QF Research

11.1K posts

QF Research banner
QF Research

QF Research

@ResearchQf

PM at large institutions. Investment decisions are your responsibility. DYODD. Analyze distributionally but non-deterministically. RTs aren’t endorsements.

Katılım Aralık 2019
373 Takip Edilen25.6K Takipçiler
Sabitlenmiş Tweet
QF Research
QF Research@ResearchQf·
1) A reminder how actually insane the runup to 2000 dotcom peak really was. $VRSN peaked at > 300x P/T12M rev. $JDSU $PMCS hit 100x. Remember ART Tech? 180x. $ZICA (Chinese text entry on handsets)? 700x. With 10-30% corrections every M or 2 as 100's stocks rose 50x 100x or ...
English
4
27
145
158K
QF Research retweetledi
Jostein Hauge
Jostein Hauge@haugejostein·
This is wild. People in *every single one* of the top US allies now think it's better to depend on China than the US. The global balance of power is clearly tilting away from the US and toward China.
Jostein Hauge tweet media
English
732
3.2K
8.4K
653.4K
QF Research retweetledi
Barak Ravid
Barak Ravid@BarakRavid·
Contrary to Trump’s statements, senior Israeli and U.S. officials said that the United States had prior knowledge of the Israeli strike and even approved it in an attempt to pressure Iran. After the Iranians retaliated against Qatar’s gas fields, Trump is now changing course
Barak Ravid@BarakRavid

🚨🚨🚨Trump on Truth Social: Israel, out of anger for what has taken place in the Middle East, has violently lashed out at a major facility known as South Pars Gas Field in Iran. A relatively small section of the whole has been hit. The United States knew nothing about this particular attack, and the country of Qatar was in no way, shape, or form, involved with it, nor did it have any idea that it was going to happen. Unfortunately, Iran did not know this, or any of the pertinent facts pertaining to the South Pars attack, and unjustifiably and unfairly attacked a portion of Qatar’s LNG Gas facility. NO MORE ATTACKS WILL BE MADE BY ISRAEL pertaining to this extremely important and valuable South Pars Field unless Iran unwisely decides to attack a very innocent, in this case, Qatar - In which instance the United States of America, with or without the help or consent of Israel, will massively blow up the entirety of the South Pars Gas Field at an amount of strength and power that Iran has never seen or witnessed before. I do not want to authorize this level of violence and destruction because of the long term implications that it will have on the future of Iran, but if Qatar’s LNG is again attacked, I will not hesitate to do so

English
676
3K
9.1K
2.2M
QF Research
QF Research@ResearchQf·
Those aren't mine but consensus which aren't 100% self-consistent as analysts provide different metrics. The point was ASPs are 1st order. Bits are 2nd order. Cons Feb DRAM ASP +42% ~ $15B+ DRAM cons. Actual ASP +mid 60s % ~ $18B actual DRAM x change in bits. Avg DDR contract Feb Q was up well above 60s % Q/Q. More stable HBM brought avg ASP to mid-60s Q/Q.
QF Research tweet media
English
1
0
4
294
siflower
siflower@siflower·
@ResearchQf Your DDR and HBM bits don't add up to DRAM? And ASP QoQ Growth % for DRAM is 19.22% for 3.92 --> 3.96?
English
1
0
0
514
QF Research
QF Research@ResearchQf·
$MU consensus. Expectations are much higher. DDR spot leveled off last few months. NAND spot continued to rise. For short term numbers, use independent product price assumptions (which is to 1st order what matters) by applying typical lags to spot. Then compare vs consensus. To repeat, what truly matters is duration and less likely short term price reaction, which is sometimes noise. If spot prices go sideways 2026 into 2027, what is your steady state earnings? And more importantly, how does memory and storage demand scale with AI applications going forward vs supply?
QF Research tweet media
English
2
15
99
15.9K
QF Research
QF Research@ResearchQf·
3) There have been many more details. That includes other suppliers at OFC as well as updates from Asian suppliers. Their 18-24 month target is $2B quarterly revenue with 40% OM. Consensus was ~$1.5B and ~35%. This target does not include impact of the new UHP laser fab with $5B annual revenue capacity. This could end up being conservative given their recent performance, but I'm more interested in 2029 - 2030. Photonics wave might just be starting.
QF Research tweet media
English
2
1
15
2.7K
QF Research
QF Research@ResearchQf·
1) $LITE is up $56 and $132 since yesterday morning. LITE presented during market hours at OFC yesterday! I may be almost 80% there on CPO scale-up opportunity through at least Feynman. There has been bunch of new info in a day. Here are 2 key LITE slides. Phase 0. Again scale-out is well understood near term. "Multi-hundred" million 1H27 alone. Quantum-X and Spectrum-X CPO switch build data later. Phase 1. That inter-rack NVL576 scale-up I've been referring to. 3x to 4x CPO links vs Phase 0. Phase 2. 3x to 4x vs Phase 1. NVL1176 also includes longer distance intra-rack due to those physical copper bandwidth-length limits (see 2nd slide). Phase 2 alone is causing those huge $NVDA (and now other customers) demand signals for LITE $COHR and bunch of other suppliers. 100% optical scale-up is inevitable. 3.2T 6.4T+. Will discuss various resulting photonics opportunities from a top down and bottoms up basis later. ASPs, units, high power (e.g. 400 mW) CW lasers etc.
QF Research tweet mediaQF Research tweet media
QF Research@ResearchQf

I might be 70% there in understanding CPO scale-up opportunity through Feynman, but some technical clarifications plus coming supplier datapoints should take that to 80-90%? Inter-rack optical scale-up for NVL576, as mentioned earlier, appears confirmed for Oberon. But that's only a first step. Scale-out near term is well understood. Orders received to date by $LITE or $COHR and their high level TAM statements seem roughly consistent. Jensen is often imprecise during presentations, but that often leads to opportunities. This is one of many AI technologies where fortunes could be made or lost over the next years.

English
6
14
139
20.5K
QF Research
QF Research@ResearchQf·
Larijani was conservative, but also a very experienced, pragmatic power broker in Iran. If Rafsanjani were alive today and in power, I'm starting to suspect we and/or Israel would've taken him out too.
English
1
0
20
4.8K
QF Research
QF Research@ResearchQf·
@vikramskr I think so because they use very simlar wording for Kyber, but these are why it feels 70% :)
English
1
0
7
554
QF Research
QF Research@ResearchQf·
I might be 70% there in understanding CPO scale-up opportunity through Feynman, but some technical clarifications plus coming supplier datapoints should take that to 80-90%? Inter-rack optical scale-up for NVL576, as mentioned earlier, appears confirmed for Oberon. But that's only a first step. Scale-out near term is well understood. Orders received to date by $LITE or $COHR and their high level TAM statements seem roughly consistent. Jensen is often imprecise during presentations, but that often leads to opportunities. This is one of many AI technologies where fortunes could be made or lost over the next years.
QF Research tweet media
QF Research@ResearchQf

@zephyr_z9 Which might be one of the reason why inter-rack CPO scale-up is being considered for NVL576

English
12
9
88
36.3K
QF Research
QF Research@ResearchQf·
@vikramskr "Vera Rubin Ultra NVL576 will combine eight separate MGX NVL racks, each with 72 Rubin Ultra GPUs, all in a single 576-GPU NVLink domain with copper and direct optical connections."
English
1
1
16
1.2K
Vikram Sekar
Vikram Sekar@vikramskr·
@ResearchQf A big question I have is whether the optical scale up for NVL576 involves pluggables or CPO. I think people assume CPO? but that means NVswitch needs CPO? which isnt there currently. so pluggables?
English
3
0
11
1.6K
QF Research
QF Research@ResearchQf·
2) Jensen also discussed the Kyber rack midplane, which also has significant M9 HLC content.
QF Research tweet media
English
0
0
13
2.9K
QF Research
QF Research@ResearchQf·
1) If ASP for 52 layer M9 Q-glass PCB is ~$200/LPU by some estimates, total PCB content would be ~$50K for a LPX inference rack with 256 LPUs. Jensen "I would add Groq to maybe 25% of my total data center." (?) Multiply that content by your Grok rack assumptions. Elite Material 2383 TT shares have been parabolic. There have been significant upward revisions for LPUs at Samsung Foundry (reportedly 4 nm). Certain prior comments from Asia about lukewarm CPX adoption also seem valid based on Jensen's presentation today.
QF Research tweet media
English
4
8
69
9.5K
QF Research
QF Research@ResearchQf·
I was parallel processing so missed this, but the C25-C27 assumption was correct. But does "$1T+ STILL GROWING" have nuance? Is there possibly some elements of backlog here (there's almost 2 years left thru C27) in which case it could grow significantly due to continuing orders à la $AVGO AI backlog discussed during its Dec earnings call? Btw this is at best the 3rd most interesting clarification I'd like. Some clarifications should come from supply chains soon.
QF Research tweet media
English
0
0
14
3.5K
QF Research
QF Research@ResearchQf·
It's hard to believe Jensen did it again. "At least $1 trillion" but over exactly what period? I assumed between C25 and C27, and if he's being self-consistent, Blackwell, Rubin & networking revenues, but I'm not 100% sure. Data center revenue was $194B in ~C25, but not all was Blackwell. That leaves something well above $800B for C26 and C27 vs $776B cons. And how conservative is $1 trillion given "I am certain computing demand will be much higher than that." Colette will clean up this evening or tomorrow morning as usual.
QF Research@ResearchQf

1) $NVDA financial team clarified those details yesterday evening to analysts. Jensen misspoke as mentioned here. If one took his statement literally, he said at least $500B next 5 quarters, but $500B is cumulative BW & Rubin over ~ 2 years thru end of C26 or ~ F27. His GTC slide is still fine and not inconsistent. For example, this is MS comment this morning after feedback from Colette's team. My back of envelope last night is roughly inline. It's not the 60% higher forward guide, but still well above consensus "so far" and doesn't including China, where Trump said "we'll be speaking about Blackwells" with Xi last night. Overall quite positive developments. NVDA also reiterated $1T sales/year, which it implied early this year, is still on track perhaps this decade. Other clarifications include BW & Rubin dies (Rubin is now NVL144) and inclusive of networking, which I would've assumed but wasn't 100% sure initially. All this is why "we really need Colette to be sure."

English
5
5
54
13.4K
QF Research
QF Research@ResearchQf·
@racetrack275 The KAIST Teralab roadmap was last summer. 40 Gb dies but implies less than 8 x 16H x 40 Gb ...
English
1
0
1
166
finánce
finánce@racetrack275·
@ResearchQf KAIST roadmap has 40Gb DRAM die for HBM5. 1d is 2028, likely 4F2. Big density jump. Rubin Ultra is 1TB. So 1.28TB is poor. NVDA wants to show 100% gen on gen jump at least. 1.28TB only ok if Feynman has NMC+stacked SRAM with tile size going to 192x192 from 128x128.
English
1
0
1
139
QF Research
QF Research@ResearchQf·
Latest $NVDA HBM. A100 80GB HBM2e H100 80GB HBM3 H200 141GB HBM3e Blackwell 192GB HBM3e Blackwell Ultra 288GB HBM3e Rubin 288GB HBM4 (384GB 8x16H 48GB later?) Rubin Ultra 1TB HBM4e Feynman 1TB HBM5 & 1.28TB 16x20H 80GB? CoPoS? Naming conventions have been confusing due to changes. HBM content shown here is per package and not per die. Both Rubin Ultra and Feynman likely have 4 GPU dies per package vs 2 dies per package today. None of this would be that surprising if confirmed. That means expected Feynman GB content won't be a big jump using this metric alone. Si intensity, HBM5 customization, ASPs and XPU units are important. The bigger change in memory content since last summer is more non-HBM DRAM and NAND (which might've gained even more momentum very recently).
QF Research tweet media
English
6
14
95
27K
iFred
iFred@iFred·
@ResearchQf The wafer thinness requirements were loosened for HBM4x, so I’m assuming a significant change in packaging with Feynman. I wouldn’t be surprised if we see “super chip” packaging become waferscale.
English
1
0
2
402
QF Research
QF Research@ResearchQf·
@latent_value7 FOCI also has LPO, which could be earlier than CPO for them. HIMX stock also diverged with FOCI despite last week. Hopefully we learn more tomorrow
English
1
0
1
228
LatentValue
LatentValue@latent_value7·
@ResearchQf Just not sure what one would expect to see before the logo officially pops up. Market has gotten comfortable that Foci is in this supply chain in the last 6 months based on their price action so something has to give. We'll see within the next 6 months I imagine, if not this week
English
1
0
1
150
QF Research
QF Research@ResearchQf·
1) $HIMX. Jensen mentioned TFC Optical as a CPO partner a year ago. $HIMX was already down significantly in the prior 2 months before last year's GTC (for good reasons). HIMX doesn't sound super duper excited about intermediate term CPO ramp today, whereas TFC remains optimistic about optical engine and related opportunities. HIMX isn't a supplier to TFC for CPO. Things can change incredibly quickly in AI infra. Perhaps there's new information tomorrow, but I haven't seen anything indicating that's very likely.
QF Research@ResearchQf

Several asked me why $HIMX was down today. One of the tech roadmap nuances I referenced was TFC or TFC Communications 300394, which is mentioned as a CPO partner during Jensen's GTC presentation. TFC supplies passive fiber array units and is $NVDA's initial supplier. Possible involvement of TFC has been mentioned in Asian supply chain. FOCI & HIMX likely participates later next year. It's still unclear to me whether scale-out and/or scale-up.

English
2
3
60
19.4K