Stock Thoughts: Weekly Reflections
Wrestling with the limits of knowledge, AI capex risks, and the realities of a sustainable career in investing
What I Worked On/Thought About This Week
I write daily stream-of-consciousness reflections about my work and thoughts, posting them on X (@stockthoughts81) and compiling them here weekly. This helps clarify my thinking and documents my process, while hopefully sparking discussions with others. Sign up to follow along, and feel free to reach out to discuss!
2/17/25
Not much was done for work. Planned for week.
2/18/25
The HBI insider buy was interesting.
Tone is changing on INTC. Wonder how Pat feels about all this.
Admittedly, I am struggling a bit with this journal process yesterday and today. Ties back to the 2/16 post. Mind is stuck on things I cannot write about.
2/19/25
Modeling out $ORCL and I am not totally sure either a) how to do it, or b) whether it is worth it to do in any significant detail. MS has done some interesting work that basically extrapolates how each of the business lines are doing right now/next couple years into the future (FY29) and then essentially asks the question of how big OCI has to be to make up the gap so that they hit their FY29 $104b rev target. OCI would have to get to like $51b and they have $34b of that coming from AI. That is a 34x over 5 years for the AI portion. And it’s a huge drag on margins. I don’t know. How can you bet on that? Sell-side is giving them FY29 rev of $99b which still strikes me as a bit high. Or high in so far as the answer to the OCI question seems pretty unknowable when the implied growth is of that kind of magnitude. Who knows what really happens with Stargate. But - where else could you make up the gap? I don’t know. I don’t really see any of the other biz lines being hugely divergent from the ballpark level of growth they are doing right now. Also - this is broken up into Growth, Stable, and Declining buckets. Bernstein has done a similar analysis. This raises another issue, in that these numbers are difficult to track. Bernstein is not terribly far from MS on their estimates regarding these buckets, but at the end of the day, they are just estimates. And e.g. gross margins are estimated for several of the underlying line items as well. You can kind of try to back into consolidated margins, but it’s just kind of a pain that the company won’t provide this level of granularity. It makes it harder to track your thesis and see how the company is performing against it. You are always guessing. So I don’t know. This is probably going to take a little while and I will continue to play around with it. I am not sure it is worth the time investment, to be honest. Because like I said - the kind of OCI number required to make this look all that attractive just seems really tough to diligence and to have conviction in.
Since I first tweeted about “small semi co”, Kokusai is up 50-60%. Extremely frustrating when this happens. But I am not sure there is much we can do. We have to go through our process. We have to do the work. I think of the quote from Druck about just buying a little before doing the work. But I am not sure that is really feasible if you aren’t Druck. I don’t know.
2/20/25
Let’s talk about ILLUSION OF CONTROL. We have a bias machine (perhaps a topic for another time) that tries to tell us when we are being biased. Its recent criticism of me was brutal but fair: “MKTX External Insights (Meeting 119): There was a desire to seek external insights for MKTX to gain better "grip" and "visibility," suggesting an illusion of control bias and overestimating the research team's ability to fully understand and predict complex market share shifts. [Source: [me], Meeting 119]”. I don’t know if “illusion of control” is exactly the right term/bias here, but let’s run with it for a second. I think this is something that I run into a lot and I need to pay more attention to. More information doesn’t necessarily always translate to a better assessment of the relevant probabilities. As I’ve been journaling about, the current competitive dynamics around MKTX are a bit difficult to assess right now. I read Tegus transcripts, sell-side, talk to management, etc. and sometimes strong opinions come through, sometimes they don’t, but it hasn’t cleared up the underlying uncertainty. On a day where I spent much time blasting out cold messages on LinkedIn, this seems like a relevant topic. The fact of the matter, for a lot of stuff, is that people just don’t know. No one does. Kokusai may serve as another example here. You have uncertainty on multiple layers - single wafer vs batch, competitive dynamics within batch, emerging technologies (spatial, selective, etc.), etc. And while there are some basic facts that point to a favorable outcome (batch being good for high aspect ratios, strong customer relationships, etc.), there is always going to be uncertainty here, and it is always going to be more uncertain than e.g. AMAT or LRCX or MKSI. And I can have all the Tegus calls I want, it’s going to be very, very hard to get to any sort of statistically significant collective conclusion from them. So what is in my control? Choosing what rate of return to demand given a certain level of uncertainty. Purchase price and knowing what questions need to be answered for the investment to work out from there. Position sizing. Updating theses and position sizing with new information. Ironically, I think avoiding falling into this trap becomes even harder if you have endless research budget/resources. It would be really tempting to do GLG call after GLG call. So what to do about this? Pay attention to how I am allocating time. Focus on what I can control. Be honest with myself about what is knowable and what isn’t. Sometimes this is tough. Take my thoughts on ORCL from yesterday for example. I am leaning toward doing another round of work on OCI to try to see if $50b rev is feasible. Is anything going to come out of that? Is that knowable? I don’t really know yet. But I figure doing the work is going to benefit our knowledge base on hyperscalers we already own anyway so why not.
2/21/25
So $AMZN Cloud AI is now multi-billion in size. I don’t know if anyone has seen a good number for this. Maybe HSD billion? $ORCL, according to MS, approaching $1b. AMZN AI growing triple digit. $MSFT reported 157% in their equivalent business. The AI stuff is a drag on margin. I feel like this should be obvious but for some reason I am having a bit of a tough time thinking it through. How can you simultaneously be capacity constrained and a drag on margins? Unless you are underpricing pretty materially? If it’s because it lacks some sort of scale - I think that is the part I am having a tough time conceptualizing. How is the scale/investment of that business separate/different than the rest of AWS? How do you cleanly separate the assets? Idk… crazy how high AWS margins have gotten. Printed 37%. Depreciation has often been a lever to drive those up. But for first time, in fwd guide, useful lives shrinking. What is the right useful life kit for the AI kits? Need to deep dive this more. The index, and so many active managers, including us, have a lot of exposure to this AI capex risk. And I am still not totally sure the right way to think about it. Or how to sketch out what the returns will be. And/or to what extent it is just an arms race and everyone has to do it to defend the moat but it’s really just a drag on returns. I wonder what sustainable margins are on AWS when this is all said and done. MSFT, meanwhile, guiding to decel in capex in 2H of this CY (1H of their next FY). And then if you think about this whole capex risk thing from a valuation/DCF perspective it’s kind of hard to know what the impact would be. E.g. risk of overbuilding is there, but maybe AMZN just slows the investment and grows into it like when they overbuilt retail. Or will competitive dynamics force them to keep investing? Idk. Probably going to continue to read, think, and write about this. I don’t know if it’s answerable. Maybe illusion of control issue again.
Is the NAND downturn all tied to overbuild/capacity/still excess inventory? What fundamental drivers may have reduced NAND demand? E.g. historically everyone bought their own storage, now doing it in the cloud. That probably, net, requires less NAND? Maybe? Store copies of digital assets if they are not unique. And then even unique stuff like e.g. pictures can get compressed.
Toured a CCOI data center today. Was pretty interesting.
2/22/25
Trying to think about what big projects are on my radar. Trying to do ORCL made me realize, both for that and hyperscalers, I need to do more of a deep dive on this AI capex stuff. Cowen note was interesting. I don’t know where this will all shake out/if I am just under illusion of control. But I feel like I need to have a better understanding of it than I do now, and I’ll see if I am able to form a stronger opinion on it. This may entail a deeper dive on NVDA as well. I actually have never really done the work on NVDA. What if it’s cheap? After all this? That’d be wild. And then - CVNA may actually qualify for our process now. It might still be cheap. The competitive dynamics and unit economics have proven themselves out to a much greater extent than when I last owned them around these prices.
There are a couple things I want to try to implement from a productivity perspective this week. Schedule when I will use (or won’t) certain techs - email, socials, phone. Batch admin (e.g. email) according to content. Be more specific with my admin blocks. Don’t just time block “admin”. If it’s just email, say that. If it’s email and following up on two specific items, say that. If it’s email/socials/phone/free-for-all, say that.
Struggling with a bit of - I’ll call it imposter syndrome for lack of a better term - lately. Sometimes I worry I chose a bad game to play. It doesn’t get more competitive than this. Can I win without being a workaholic? Should I have stuck to microcaps or something else entirely that is less zero-sum? The normalization of workaholism in this industry is something I am trying to make sense out of. Something seems off. These thoughts come and go. Maybe that’s normal, maybe it’s not, I don’t know.
2/23/25
I tried something new today - the below is a summarized capture of a conversation I had with ChatGPT using voice notes. Figured I'd experiment with that for journaling.
Been thinking a lot about work-life balance and what "enough" looks like in terms of both hours and ambition. It's interesting how relative these concepts are - put me in a room of NYC pod shop guys and I might seem laid-back, but put me in a room of typical professionals and I'm probably seen as quite driven.
I think there is a lack of honest, nuanced discussion about sustainable success. The people at the extremes yell the loudest - either hardcore workaholics promoting 100-hour weeks or those completely rejecting ambition. But where are the voices in the middle? I wish more people would openly share their real daily schedules, their actual work hours, how they structure their weeks. Not the curated, social media version, but the messy reality of trying to balance meaningful work with a full life.
We see the extreme examples - the Warren Buffetts and Michael Jordans - but even those stories make you question whether that level of singular focus leads to fulfillment. For every success story of extreme work hours, there are countless others who worked just as hard but didn't land on top.
I've been wrestling with what the right balance looks like. I enjoy investing, but I don't think I want to do anything 80 hours a week - that's not a full enough life. I want space for fitness, sports, friends, church involvement, and eventually family. Success for me is multifaceted - it's about living according to my values, being able to provide comfortably for myself and future family, and creating genuine value through my work.
I suspect there's a middle path where you can achieve meaningful success without sacrificing everything else that makes life rich and meaningful. But finding that path is harder when there's so little transparency about what it actually looks like in practice. I wish more people would talk honestly about their struggles with this balance, their doubts, their experiments with different approaches.
What I do know is that I want to optimize for a full life, not just a successful career. The goal isn't to make as much money as possible, but to make enough to live comfortably while maintaining space for relationships, health, faith, and personal growth. Beyond basic comfort, I'm more interested in the quality of my hours than the quantity of my earnings.