# A Week of Serenity's Public Posts, Lined Up: The Bottleneck Moved to the Source, and the Money Is Borrowed

Source: Realpha Blog (blog.getrealpha.com)
Original article and charts: https://blog.getrealpha.com/en/blog/serenity-x-week-2026-11-02/

> Serenity, an independent researcher on X, posted a batch of public notes between 26 and 30 October 2026. This one is written from both sides: first the strongest version of his case — roughly $40B of financing turning into accelerator purchases in the first half of 2027, and an upstream bottleneck that has moved from modules to indium phosphide substrates and laser dies — then the hardest objections and counterexamples, then which side I think holds. Notes on industry structure: no tickers, no recommendations, no price targets.

Published: 2026-11-02
Locale: en
Tags: indium phosphide, lasers, memory, AI infrastructure, vertical integration, how to read information

TL;DR: This week the tight spot moved one tier upstream, from modules and lasers to raw substrate material — and the money buying all of it comes from the capital markets. When both of those hold at once, the field worth watching shifts away from pricing and onto something else: whether anyone upstream has started collecting prepayments from buyers.


![Inside a compound semiconductor crystal-growth plant, the nearest furnace glowing orange and running while a row of dark furnaces waits behind it, the aisle receding into depth past where the eye can follow](/covers/serenity-x-week-2026-11-02-cover.png)

> Half an acre of square pond, a single opened mirror,
> where the light of heaven and the shadows of clouds drift together.
> How does the water keep so clear?
> Because a living spring feeds it at the source.
>
> — Zhu Xi, "On Reading" (my own rendering)

The batch of public posts Serenity put out on X between 26 and 30 October 2026 sets two things side by side: roughly $40B of financing that will turn into accelerator purchases in the first half of 2027, and an upstream bottleneck that has moved from modules up to the source — indium phosphide substrates and laser dies. I drew only on his public timeline and touched nothing he charges for; the figures come from public earnings calls, public interviews and public news, with a few resting on newspaper quotes and sell-side views, which are a notch softer, and I flag those in place. Two limits on all of it: the next leg of that money depends on whether the credit market stays open, and so far "raw material is short" comes only from suppliers, with no sign yet of buyers paying up front.

## Standing on His Side First: Where That Money Goes

The post he opened the week with lays out a short path for a pile of money: a space company closed roughly $40B of financing, and the primary use of it is buying accelerators in the first half of 2027. The line he added next was, to me, the most useful thing he wrote all week — that purchase indirectly lifts demand for memory and networking.

What makes it useful is that it replaces the vague phrase "AI capex" with a path that has an order to it. The money lands first at a company that wants compute and becomes an accelerator order; each accelerator board needs high-bandwidth memory (stacked memory dies packaged right next to the compute die), and the racks need optical transceivers between them, so the same pile of money shows up on three different industries' order books in three different quarters. Once you have the order, you get the lag: product shipping in the first half of 2027 means capacity reservations and long-term agreements had to close in the second half of 2026.

![A chain runs left to right along a timeline, starting from a funding round, passing a node on the earlier half-year mark, and ending on the later half-year mark where it splits into three lines.](/figures/funding-path-and-lead-time-en.svg)

That path had two independent landing points this week. One is his thank-you to a company for turning free cash flow positive at roughly $3.4B in the first half, followed by a complaint that the AI infrastructure names from memory through optical sold off far too hard — his reasoning being that the billions raised by frontier labs and hyperscalers end up flowing back along that same path into accelerators, networking and memory. The other landing point is memory. On 28 October he relayed a Korean memory maker's earnings call, which carried three qualitative claims: the memory supply shortage persists through 2030; its own DRAM and HBM volumes through 2029 are effectively sold out; and it has signed long-term agreements covering more than 50% of 2027–2028 capacity.

That same post also listed revenue and operating profit figures, with an operating margin of 79.7%. I cannot reconcile those numbers against any public earnings call, and the shape of them does not look like a memory maker, so I did not use them as material. Those three qualitative claims are the only thing I took from this section.

## The Hardest Part of His Case: The Bottleneck Moved to the Source

Demand getting bigger is not enough to carry a week of reading on its own. His second thread this week is about supply, and the location has changed.

On the same day he quoted a Taiwanese epitaxy maker in the press: indium phosphide substrates are in serious shortage. Indium phosphide is a compound semiconductor substrate material, and most of the laser dies that convert electrical signals into optical ones inside a data centre are grown on that kind of wafer. He also quoted an investment bank's view that the primary bottlenecks ahead will be three things: indium phosphide substrates, electro-absorption modulated lasers, and continuous-wave lasers — the first of those modulates data onto the light, the second supplies the steady light source.

I only saw the weight of this after lining it up against last week's public news. Last week the scramble was over advanced packaging, substrates and test — the "make it" tier. This week the tier being named is raw material. Every step further upstream stretches the lead time for adding capacity: module lines run in quarters, laser dies have to clear customer qualification, and crystal furnaces and substrate plants run in years.

The epitaxy maker added one more line, which he copied down: whoever can secure more raw material supply will have greater pricing power and influence in the market. That sentence turns an industry structure into a testable prediction — if the raw material tier is the bottleneck, companies close to raw material start getting fought over as strategic assets.

On 29 October the matching move showed up. He posted that a networking chip maker is in the final stages of acquiring a laser and optical components maker, with the comment that this is an attempt to vertically integrate the most supply-constrained chokepoints, with lasers. He then reasoned that it makes other laser players strategically more valuable.

That step upgrades the whole thread from "some things are short" to "someone will buy a company to solve the shortage." You cannot buy enough volume on the open market, so you buy ownership of the capacity instead. An acquisition's motive is harder to fake than ten supplier complaints — complaining costs nothing, while an acquisition has to be signed, cleared and booked.

## Taking a Different Chair: The Same Evidence, Read Another Way

Having read all that, the part that unsettles me sits in the first section, and the supply half comes out comparatively fine.

**The money is borrowed.** That $40B is financing, not operating cash flow. He spells the path out himself: financing comes in, becomes accelerators, flows on to memory and networking. But the first link in that chain hangs on whether the capital markets' door stays open. If purchases in the first half of 2027 depend on financing closed in late 2026, then a move in the price of credit can push that first link out or shrink it — and the three tiers behind it cannot see that happen. All they see is orders arriving a few months late. Worse, there is a self-reference here: financing gets counted as evidence of demand, while the basis for the financing is optimism about this industry, so optimism becomes its own evidence.

![On the left a short chain rests on a ground line; on the right three nodes form a closed loop with one hinge marked on it.](/figures/funded-demand-self-reference-en.svg)

**"Sold out" and "retaining pricing flexibility" sit in the same sentence.** The memory maker signed agreements covering more than half its capacity while retaining pricing flexibility, and those two things together establish only that volume is locked. The strength of a phrase like "sold out through 2029" rests on how much of the agreement is noncancellable and what the penalties are, and none of that is public. Flexibility also cuts both ways: in a good market the seller uses it to raise prices, and in a turning market the same clause becomes the buyer's door into asking for a discount.

**The evidence for the raw material shortage is softer than the demand half.** The claim that indium phosphide is in serious shortage comes from a company that sells epitaxy and related materials, said in a newspaper, and "whoever holds raw material holds pricing power" comes from the same mouth. Shortage talk has always favoured sellers, so that sentence counts as an independent confirmation only once it comes out of a buyer's mouth. The bank's bottleneck list is sell-side research too. There is nothing at earnings-report hardness in this tier yet: no substrate maker has disclosed capacity booked out to a given year, and no buyer prepayments have surfaced.

**The acquisition is still a rumour.** He wrote "final stages," and deals in final stages break. There is also a direction he does not address: vertical integration can cut the other way for peers. The acquired laser maker's capacity will favour its new parent's own designs, which removes a slice of supply from the open market — and to everyone else buying lasers, that reads as supply disappearing. As for "other laser players become strategically more valuable," that holds only if the acquirer keeps selling outward, which is decided after the deal closes, and nobody knows it now.

![On the left a pool holds five blocks; on the right the same blocks appear again, with one boxed inside an acquirer and four still left in the pool.](/figures/vertical-integration-removes-supply-en.svg)

## The Second Cut Against Him: That Basket Undercuts His Own Case

His highest-volume output this week was a basket, and it lands right on the soft spot in his own case.

On 27 October he announced an "It's All Over" portfolio with five screening conditions: fell off a cliff, trading below where its catalysts were announced, improving fundamentals, the chart looks a certain way, and short interest spiking. He was blunt about it: this is not a recommendation, he just wants to see how the group that fear hit hardest performs into year end. When someone criticised the rigour, he replied that he was not doing a rigorous analysis. I respect that posture, so what follows is not a search for slips of the tongue. What I want to work out is what the basket can prove.

Very little. The list spans space, mining, storage, silicon photonics materials and data centre power — several unrelated industries — and half the entry conditions are price and positioning shapes: drawdown, chart, short interest. A basket like that measures the degree of fear in the market and the force of high-beta (amplifying the index's moves) rebounds. If it rallies into year end, that proves fear receded; it says nothing about indium phosphide being short. If it keeps falling, it refutes none of the supply reasoning above. The basket and his upstream bottleneck thesis carry no information about each other, so neither can vouch for the other.

![Two circles of the same size sit apart with a gap between them, each labelled inside with what it measures.](/figures/two-circles-no-overlap-en.svg)

The sharper cut is in another of his own posts. A silicon photonics materials company fell from $18 to $4; he said he had been negative on it, but was surprised to find it below where it was when it announced a partnership and below last year, so he put it in the basket. The next day he added: I said that at $12, and now it is $4, and I do hold a small position.

I stopped on that one. Here is a ready-made case of the narrative direction being right while the price still got cut by three quarters. A judgement about industry structure and whether your holding period can survive it are two separate things: you can correctly identify a bottleneck tier while a particular company in that tier gets ground down inside your correct window by financing terms, dilution or customer losses. His conditions "improving fundamentals" and "fell off a cliff" holding at once describes, on the kind reading, a mispricing — and on the unkind reading, a market that has seen the fundamentals turn before you have. The same screen cannot separate those two.

![Two lines part ways on the same chart: one runs almost flat to the right, the other slides from a high point all the way down toward the bottom.](/figures/narrative-flat-price-crash-en.svg)

One more thing from this week has nothing to do with industry judgement, and I wrote it down anyway. Someone used their own reach to name companies that would die off, and he replied that hurting smaller competitors that way might set an unhealthy precedent. I take it as a note on source quality — an account willing to say that on behalf of a competitor it is itself negative on looks less like one managing a position.

## Which Side I Take, and the One Condition It Turns On

With both sides laid out, my judgement is that the supply half holds, the demand half takes a discount, and one condition decides it.

The discount first. All the demand-side evidence is either "someone announced a pile of money" or "someone said it will be short until year X," and what those share is that saying them costs nothing. One piece of supply-side evidence is different: the acquisition. It has to be signed, cleared by regulators and booked, and the motive needs no guessing — you cannot buy enough volume on the open market. So I treat "upstream is tight enough that someone will buy a whole company" as the hardest thing this week, and "$40B becomes accelerators in the first half of 2027" as a conditional prediction, conditioned on the credit market.

The deciding condition I write as a shape you can look for: has the raw material and laser die tier started showing buyer prepayments? Last week the packaging, substrate and test tiers showed exactly that shape — downstream customers putting up cash to expand a supplier's capacity and lock quota. That is what this week's raw material tier lacks: so far only the suppliers themselves are calling it tight, in a newspaper. If over the coming months a substrate or laser die maker collects a capacity reservation payment, or discloses capacity booked out to a given year, then "the bottleneck moved to the source" graduates from supplier talk into buyer behaviour, and his case holds. If several quarters pass with nothing but verbal tightness and no move in pricing, that tier is most likely borrowing momentum from the tier below it.

The next part took me the longest to think through, and I suspect plenty of people are stuck where I am: fundamentals are improving, the stock is cut in half, so how do you separate a mispricing from something genuinely broken?

I will admit up front that I have no clean answer. But this week's material helped me break it into three layers to work down through.

The first layer asks about cash. "Improving fundamentals" has to name the field that is improving: revenue, gross margin, or free cash flow. The company that reported roughly $3.4B of positive first-half free cash flow and a company growing revenue with negative cash flow are two different animals in a drawdown. The first is losing valuation; the second is moving toward its next financing round while it falls, and the terms of that round are set by the market — so the price decline itself becomes a cause of deteriorating fundamentals. Separate that layer cleanly and half the question is solved.

![On the left a two-step chain runs to its end and stops; on the right three steps curve into a ring with an arrow folding back to the start.](/figures/two-creatures-in-a-drawdown-en.svg)

The second layer asks whether the company's position in the chain can be routed around. That vertical integration story demonstrates one version of the risk: once a chokepoint is bought, the people who used to buy from it go cultivate a second source, or switch to a technical path that skips the chokepoint altogether. So I ask whether demand exists because only this company can make the thing, or because only this company currently has the volume. The first keeps customers through a drawdown; the second gets diluted the moment customers widen the supplier list.

The third layer asks whether time holds. Clear the first two and a practical problem remains: the window in which you are right may be longer than the window you can hold. That company down from $18 to $4 still has not had its narrative direction refuted. The approach I have tried is to write "why I think this is right" and "how long before it has to show me something" as two separate sentences — and if the second one cannot name a concrete event and a date, I treat the judgement as not yet ripe. It has not spared me a single drawdown, but it does tell me which field to look at while something is falling.

## One Thing to Take With You

Next time you see a headline about some raw material being in serious shortage, ask one question first: who said it, and who has paid money for it.

A supplier calling things tight, and a bank adding it to a bottleneck list, cost nothing between them. Signals worth something come in two shapes: a buyer putting down a capacity reservation prepayment, or someone buying the companies in that tier outright. This week I saw the second and not the first, so I file the thread as one hard piece of evidence plus a field still waiting to be filled, short of a conclusion.

![On the left two short blocks sit on the ground line; on the right two tall blocks stand, one filled in solid and the other only a dashed outline that is still empty.](/figures/who-paid-for-it-en.svg)

The practice I have been trying is to add one column, in my notes, beside every claim that supply is tight, and write only "who paid." Most of the time that column is empty. Empty does not make the claim wrong; it means the thing is still at the talking stage, and the distance between talk and money is how much time you have left to look at it slowly.

If you want the raw material behind this piece, find Serenity's account on X — the original threads from 26 to 30 October are all still up, including that bottleneck list and the screening conditions for his own basket.