← 返回

2026.10.01 | Substack | Semis | Irrational Analysis Q3 2026 Holdings

原标题：Q3 2026 Consolidated Holdings, Performance, and Ideas

#semis #markets

Substack · 2026-10-01

The trading account bottomed at ~900K as Leopold blew up but it’s mostly back.

Best part is I don’t owe taxes lol

The trading account is benchmarked against the market-cap semiconductor ETF, $SMH , and my bare minimum goal is to outperform by at least 30%.

At the time of writing this outperformance is 51.55%, an acceptable outcome.

Could have been way higher but whatever. Mistakes are learning opportunities.

Account statements for the last three months and full YTD transaction history available here. Have fun looking at all the stupid mistakes I have made.

Ia Trading Stocks 3q2026

64KB ∙ XLSX file

Download

Ia Trading Options 3q2026

23.4KB ∙ XLSX file

Download

Ia Trading Jun 2026 Clean

31.3MB ∙ PDF file

Download

Ia Trading Jul 2026 Clean

26.9MB ∙ PDF file

Download

Ia Trading Aug 2026 Clean

23.8MB ∙ PDF file

Download

I have started to think of the trading account as four categories as I re-build after almost blowing up alongside the hero of our time, Leopold AGI cultist Ashenwhatever.

Market Memo] A tale of two

heroes.

Irrational Analysis

·

Aug 1

![[Market Memo] A tale of two heroes.

Irrational Analysis is heavily invested in the semiconductor industry.

Read full story

[Tier 1] Core High Conviction (~$250K and above)

Nothing new to write here. Go look at backlog.

[Tier 2] Secondary Holdings (~$100K to ~$200K)

Presently not happy with sizing. Want to double all of these but alas busy with dayjob and exercising some restraint. No new names other than Credo.

I believe their SerDes (porfolio-wide) is second only to Broadcom and Nvidia. Yes, I really do think Credo is better than Marvell. In order to make a good slow-and-wide uVCSEL system, you need a team that is cracked at clock-forwarded 32-64 Gbps NRZ ultra-dense ultra low-power custom SerDes. None of that UCIe shit.

Excited to go to OCP and interrogate the Credo people. They are rumored to have a uLED demo and I know for a fact from 5+ finance-land contacts (3 L/S hedige, 2 crossover) that they are pivoting to uVCSEL which is the right thing to do.

You can see the lead of that program admitting reality at ECOC.

FUNDA

Research|Optics: ECOC 2026 Takeaways, NPO Gains Momentum as Slow and Wide Takes Shape

We attended ECOC 2026 in Málaga last week. The debate between slow and wide and fast and narrow was a recurring topic in presentations and plenary sessions. Companies presented microLED and micro-VCSEL designs and discussed their performance, manufacturing challenges and production timelines. NPO was also in focus, with live NPO demos on the show floor and more discussion of how and when it ramps…

Read more

2 days ago · 18 likes

My anti-uLED jihad below.

CBRS Update, WOLF/NVTS Lawsuit, HBF Spec Analysis, Aeva WTF, Taalas/AMD, AAOI, TSEM, uLED Jihad

Irrational Analysis

·

Aug 9

Irrational Analysis is heavily invested in the semiconductor industry.

Read full story

Credo Sr. Director Mohsen Asad (Hyperlume founder) is tacitly admitting my arguments against uLED are correct.

SLOW AND WIDE MUST HIT 1E-12 RAW BER ON EVERY SINGLE LANE MINIMUM. FRANKLY 1E-14 PREFERED FOR MARGIN.

IF YOU FAIL TO MEET THIS BAR YOU NEED FEC AND IT DEFEATS THE FUCKING POINT OF SLOW AND WIDE.

uLED FAILS THIS BAR AT EVEN 2 GBPS. FIBER ATTACH EXPENSIVE.

uVCSEL OBJECTIVLY BETTER IN EVERY WAY. THERE IS NO ENGINEERING OR LOGICAL ITEM IN FAVOR OF uLED GIVEN THAT uVCSEL EXISTS.

[Tier 3] Degen (~$50K to ~$100K)

I want to 2x/3x/5x all of these but again exercising some restraint.

Only new name is QMCO. Someone pinged me about this. Tape drives. Spent half an hour looking into it and its hilarious.

For those of you who are unaware, there are these things called tape drives. Much cheaper than HDD but also much slower. Most people have no need for such cheap and slow storage. However, the AI boom seems to have generated so much data, there is enough spillover where tapes start to make sense for more use cases.

The kicker is long-term storage and cybersecurity.

AI is causing chaos in cybersecurity. So how to secure data?

Well you could air-cap some HDD servers but that is not perfectly safe.

You can’t unplug the HDDs and store them in a vault or something but the lubricant and mechanical bits degrade. At best HDD is safe to store 3-5 years while powered off and that is honestly kind of questionable. SSD/NAND has the same problem. Bit rot. Look it up.

Properly stored tape drives can last for up to 30 years in offline storage.

[Tier 4] Sophisticated Activist Campaign(s) (1 share each)

Cerebras is super fun. Glorious entertainment value in this ticker. It’s actually the most interesting semis/AI stock in my opinion. There are credible arguments that can be made to be extremely bullish and extremely bearish, depending on duration and what you believe is achievable in their roadmap.

Very excited to make real progress on the practical parametric yield post. It has been sitting in my drafts for over a year. Cerebras (which is disproportionally harmed by this industry-wide phenomenon) has pushed this technical-deep dive to the top of the queue. It’s gona be great. Regardless of if you love Cerebras or hate them, you will learn a lot and be entertained as well.

For the brain-dead and lazy, here are my quick opinions without any justification. You just have to wait for the full deep-dive to understand where these opinions come from.

Hot Chips 2026: Irrational Recap

Irrational Analysis

·

Sep 6

Irrational Analysis is heavily invested in the semiconductor industry.

Read full story

Subscribe to get the parametric yield deep-dive as soon as it is ready. I kinda can’t contain my excitement.

Pledge your support

Share?

Share