughly the price on your screen right now.

Subscribe now

Read more

This article is for informational and educational purposes only and does not constitute investment, legal, tax, or financial advice, or a recommendation to buy or sell any security. The author may hold a position in Worldex (KOSDAQ: 101160) and could transact at any time without notice. The analysis reflects personal opinion and includes forward-looking estimates and assumptions that may prove wrong; figures are drawn from sources believed reliable but not independently guaranteed, and some (including legal interpretations and probability weights) are the author's own judgment. Korean securities, tax, and corporate law are complex and fact-specific. Do your own research and consult a licensed professional before investing. You alone are responsible for your investment decisions.

Disclosure: I own this stock.

Also* The stock has been moving while I have been writing this up- so some prices/valuations are stale- but still not close to my bull case- the “catalyst” has still not activated.

I think I have found the Korean version of Succession, except instead of a media empire, the family drama is sitting on top of a deeply undervalued Korean semiconductor consumables company exposed to one of the biggest memory capacity expansions in history.

This company, on a comparable basis, should trade at around 3X its current valuation. Easily.

But first we need to understand the fun stuff.

The company is Worldex and the CEO and founder is Bae Jong-sik- he’s our Korean version of Logan Roy. A Tycoon and a real turd to be honest. (Sorry Bae- but you are).

He was born in 1951, so he is 75, and he has still not properly implemented succession to either of his sons. The two sons currently own 0% of the shares, which is the entire game. The founder needs to gift his shares to his son, and he wants to gift them at the lowest possible price. A low price means a lower gift-tax bill, which could save the family billions of won. But time is running out.

After the succession happens, the incentives flip discretely to align with minority shareholders- and this is, along with a recent passive shareholder now flipped activist, is why the governance situation is so interesting, and so ugly, and so ripe with opportunity once you understand the outcomes.

(I will go into depth into these governance events, and how I am going to play it later in the article).

The tactics used by Bae Jong-sik are dirty, but he is fighting what I believe to be a losing battle- against activists, and against the new Korean pro-minority shareholder policies.

The company previously tried to push through a golden-parachute style clause that could have allowed special severance payments of up to 20X the three-year-average salary, on top of the ordinary retirement allowance, if directors were dismissed before the end of their term due to hostile M&A. That clause was removed after shareholder opposition, including opposition from VIP Asset Management- which is our Knight in shining armor.

In Succession terms, VIP Asset Management would represent the largest activist shareholders who see massive value in the company, but need to fight Logan Roy to get it. And last week, on the 9th of June, they went even more activist- VIP filed to change its holding purpose from “simple investment” to “general investment”, increased/confirmed its stake at 15.64%, and formally pushed for buybacks, cancellation, higher shareholder returns, and TSR-linked executive pay.

This followed the 26 March 2026 AGM in Gumi, where VIP’s CEO reportedly attended in person, opposed Worldex’s director-compensation proposal, and, with the controller’s votes restricted, helped minority shareholders defeat it with 69.2% voting against.

There is a caveat: Important voting wrinkle: the family’s ~35% stake did not have normal voting power here. Because this was a director-compensation proposal, the founder/controller was treated as a specially interested party and his votes were restricted. So, for once, the minorities actually mattered — and they voted it down.

This is the new South Korea, baby. The age of tycoons is over. How are you going to make money with this?

Let’s first understand the juicy undervalued gem that is Worldex. Then I will get into how this situation likely plays out.

Worldex

The company is called Worldex and they make Silicon parts mostly (73% revenue), quartz parts, and fine ceramics.

Their closest competitor is Hana Materials and trades at around 3X the valuation- except that Worldex has better financial operating metrics- Worldex will benefit from the demand draw created by SK Hynix and Samsung Electronics the memory makers expand capacity. The most recent financials for Worldex showed that 45% of their revenue came from SK Hynix and Samsung Electronics.

This is the valuation sorted by closest peer. The valuation is rather striking.

Silicon Parts are used in semiconductor process tools (machines) that make wafers. Plasma or etch chambers that could be running DRAM etch, HBM, 3D NAND etch, logic or foundry etch, other processes.

Worldex is an “after-market” supplier of silicon parts, quartz parts, and fine ceramics and other, whereas Hana Materials is a “before-market” supplier.

What does this mean? It means that when the memory makers like SK Hynix, Samsung Electronics, expand capacity, players like Lam Research, Applied materials, (and a few others) who make the machines you see below order directly from Hana Materials for the silicon parts- the silicon focus rings, silicon electrodes, etc, that go into the machines. They order from Hana first.

This is why Hana Materials is called a ‘before-market’ supplier. Their Silicon products are used to fit the machines first.

Worldex is mostly used after the products depreciate, and need replacement.

Worldex stated that 45% of Q3 revenue came from Samsung Electronics and SK Hynix. Worldex was registered as an SK Hynix supplier in 2001 and Samsung semiconductor supplier in 2002, they have been qualified a long time.

Don’t get me wrong now, there should exist some kind of a discount, which I will explain later, but in the face of one of the biggest memory capacity expansions in history, Worldex trading at a P/E ~10 and EV/EBITDA of 3.8, is far too cheap.

The reason for the massive discount, is governance, and which we delve into deeply later. The story is changing very quickly.

Worldex at a EV/EBITDA of 3.8 has a massive margin of safety due to its current valuation and a surprisingly clear path to future revenue growth- SK Hynix said to double memory capacity in 5 years and I wouldn’t be surprised if we hear the other memory makers following suite.

This tweet by Jukan is also favorable in the general direction for Worldex, not because it means Worldex will absolutely get price increases, but it points to the actual demand benefiting the entire chain.

Where things get even weirder- valuation-wise.

Hana Materials is considered Worldex’s most direct peer or competitor, BUT

Worldex does more revenue, more operating profit, and more net profit

YET Hana trades at a market valuation of 1.34T KRW.

Worldex trades at a market valuation of 424.33B KRW.

Hana Materials trades at a P/E of 27.

Worldex trades at a P/E of ~9.

Today the company released a “value plan” which was bearish, in my opinion, but the bottom seems to be in for this company, so the market took it as bullish, and the stock rose +10%

The reason for this massive discount, is governance. But that’s about to change, you just need to understand how to play it.

But there’s more- if the price rises before a “gift” is done- the bull case may only materialize much much later. It’s complicated.

The point of this article is not to convince anyone to rush and buy the stock, I believe there is opportunity here, but within bounds, and the situation needs to be understood completely to extract that opportunity.

Read more

This research is not complete, I will be editing this document as I discover more, but wanted to get this out because the market is onto these two.

I think Nippon Chemical is a superior play. I just started analyzing Sakai first so that’s why it reads in this order.

Sakai Chemical (4078.T)

2027 numbers are based on management’s estimates which I believe might be far too conservative given the growth in AI server MLCC.

The TL;DR is that:

1- 30% of Sakai Chemical Operating income, (Electronic Materials segment) is exposed to high-end MLCC most of which through its hydrothermal Barium titanate. (Not all of this 30% is AI-Server!)

2- The AI server MLCC market is expected to grow at 80%+ CAGR (, and the general server MLCC market will also accelerate due to agentic AI increasing CPU demand (around 30%-40% CAGR).

3- The company is cheap ahead of the biggest expansion in MLCC we have ever seen.

From the Zephyr tweet below:

“The AI server MLCC market is growing at 80%+ CAGR, and the general server MLCC market will also accelerate due to agentic AI increasing CPU demand (around 30%-40% CAGR)

MLCCs for servers were a $1.3B market in 2025 ($600m for AI servers, $700m for general servers.

We will see negative growth in the smartphone/mobile MLCC market for at least 2026-27.

Humanoids are another future high-growth market for MLCCs
Book-to-bill ratio for most MLCC suppliers is over 1 now

“MLCCs for servers were a $1.3B market in 2025 ($600m for AI servers, $700m for general servers.” -Zephyr

What is important to note here is that the MLCC market is around $15Bn. Sakai and Nippon both sell into automotive, and electronic consumables- it’s not all AI-Server. But demand is very likely to increase across the board.

Sakai uses hydrothermal and Nippon uses Oxalate- these processes both produce BaTiO3 of extremely high quality- which is most likely the ones used in AI-Servers.

These are the materials that Sakai Chemical makes that goes into making an MLCC (multi-layer ceramic conductor)

Or a clearer diagram for you:

The dielectric layers:

1. Barium titanate. BaTiO₃. The flagship. This is the core MLCC product.

2. Other perovskite-type dielectric compounds.

Dielectric raw materials & composition tuning:

3. High-purity barium carbonate. BaCO₃.

4. Barium and strontium salts.

5. High-purity titanium oxide

Electrode-related:

6. Electrode materials.

It’s probably 99% MLCC at this point.

I am very certain they are sand-bagging their 2027 estimate for electronic materials revenue because the demand for the AI server MLCC market is growing too fast.

How exposed is the company to MLCC? Operating profit for 2027- according to management estimated, they are about 40% exposed to mostly high-end MLCC.

Why do I say high-end? Because there are three ways to make BaTiO3.

1- Hydrothermal. Sakai’s focus. Grows BaTiO₃ particles in a hot, pressurised liquid reaction. Produces very fine, highly uniform particles, ideal for thinner, higher-end MLCC layers.

2- Oxalate. Uses a chemical precursor that is precipitated and then heat-treated into BaTiO₃. Very uniform composition and can compete in high-performance MLCC applications.

3- Solid-state. Mixes solid powders such as BaCO₃ and TiO₂, then fires them to form BaTiO₃. Cheaper and good for scale, but particles are generally less fine and uniform.

Hydrothermal is 13.5% of the market.

Oxalate is 21.4% of the market share.

Solid state/ or phase is 61.8%.

Given that “the AI server MLCC market is growing at 80%+ CAGR, and the general server MLCC market will also accelerate due to agentic AI increasing CPU demand (around 30%-40% CAGR)”-Zephyr

The demand will most likely be skewed towards Hydrothermal and Oxalate because AI servers increase demand for high-capacitance, high-reliability MLCCs.

That should benefit multiple premium BaTiO₃ routes, like Sakai.

Nippon Chemical (4092.T)

Nippon Chemical is an even cheaper play to Sakai.

And it also likely has even more exposure to AI-server MLCC material through its own Electronic Materials segment. Which is at 25% of total revenue.

But operating profit for this segment dropped even amidst a 23.4% rise in sales for Electronic Ceramic Materials (Mostly MLCC). Caused by problems: mainly battery materials, temporary prior-year inventory benefits reversing, and restructuring, consolidation costs.

If we assume that Nippon’s Electronic Ceramic Materials segment can do the same 20% operating margin as Sakai’s, then at a 12,000(mYen) 2027 segment revenue (my own estimate) we get a contribution of a 2400 to operating profit.

This contains a lot of assumptions though, because it means the problems they had in battery materials, temporary inventory, restructuring etc, get cleaned up.

But that would push them from a EV/EBITDA 8x to EV/EBITDA of 4x in 2027. (unless share price re-rates and all else equal which is a big ask).

This 2400 is significant. The companies operating profit in its entirety was 2400 (mYen) for 2026.

They could very quickly re-rate 100%, or become far too cheap.

Nippon Chemical also has capacity to produce high-purity red phosphorous, used for optics, but the amount is not quantifiable from the statements.

I personally, have made some major mistakes. Not just this year, but I have become very aware that if I just held the stocks I pitched longer (since I started writing) I would have done drastically better. If I held one of the first stocks I ever pitched- $HRTG, I probably would have been retired 2 years after. (But I would never stop looking at stocks).

I have always sold too early, and I think this comes from an inherent risk-aversion. For the majority of my investing career I also held 50% of my portfolio in cash, waiting for a crash that never came.

Initially my rationale for holding so much cash was a good one, I was painfully aware of the Dunning-Kruger effect.

I also knew that getting to the point where you have looked at enough stocks to pick the good ones takes years of practice, extending the ‘peak of Mt. Stupid’ further out.

Investing is not like a tennis game where you hit 1300 balls a game and your neurons adapt after each hit.

It takes much longer because you need to wait months and possibly years to verify your initial thesis as new information is slow to come out. The price of a company shooting up or down short-term, is not a verification of a thesis, it is the market voting on the outcome.

I have been looking at stocks daily for almost 13 years now. And I do feel I am out of the valley of despair. Finally. That doesn’t mean the lessons are over, though.

Lesson I need to apply: I need to hold my winners.

My new strategy will be to sell 25-50% and sometimes 75% of my winners after massive runs (70-100%+). But to always hold some and let the market do its thing.

Subscribe now

JCU Corp (+70%)

We will start with the pitch that began it all- JCU Corp.

JCU Corp was never meant to be a 10 bagger. I saw it as an extremely positive reward to risk play and I mentioned that in the pitch. The market has re-rated the company since then. I positioned myself extremely heavily in JCU because I thought the downside was minimal, and the upside had 50-100% in it. I have taken off most of the massive position since but now still hold a full position in the stock.

Recent earnings show increasing profitability, and most importantly- the increased memory prices have had a negligible effect on their via filling (80% of their operating profit) growth. I had assumed the softness in the smartphone market would lead to a decline in their biggest segment. It had minimal to no effect. Chemicals for high-end PCBs (mostly smartphones) have surprisingly been on the rise.

What about AI-Datacenter exposure? Our proxy for this is Taiwan, Japan, and Korea. Taiwan more specifically. I think it is lagging and we will eventually see out-sized growth from Taiwan because it takes a long time to qualify the chemicals in the packaging facilities. They have yet to come online.

Ibiden, Unimicron, are expanding capacity. They make the package substrates that are then filled by chemicals like CU-BRITE. They have to build the facilities and then they have to qualify the chemicals in the facilities. It all times a lot of time, and the bulk of the facilities will come online later. When these facilities are qualified, that is when I believe we may see a big ramp-up in the chemicals used to via fill PCBs.

Then there is the TIPHARES, JCU’s line of surface-treatment chemical solutions and the Kumamoto facility near the TSMC Gigafab in Japan.

This is very important to the company, its the first thing they mention in their update.

JCU presents us with a lot of optionality too, as most of their current capex is to increase their exposure to the highest end semiconductor package substrates- as mentioned in the image above.

This all makes JCU look like a long-term hold. They are quietly positioned to do very well as capacity increases.

Current EV/EBITDA at 11.46x is a fair to bargain multiple to pay for a business that will probably see outsized growth into the future. I still don’t see much torque for the stock, the business is not like that, but I do see a steady compounder.

Position update: sold most of my super overweight position, now holding a full position of 7%.

RS Technologies (+80%)

When I pitched RS Technologies it looked like quite an obvious and safe long. The company has surprised me. It re-rated quickly as the market realized it shouldn’t be lumped in at the same multiples as other prime-wafer businesses.

Reclaimed wafers are a different game, and the company does in fact have a qualification-like-moat at competitive price points.

My favourite thing about RS Technologies was the reward to risk profile. It was the kind of company I love to pitch because at the valuation (at the time) it had minimal downside (and 50% cash as a % of market cap). It’s hard to lose on companies like this.

For Q1, the overall outlook is positive.

The company saw an increase in revenue and margin in its reclaimed wafer business YoY and a 31% YoY increase in its prime wafer segment. It puts the company at a P/E of 18. Still relatively cheap due to the amount of cash it has on the books, but it does not have the same appeal now as when I pitched it initially.

Position update: I have sold 75% of my position.

Shin Etsu (+23%)

Shin Etsu’s electronic’s segment, now its biggest segment and highest operating margin segment, is exposed to almost everything chip-related. They make products for:

• Front end wafer fab: Silicon wafers, SOI wafers, photoresists, hardmasks, photomask blanks, synthetic quartz mask substrates, pellicles

• Back-end packaging, thermal protection: Encapsulants, liquid encapsulants, die attach, thermal interface materials, silicone coatings, low-dielectric materials

• Adjacent electronics, infrastructure: Rare-earth magnets, HDD VCM magnets, optical-fiber preforms/coatings, LT wafers, PBN, compound semiconductors, QST/GaN.

Shin Etsu is a bet on longer term AI and datacenter demand.

The company is also a photonics play. They have the largest SOI capacity worldwide (according to Citrini) and we are likely to see them start mentioning this market soon as the photonics buildout ramps up. They license Soitec’s smartcut technology and we have seen Soitec run up 245%! I believe Shin Etsu’s time is yet to come.

This from Citrini’s “Let there be light” article.

I think Shin Etsu will experience a similar dynamic to JCU Corp- where most of the growth lies ahead of it.

We are also watching the company slowly shift to a higher margin business as its “Electronic Materials” segment sees gains from the datacenter buildout and its infrastructure drags.

Position update: no change, holding a full position.

Soitec (+245%)

I found Soitec from reading when he will still criminally underfollowed. I have no qualms pitching stocks from others when I believe in them— as long as I give credit where it is due!

I did this for Wasion Holdings, ChongQing Electric Machinery, and Thakral Holdings from . Wasion Holdings and ChongQing both did 80%+ and I have since sold both, although ChongQing might warrant another look.

Thakral Holdings has an extreme margin of safety and is only up 16% since the pitch. Deep value.

With the amount of Substacks and X accounts pitching stocks soon every stock on the market will have a pitch, and it will be our job again, to filter noise.

I have always wanted to build my own Substack around one goal- to make you money with limited downside while not posting noise. I will continue to pitch other people’s theses if I think it will make you money.

Thank you Jason for finding Soitec. I hope my subscribers made some money on it and didn’t sell early like me!

Soitec is currently up 245% since I pitched it. I learned a very expensive mistake on Soitec. I started weighting it more and more and ultimately paper-hands half of my position on a dip, and then proceeded to watch the stock double from there- then sold my entire position and watched it double again!!!!

(This is a major part of the reason I tend not to announce trades I do personally unless I get 100% out of a position or the thesis changes- because my analysis shows that just buying and holding the stocks I pitch perform better than I do myself!!! Lesson learned.)

I will admit that Soitec now has a much steeper valuation than when I first pitched it.

But things have fundamentally changed too- the photonics demand TAM expectations have increased drastically.

It is for this reason I wish I held. The optics buildout is truly gargantuan. Beating all previous expectations.

The Goldman-style thesis says AI networking shifts from “some pluggable optics at the network edge” toward massive numbers of optical links inside scale-out and scale-up AI fabrics. That creates demand for silicon photonics PICs, and those PICs are commonly built on Photonics-SOI wafers. Soitec is one of the key suppliers of those engineered substrates. They also benefit through the licenses they license out- to Shin Etsu, for instance.

Position Update: Sold out of all, but I wish I didn’t.

Argosy Research (+20%)

Argosy gives me a similar feeling to RS Technologies in its setup. It’s largest segment has bottomed out (SO-DIMM) and Server (R-DIMM) is growing at 30% annually.

I think the market hasn’t woken up to the fact that R-DIMM benefits from the CPU bottleneck- adjacently.

Agentic AI makes CPUs more important. More CPUs and CPU servers mean more DDR5 server memory. In conventional x86, that memory is usually RDIMM / MRDIMM.

It benefits because agentic AI likely increases:

1. CPU server count

2. CPU:GPU ratio

3. memory per CPU socket

4. DDR5 speed requirements

5. high-capacity RDIMM, MRDIMM adoption

6. RCD/PMIC/SPD hub content per module

As we see M2 become a bigger portion of Argosy’s revenue (it’s also higher margin with a bigger moat vs SO-DIMM and their M.2) we will see the company re-rate and probably do what RS Technologies did- a double.

GnCenergy (-20%)

Would’ve been nice to buy this dip but immediately after posting Korea, and a lot of Japan entered into a brief correction.

They also released earnings which I reported on here:

Think this one needs a bit more time to play out as their backlog comes into their income statement at much higher margins (my thesis).

Position Update: still holding a full position

Please leave me feedback on what markets you would prefer I focus on, and what kind of value you get out of this Substack.

It’s a small-cap.

Holds 70%+ market share in its niche.

It’s vastly undervalued at EV/EBIT 10 on a single name basis and vs peers which trade at 20-30X EV/EBIT. The name could double and still be valued at less than peers.

It’s customers are the big, expanding, capex heavy names in South Korea.

I believe it’s undervalued because it’s small and the market has yet to “discover” it properly as it has been doing daily with other names in the space.

I believe the business has 100%+ upside given its strong historic growth, and strong position in one of the biggest bottlenecks we’ve ever seen.

The name of the company is…

Read more

The market is treating it as if future growth is unclear, but its not because the business is sitting in a tailwind created by the expansion of capacity of RAM memory by SK Hynix, Samsung, and Micron.

-Second-order beneficiary of the RAM shortage.

-Growing topline revenue at double digits.

-Two segments growing at 30%+ annually with customers like Meta, and now recently Google. A subsegment expected to double shipments this year.

-6% dividend yield.

-It trades at low multiples. PE 12~ (Below its historical average).

-Its biggest revenue segment is currently supressed, partly due to a slowdown in the sale of laptops- because of the rise of memory prices, not an actual demand destruction of laptops.

-The business itself is also evolving to one with more of a qualification moat.

As the other segments experience 30% and double digit growth you get a more diversified company- and when their biggest segment bounces back eventually you will most likely have growth from every product.

This is not a moonshot. And its Moat is not impenetrable, but it’s an undervalued company based on three factors:

1- Future Growth in memory will clearly benefit them.

2- Growth is deviating towards higher-margin segments.

3- Valuation is below historical mean, and this doesn’t take into account point 1 and 2.

Read more

Here we try to understand:

1. Sive Semiconductors

2. Applied Optoelectronics

3. Soitec- I am long

We can try our best to wrap our heads around the valuations.

Sivers Semiconductors AB

The most hyped of the bunch.

They are basically fabless. They make lasers.

These lasers are individual parts that are used to make up a photonics ‘module’. The lasers are just one part of the module. And everyone wants to use Siver’s lasers in their own product.

Their photonics lasers:

• InP DFB lasers

• laser-array specialist for SiPh, optical I/O, CPO,

• lasers for pluggables

Closest peers are Coherent and Lumentum who offer a much broaded range of optics. Coherent and Lumetum make the lasers AND the modules.

Sivers photonics are very focused on the three mentioned above.

Sivers is similar to NVIDIA in that they hold the IP and get TSMC to make the chips.

Lets briefly try and undestand what all these partnerships mean for Siver’s Semiconductors.

-Ayar Labs partnership to use Siver’s lasers in their own products.

-Win Semiconductors, basically the TSMC for Siver’s, add real capacity to scale of production (of the lasers).

-O-Net is there to use Sive lasers and make modules (External Laser Small Form Factor (ELSFP) optical modules.)

-Poet Technologies also using Siver’s lasers to make a full module product (eg Pluggable transceivers)

-Enablence also using the lasers to make a full product.

How you should think of all these partnerships is that Siver’s lasers work well and now they are partnering with companies so they can make a full product using their lasers.

They are essentially preparing themselves for the launch of light. This is what the market is pricing in.

They are busy sampling their product, qualifying their product for manufacturing. The revenue has not arrived yet, but it will.

The quantity? That’s hard to tell.

From Marvell

“CW is like a lightbulb…It just shines a constant light. It is easier to make and is available from multiple sources and it’s inexpensive,” Nguyen said. “All of the high speed magic for modulating data happens in the silicon photonics.”

Valuation

You cannot value this company looking in the rear view mirror.

The best thing we can do is provide a wide band of what it should obviously not be worth- too expensive, or too cheap.

Price to sales and EV/EBITDA do not tell you anything now.

ALL the company’s real revenue will start next year, in 2027.

The market is trying to figure out what quantity that could be.

So what could it be?

Currently market cap sits at ~$400m (USD) EDIT*- now changed to $270m within like a day of me writing. yikes.

So how do you value it?

You need to try your best to quantify the 2027-2030 and beyond revenue opportunity. Siver’s gives us their DFB array TAM below.

2027E - $564M to $1.1Bn

2028E - $1.2B - $2.5Bn

BUT then they tampered these expectations down- $2bn in 2028?

If they don’t know their TAM, how are we supposed to?

Below is from their capital markets day presentations:

But its the oldest, November 19, 2023. We have an anchor of $10bn for Sivers as whole (not just lasers).

Lets try make some sense of this

The market currently is trying to price in their photonics opportunity.

If we use their LATEST Tam projection of $2bn + in 2028. And they get, like, 10% of that?

$200m revenue.

Latest results- Q4 2025, they did $8.4m. Annualized for 2026, lets say ~40M revenue. This is before they ramp up right?

That’s not bad, then you have a company trading at 2028 P/S of 1.

It’s okay.

The company right now, with all this information is not worth $1Bn today.

$500m is expensive and pure speculation.

$200m is ‘meh’ given the information is so sparse, the TAM got LOWER according to management.

$100m is where I would start looking to get in, to me personally, as an investment, or at least thinking hard about it.

Why the wide wide range? Because the information is so unclear.

It is worth noting that they have NOT started ramping yet, so we cannot look at current revenue. We are looking at their TAM and what they think they can acheive.

AAOI

Similar to SIVE, AAOI depends on some insane projections. But looks better.

Zephyr is saying that in the most aggressive optical transceiver rollout, the market is expected to do 100m - 120m units by 2030. Remember, Zephyr and Citrini were both bearish on AAOI in their recent photonics post here on Substack.

This bearish position is in stark contrast to the bullish projections AAOI management has given.

They say that they can do $378m by Q2 2027.

That’s a revenue of $1.5bn on a market cap (currently) of ~$7bn.

Let’s go with what they say, and assume they can do 40% in 2028, with a growing revenue.

By 2028 they could be doing a $2bn revenue and 40% margins. close to $1bn EBITDA.

Add the cash they made in 2027 from their bullish projections = ~500m

If market cap is constant, cash increases then by the beginning of 2028 you have a company trading at EV = $6.5Bn

Forward EV/EBITDA = ~6.5x

Not bad if you must believe the management.

These numbers are not super accurate, and it doesn’t matter in this enviroment because these numbers are going to change drastically.

It gives us a general feel. It feels a lot more ‘investment’ vs SIVE.

Soitec

Soitec is probably the most undervalued even at current metrics when you incorporate its:

1- Moat

2- TAM

3- Current and future multiples.

It’s probably the most undervalued photonics play.

With the best risk to reward profile.

Trading at EV/EBITDA ~7x is insanity for a company like Soitec. I am long. It is my biggest and only position in the photonics space for now.

Soitec

Soitec caught my eye after reading Jason’s article.

This strapping young lad (Jason) is doing some great research, I suggest you read his article before you continue with mine.

Jason's Chips

The Soitec Series | Part (1/7): Introduction to the French Photonics Substrate Monopoly

Opinions are my own and do not represent past, present, and/or future employers. All content is based on public information and independent research. This newsletter is not financial advice, and readers should always do their own research before investing in any security. I am invested in the semiconductor industry. As of the date of this publication, I…

Read more

7 months ago · 47 likes · 8 comments · Jason's Chips

As a generalist, I want to understand Soitec’s operating leverage, bottleneck potential, moat, and valuation.

I do not have a specialist know-how, but I have relied on my fundamental knowledge of companies to make money.

Before we can even get to valuation, we need to understand a few things, quite briefly:

1- What Soitec actually makes?

2- Is there a moat, is it hard to make?

3- Is there operating leverage? If there is enough demand, can they pump them out and capatilize on that with economies of scale?

4- Roughly, what will the demand for SOI wafers be in future

Once we know this we can get to

5. Valuation- and how that relates to how fast the company will grow.

Fortuantely, Jason already answers most of our questions in his article.

I finished reading his article and thought- “I need to understand the nitty gritties of the valuation, how many of these SOI wafers the market will demand (roughly) and whether that leads to a blast off in operating leverage”.

I look forward to reading his parts 2-7, but I’m too impatient and I want to undrestand this business more thoroughly, right now.

Jason's Chips

The Soitec Series | Part (1/7): Introduction to the French Photonics Substrate Monopoly

Opinions are my own and do not represent past, present, and/or future employers. All content is based on public information and independent research. This newsletter is not financial advice, and readers should always do their own research before investing in any security. I am invested in the semiconductor industry. As of the date of this publication, I…

Read more

7 months ago · 47 likes · 8 comments · Jason's Chips

I’m going to recap a bit, because as a generalist I still needed to wrap my head around what this company actually does.

1- What Soitec actually makes?

We’ve got a Soitec SOI wafer on the left, and normal wafer on the right.

Companies we’ve spoken about- like Shin Etsu, and RS Technologies, (and there are tons of others), make wafers. RS Technologies makes the wafers on the right. Shin Etsu makes both. Soitec makes the ones on the left.

The big risk to our ‘normal’ silicon wafer companies is that everyone now knows how to make them. China is increasing capacity a lot.

When it comes to SOI wafers, Soitec has a special process- its Smart Cut Technology. It actually licenses this out to Shin-Etsu Handotai (owned by Shin Etsu). So great! Shin Etsu also makes photonics wafers! We are positioned well! (But probably a small % of revenue…for now…).

Back to Soitec- they ALSO manufacture these SOI wafers themselves.

2- Is there a moat, is it hard to make?

Jason has described the moat quite nicely.

Smart Cut™ - This is the foundational enabling technology for engineered SOI substrates and Soitec says it underpins nearly all SOI wafers sold globally.

The moat exists.

3- Is there operating leverage?

There is. Via its own manufacturing capacity for SOI wafers and also via its royalty stream.

And what is even better is that the licenses are royalty based- so Soitec will benefit from any increase in demand for SOI wafers. This means if there is a ramp up (which is coming next year and beyond), Soitec will get close to pure profits from Shin Etsu, and other companies that hold Soitec Smart Cut licenses.

That is pure operating leverage if you ask me.

The company also has operating leverage in the form of its own manufacturing capacity.

The company has publicly disclosed their capacity too, on their site.

It’s very likely that Soitec can rellocate their mix within the SOI lines, I’m just assuming this because BERNIN 2 for instance says that they can produce multiple SOI types.

But the royalty revenue will impact the bottom-line (net profit) directly, and that is pure operating leverage without the risk of expanding capacity.

Pure profit play.

Which means any potential bottleneck that leads to a price hike would also be captured via the royalty stream.

From Soitec’s filings: in FY2024-2025, Soitec reported €891 million of consolidated revenue, and its customer mix table shows “Other customers/royalties” = 13% of revenue.

We don’t know how much of “Other” is purely royalties. But 13% isn’t small, especially when it comes with almost 0 cost. Also, most of the end markets where these SOI wafers are used in, are only actually coming to market in 2027. So this is very early.

4- Roughly, what will the demand for SOI wafers be in future

These SOI wafers are used to make Photonic Integrated Circuits (PICs) which are used in things like Transceivers (think AAOIs recent management call about the insane demand for transceivers for 1.6T) as well as is CPO (Co-packaged Optics).

We have already had real world proof from AAOI incoming orders and management commentary that this space will be insanely massive.

Companies like Coherent, Lumentum, AAOI, make products that use PICs (photonic integrated circuits). The insane demand we’ve seen in this space will certaintly lead to a massive increase in SOI wafer production.

The fact that most SOI wafers produced use Smart-Cut technology mean that Soitec will certainly benefit.

Let’s try and quantify it, roughly.

The evolution of PICs used in Datacenters:

I will publish my full demand/supply model for the wafers in the next post (free). It will require more work because their are different kinds of SOI wafers to account for, which have different end uses.

The setup is compelling and demand appears to be building rapidly, and Soitec looks unusually well placed to capture that upside. At roughly 6.5x EV/EBITDA, the stock still does not strike me as priced for that possibility, which made it easy for me to add to the position.

I am long.

My conviction really strengthened once I built out the valuation, which is where I think the numbers start to get interesting.

5- Valuation- and how that relates to how fast the company will grow.

This is how I have modelled the business based Soitec’s own SOI wafer capacity. I have also used bullish assumptions.

Read more

I am quite chuffed with JCU Corporation, up 47% since my post in December. It hasn’t blasted off like other bottleneck plays- but it was always the most compelling Risk to Reward play for me, and I had this sitting at a 40% position initially.

I have reduced exposure because my portfolio suddenly became a one-trick pony, and I’d rather not have all my eggs in one basket.

JCU Corporation (4975.T) Update

• Increased operating profit by +16.4% YoY with the chemicals segment up +9.5% YoY.

• One question I have been trying to answer relentlessy is the bottleneck potential for names like JCU Corporation and C. Uyemura.

• JCU Corporation has more optionality towards hybrid bonding in HBM with its TIPHARES brands - so it might see more upside late 2027, 2028.

• C. Uyemura has more direct exposure to the current AI buildout in its chemicals segment “Wafers & PKG”.

But could the chemicals used hit a bottleneck?

The graph below is for C. Uyemura specifically, BUT, and huge BUT, this information relies heavily on my understanding of their processes.

A bottleneck would occur for chemicals because qualification takes long, there could be a point at which engineers are waiting for new capacity to be qualified but chemicals production capacity hit a bottleneck. But- huge disclaimer- I need to confirm this. If I can, JCU and C. Uyemura could be an early bottleneck play.

I am looking for any expert in this field to please reach out to me regarding the bottleneck potential here.

The other AI adjacent plays I profiled: Wasion Holdings and ChongQing Machinery & Electric Co LTD have both done even better. - Sometimes just recognizing the value others are putting out on Substack and X is all you need to beat the benchmark twice over.

Ever grateful to for ChongQing for Wasion Holdings. If you want some alpha- follow these guys. with only 456 subs on Substack? Undervalued.

Wasion Holdings (3393.HK) Update

Wasion Holdings doubling in share price since the mention. And I took some profits yesterday. But I am also still holding a position as Wayon Energy gets spun-off.

Wayon Energy has three business segments:

1- Smart Distribution Network: Smart Switchgear, high-efficiency transformers, and “Smart Distribution” solutions.

2- Data Center: Power infrastructure projects for power distribution modules, IT containers, HVDC power supply systems. There will be massive growth for this many years out.

3- Energy storage. Energy storage systems and PV + storage microgrids.

• The big story is value-unlocking. Wasion is proposing to spin-off Wayon Energy.

• I believe the share price has increased in part due to the value-unlock and most-part due to the fact that the market realizes its an AI buildout beneficiary.

ChongQing Machinery & Electric Co LTD (2722.HK)

2722.HK doing a solid 40% but still looking fairly priced and still largerly exposed to the AI buildout in China, which seems to lag behind the U.S.

So what’s next? What do I do with my big bag of profits?

Well I have two new picks. One is quite under-followed, the other has very recently started making the rounds on X.com

I believe both are worthy investments.

Read more

For my non-paying subscribers, the paywall initiates just before the meat of the content- Table of contents item 5- Meeting with management. They have provided us with some valuable information that updates our thesis, and gives us relevant information as to where we need to pose our next questions for when we meet with them again.

Table of contents

• Executive summary

• TAM map: advanced copper interconnect chemistry (2025–2030)

  • PCB: HDI microvia fill + fine-line (mSAP/SLP)

  • Packaging (PKG): RDL, Cu pillars, TSV

  • Semiconductor: hybrid bonding + TGV (early-stage)

• Management update (paywalled)

  • What changes vs. the original thesis

  • What was reinforced about business quality

• What it means for the thesis

• Disclosure & disclaimer

• Appendix: sources used for TAM charts

Thank you

I am grateful to my paying subscribers, who enable me to do this work. Quick note before we start: I’m trying to keep this Substack high-signal. I’d rather publish less often and only when there’s genuinely new, decision-useful research to share. If you have preferences on format or frequency after reading this post, message me, this is funded by you, and I’ll adapt.

Executive Summary

Advanced copper interconnect chemistry is a small, high-ASP (average selling price), oligopolistic niche. Each layer of the stack, PCB, PKG, and Semiconductor, has a different growth and moat profile.

However, most of the money (medium-term) will be made in high-end PCBs where HDI Microvia Fill, and Fine-Line Pattern (mSAP), the chemical processes which are used in tandem, lead the growth.

The electronic consumer industry is at the forefront of this demand, with smartphones and their high-end PCBs, at the helm.

Guess who dominates that market? You guessed it… yours truly- The Chosen Stock.

An overview of the TAM for advanced copper interconnect chemistry (via-fill + fine-line + packaging interconnects).

Advanced copper interconnect chemistry is a niche, high-ASP consumables market spanning PCB microvia fill + fine-line patterning, plus advanced packaging interconnects (RDL, Cu pillars, TSV, and eventually hybrid bonding / TGV). Suppliers exist in an oligopoly and tend to be concentrated in the highest-spec processes, and growth is driven by tighter interconnect requirements across smartphones, AI infrastructure, and advanced packaging.

To make it simple, we are looking at three layers of the stack.

1. PCB (Printed circuit board)- The big green slab.

2. PKG- The middle green slab.

3. Semiconductor- The top grey slab.

• PCB (Printed Circuit Board): Think of this as the foundation. It is the sturdy baseboard that physically holds the entire system together and connects the memory (RAM), USBs(ports), power, storage (SSD).

• PKG (Package Substrate): Think of this as the Brain Stem / Translator. The Chip speaks a language of "nanometers" (microscopic), while the PCB speaks "millimeters" (visible to the eye). You cannot connect them directly; it would be like trying to plug a lightning cable into a wall outlet without an adapter. The PKG translates the Chip’s microscopic, fragile signals into a larger, tougher format that the clumsy PCB can actually grab onto.

• Semiconductor (The Chip/Die): Resting on the very top, this is the brain. It is the actual silicon slice where all the computing happens, but it is too fragile and intricate to touch the main circuit board directly without the PKG bridge beneath it.

Within the stack, you can see the copper interconnects.

We’re going to focus on the copper plating consumables market, starting at the PCB layer (microvia fill + fine-line).

PCB (Printed circuit board)- The big green slab.

The image below is my best attempt to conduct a market analysis on the TAM for the chemistry that goes into PCB, and PKG - the TWO green slabs, specifically.

Use the chart as a map: the PCB profit pool is dominated by two things — (1) vertical connectivity (HDI microvia fill) and (2) horizontal wiring density (fine-line patterning via mSAP, often inside SLP boards). Everything else is smaller today, even if some categories are growing fast.

Most modern smartphones use HDI (laser-drilled microvias + build-up layers). The highest-end smartphones use SLP, which is still HDI, but with much finer wiring and is typically enabled by mSAP (Modified Semi-Additive Process- it is a manufacturing technique).

1. HDI (High Density Interconnect)

• What it is: The standard for modern smartphones and laptops for the last decade.

• The Tech: Unlike old-school green circuit boards, HDI uses “Microvias” (tiny laser-drilled holes) to connect different layers of the board. This allows components to be packed tighter together.

• The Process: Circuit formation is often subtractive, but real boards can mix approaches by layer depending on cost/yield/design targets.

2. SLP (Substrate-like Board). (Fine-Line Pattern mSAP).

• SLP is best described as an ultra-high-density form of HDI that moves PCB manufacturing toward package-substrate-like feature sizes and process control.

• What it is: The “Ferrari” of motherboards (or main logic board for you technical guys). It is the next evolution after HDI, used in flagship devices, like the latest iPhones, Samsungs.

• Why “Substrate-like”? It is still technically a motherboard (PBC), but it is built using techniques borrowed from semiconductor packaging (IC Substrates). It bridges the gap between a chip and a board.

• The Tech: It allows for materially finer line width/line spacing (often enabled by mSAP). This shrinks the board size dramatically, leaving more room inside the phone for a bigger battery.

Following the money, HDI Microvia fill and Fine-Line Pattern mSAP (SLP) are where we should keep our focus on. SLP sits on the advanced end of the HDI continuum.

The chemicals for HDI/SLP’s biggest market is mainly smartphone PCBs, but growth is also spurting out in nascent industries that, over time, will add a considerable amount of TAM to this market:

1. Self-Driving cars.

2. VR Headsets.

3. Data centers (this is specifically regarding PCB boards, not PKG substrates)

This is an industry that is positioned to grow however which way you look. I didn’t add smartphones here because it would dwarf the categories below and make the chart look a bit silly (as shown after).

Smartphones currently dominate this market.

The takeaway here is that via filling for high-end (PCBs) for smartphones lead the nominal growth in the advanced copper interconnect chemistry market, but other industries, like automotive ADAS modules and VR/AR, are growing at such a rapid clip (30-50% CAGR) that they are bound to eventually contribute meaningfully to via filling too. There are other nascent industries like advanced robotics, which will, over the long-term, significantly contribute too.

PKG- The middle green slab

The PKG Substrate is where things get more advanced. I do not know different ASP (average selling prices) yet.

The largest advanced packaging interconnect chemistry pools today include RDL and Cu pillar plating; TSV (Through Silicon Via) is the main ‘via’ analogue at the package level.

The PKG substrate is where interconnect chemistry gets more specialized. Here, the big consumables pools are less “via filling” and more advanced interconnect plating, especially RDL (fan-out / redistribution wiring) and Cu pillars (bump replacement for performance packages). TSV is the closest analogue to true via-fill at this layer.

Note* this is a different chart, it is the analysis of the chemistry market for the PKG-Semi layer

*Note: for TGV and hybrid bonding, CAGR is shown as “0%.” (not meaningful) because the markets are too early-stage for clean growth math off a small/zero baseline.

RDL (Redistribution Layer):

• TAM (~$150M, CAGR 19% est 2025- 2030)

  • Status: Common Practice.

  • Context: This is widely used in “Fan-Out” packaging (used in smartphones and processors). While the technology is established, the industry is pushing for “finer” RDL (smaller lines), which keeps the chemical requirements high.

Pillars (Cu Pillar):

• TAM (~$250M, CAGR 13% est 2025- 2030)

  • Status: Standard.

  • Context: These have largely replaced older “solder bumps” (C4 bumps) for performance chips. They are the workhorse connection for the last decade of processors.

Semiconductor Level

Hybrid Bonding:

• TAM - Yet to be seen. Average selling prices will probably be high, but undetermined as of now.

  • Status: New / Ramping Up.

  • Context: This is the “next big leap.” It abandons solder bumps entirely to fuse chips directly (copper-to-copper). It is currently used in high-end 3D stacking (like AMD’s 3D V-Cache) and is the future for “stacking” logic and memory. In the “Advanced Packaging Interconnect Chemistry Market Model” Below TAM is very low- but only because there is very little information out right now- its too new commercially.

Through-Glass Via (TGV):

• TAM - Yet to be seen. Average selling prices will probably be high, but undetermined as of now.

  • Status: Emerging / In Development.

  • Context: This is the most “futuristic” item on the chart. Glass substrates are being developed to replace current resin/organic substrates because they are stiffer and handle heat/signals better. Intel, and others are targeting this for ~2026–2030. It is largely pre-commercial for mass logic production.

Putting it together: the medium-term profit pool for advanced chemistries is concentrated in advanced PCB workflows (HDI to SLP/mSAP), while the packaging and semiconductor layer adds longer-dated optionality (RDL, pillars, TSV, and eventually TGV/hybrid bonding).

With that context, the key question is simple: how much of the “cash engine” does The Chosen One control today, and what does management’s latest commentary change in our original thesis?

I think you will be pleasantly surprised. Let’s dive in.

5. Meeting with management

Subscribe now

Read more