Marvell Losers Folding, ARM Shareholders Baghodling
Where there is smoke, there is [dumpster] fire.
Irrational Analysis is heavily invested in the semiconductor industry.
Please check the ‘about’ page for a list of active positions.
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
Admittedly, I am rather drunk right now celebrating something that happened at dayjob. Someday I will be able to tell this story.
Today lets discuss two epic fails within the semiconductor industry.
Today, I choose [extreme] violence. 😈😈😈😈😈
Let’s start with Marvell, the losers club of AI semiconductors.
They delayed their investor day.
This comes as several high-ranks have folded.
For months, I have been tracking rumors that Marvell fucked up their TSMC N3P 224G SerDes.
Here is a list of circumstantial evidence:
3+ sources tell me Trainium 2/2.5 stability is poor.
Anthropic and Amazon are both unhappy.
Network stability was repeatedly highlighted as an issue.
Public demos of Marvell’s 3nm 224G showed poor performance.
I have seen DesignCon and OFC demos.
Post-FEC results are quite underwhelming given the favorable channel and aggressive active cooling of the demo setups. (unrealistic)
Multiple hedge-fund contacts have told me that Marvell stuffed the channel with oDSP.
Multiple contacts have told me that Marvell oDSP customers are unhappy with performance.
This week, I received very concerning performance data that corroborates the above circumstantial evidence.
I believe Marvell botched porting their 224G SerDes from N5 to N3P.
Hock Tan about to perform a colonoscopy on Marvell’s optical DSP business, without anesthesia.
Right now, oDSP market share is roughly as follows:
Marvell: 80%
Broadcom: 20%
Maxlinear: LOL
Broadcom is going to gain a lot of share. Maxlinear has a change to hit high single digits share next year based on power numbers I heard and the fact Samsung Foundry is practically giving away SF4X wafers.
ARM had a bad print today.
This massive guidance miss is because of the epic ass whopping delivered by Qualcomm lawyers in the trial ARM decisively lost.
Why did ARM not telegraph this guidance miss better? No idea lol.
Their behavior on the call is funny.
They are slowly trying to get rid of this disclosure. V9 royalty growth is stalling because Qualcomm will stay on V8 before pivoting to RISC-V.
**COUGH COUGH TARRIF PULL-FORWARD COUGH COUGH**
My understanding was that ARM involvement in Stargate was just Grace part of Grace-Blackwell which is obvious. Thus, the announcement as “technology partner” was more of a publicity stunt to pump ARM shares so Masa can use as collateral to lever up. The highlighted sentence is very interesting. I would have thought Stargate would use x86 CPUs for general-purpose storage and data pre-processing servers.
This Morgan Stanley analyst is a moron and should be fired. Bro you get to ask questions and whiff on something this obvious? Why are you covering a sector that you clearly know absolutely fucking nothing about?
Congratulations on the win at ur day job.
Fair to say then it is highly unlikely MRVL wins 2nm Trainium? When do you expect this decision? I hear estimates varying from 4Q 25 to middle 26.
About that Morgan Stanley analyst and his softball question: is Morgan Stanley handling any of Softbank's activities, maybe in the M&A space? Maybe he didn't want to upset that lucrative apple cart 💰😁.