Stranded Indium-Phosphide
Degenerate stupid idea.
Irrational Analysis is heavily invested in the semiconductor industry.
Positions will change over time and are regularly updated.
Opinions are authors own and do not represent past, present, and/or future employers.
All content published on this newsletter is based on public information and independent research conducted since 2011.
This newsletter is not financial advice and readers should always do their own research before investing in any security.
Feel free to contact me via email at: irrational_analysis@proton.me
Welcome to a very stupid follow-up to an even stupider idea.
For technical background, see this post.
Practical Semiconductor Optical Amplifiers (SOA)
Irrational Analysis is heavily invested in the semiconductor industry.
Idiot-proof summary:
SOA = optical amplifier = a DFB/laser without gratings
Above point is over-simplified. Designing an SOA has some unique complexities. Pre-amp section, tapering, weird optical beam shape…
Only a very small number of LiDAR shitcos know how to make high-power SOA.
InP world is apocalyptic. Shortage is so bad. People really don’t understand. Last week I met a new person (whos name I promptly forgot) and he gave a nice intuitive analogy as to why InP situation is worse than DRAM.
“Only 25% of the worlds DRAM goes to AI. Can just take more from consumer. 80% of InP already going to AI.”
Ignore the absolute percentages he said. The core argument is true. InP has ZERO use-case for consumer markets. DRAM… plenty. We can always just (metaphorically) shoot the smartphone industry and annex their DRAM allocation. Can’t do that with InP.
Suppose you have a design objective to make an ELSFP module for NPO/CPO that has the following core attributes.
Single wavelength (1310nm)
22 dBm power coupled to each output fiber.
8 output fibers.
Here is my mock-up of the normal way (8x discrete DFB) and the LiDAR shitco way (SOA strat).
It’s not completely stupid.
Let’s do the math to figure out:
BOM + assembly cost
InP intensity
gross margins
I will post screenshots of the model and a full XLSX for spreadsheet enthusiasts to play with.
As you can see, a 650 mW SOA has better InP area efficiency (more photons per unit area of InP) but this basic model at a device level misses two critical attributes.
You still need a CW laser to seed the discrete SOA.
Coupling losses from many lenses and isolators hurt a lot. Especially coupling losses after SOA output to accommodate beam shape issues.
Now on to product-level (ELSFP) cost modeling.
Normal way gets 51% gross margin while the LiDAR shitco way gets only 17%.
Given that LiDAR is fucking dead, I think they would be delighted to have massive revenue even if the margins are ass.
The math works. Kinda hilarious to be honest.
Here is spreadsheet. Have fun.
So now that we know this stupid idea can work, how to play it?
As I have said many times, only LiDAR losers know how to make high-power SOA. The usual suspects (Lumentum, Coherent, Broadcom, Sumitomo, Furukawa, AAOI) do not know how to make this niche device because they have been too busy making money in real markets.
The vast majority of LiDAR shitcos (both public and private) are fucking liars. I don’t want this to end up being a situation where people ping me “what about this one”.
There are only two that have viable, high-quality SOA appropriate for NPO/CPO market. One is privately held IDM (designs and manufactures) and I’m not going to tell you their name because reasons.
The other is AEVA 0.00%↑ who manufactures with Seivers as the InP fab partner and whose public claims (which I have independently verified) were used to build the COGS and InP model.
Thank you to the anon person who pointed out the public info that strongly indicates Aeva uses Seivers.
Seivers is one of my mute works on X/Twitter because I find their fans (particularly the Serenety simpleton) to be quite annoying.
Getting tired of shooting down their 8/16 wavelength monolithic array and supposed link to Ayar/AMD. Idiots, Lumentum cut everyone except Nvidia off. Ayar did not remove Lumentum from their website. Lumentum removed Ayar.
Go fucking ask Matt Sysak what is going on.
You can also ask literally anyone who has actually used the Seivers DWDM laser or Ayar Labs Supernova what they think. Everyone in industry knows what is going on and you clowns keep making the opposite conclusion from reality.
Frankly, I can still hate Seivers DWDM product and be mildly bullish the company because of Aeva SOA. The yield on the DWDM monolithic array is horrific. Seivers will may way more money (in revenue $ and gross margin $) off Aeva than they ever did off Ayar Labs. LOL.
But frankly Aeva more interesting buy than Seivers and AXTI 0.00%↑ more interesting that both of them.
Have fun degenerates. I only own Lumentum, AXTI, and AAOI.
Not bothering with this crap but dogshit is fun to write about.
Shit is so bad that Sievers got re named to Seivers. LOL. Fun take. Have to agree that if you are not at least in Lumentum what are you even doing? As for AXTI, If 80% of InP already going to AI, let’s shoot down the remaining 20%.
A forward-looking footnote on the wafer layer, since the shortage may look different a few years out. JX Advanced Metals — which, together with Sumitomo Electric, is estimated to hold roughly 40% of the global InP-wafer market — announced on June 16 a ¥120B (~$750M) investment to raise its InP-wafer capacity to 7–10x the FY2025 level over four years.
https://www.jx-nmm.com/newsrelease/2026/20260616_01.html