Buy Low, Live Free

Sometimes you get lucky

Less than a year ago, I shared that this blog would go through a reboot switching from writing about crypto to personal financing. Well, it seems that post was quite prescient: 

Today's stock is Alphabet (Google). I opened a new relatively large position of about 550 shares at an average price of $182 / share.

I'll write an in-depth article soon on why I like Alphabet, but as Soros used to say "Invest first, investigate later". 

Less than a year later, that position has more than doubled and I have yet to write an in-depth article about why I like Alphabet. Well, let me provide a short summary of why I like it as a shareholder and where the risks may lie:

What I like:
Google, unlike many of the frontier AI labs, owns the full stack. From TPU chips, Google Cloud hosting, closed (Gemini) and open (Gemma) models to consumer devices, they own everything, including the ability to train the AI models from their users' data (hoovered up from Google Maps, YouTube, Search data, etc.)

From 2022 - 2024, the consensus thinking was that Google was late to the game, which in hindsight, was true. But what most people did not recognize at the time was that Google was swimming like a duck. Above the water's surface, things seemed to be status quo but underwater, the team was working around the clock. The Code Red announcement -- which Sundar must have known would be leaked -- only reinforced the urgency of the situation.

Now, following this week's earnings, it's clear that Google Cloud is a key advantage over the independent AI labs (namely OpenAI and Anthropic), who have to pay a hefty margin on top of the already competitive API pricing they are charging its customers. Since Google is vertically integrated, they need to balance serving its internal customers (Gmail, Gemini users) and its external customers (Anthropic and other hosted models). Per Sundar on a recent podcast interview, internal customers have much longer LTV than external customers, who can switch with little notice.

I like owning Google because it gives me wide exposure AI. It's like owning a basket of AI stock all under one roof. I don't have to worry about excess capex capacity because its external cloud customers can soak it up; I can't say the same for Meta or Oracle, who have similar problems but different end customers to serve. 

Given my bullish views on AI, holding Google for the long-term is a no brainer to me. 

It's quite amazing how quickly the public narrative can shift once execution is on display in the public markets. I think having quarterly reporting is a strength, not a weakness, of the U.S. stock market. 

What I'm keeping an eye on:
Given my bullish views, my concern is primarily focused on execution and over investment in the near-term. Google is projecting capex of $180-190B in 2026, while their TTM cash flow (before capex) is $175B. This year they will essentially breakeven on cash flow. Another way to think about overextension is if they allocate their internal resource to pet projects or lesser known ROI projects. AI research can be a slippery slope: How much capex (or human researchers) do you allocate to a project without knowing what the return may be? You know you need to allocate resources because the math shows that by throwing more compute and people behind a problem, the better the solution but what does that look like?

Already we're hearing that researchers, who are assessed based on their server utilization, are re-running programs to artificially boost internal metrics lest they draw their manager's ire. How much of that is going on at Google today? If AI developments hit a plateau, how will Google know if its researchers are doing real work vs. fake work?

The AI race now appears to be Google's to lose. They have all the cards, now they just need to play it smart. 

Intel


I sold Intel just a few days before the U.S. government announced their investment. It was sold via a covered call, which I thought was a fair price at the time. 

My thesis was that Intel being the sole national champion of semiconductors would never find itself facing bankruptcy and that the U.S. government would do whatever it could in its power to support it. I never imagined that the government would become an equity investor, though. My holding period was quite short, at around six months. 

The lesson I learned is that it takes a long time for a thesis to play out, especially for a capital intensive business like Intel. I should not have exited after making only a 20% gain if I had conviction in the stock. Drunkenmiller said it best: Position size matters. In this case, it wasn't a massive position but when I sold, I felt a bit of validation that I was "right" and made some money.

Drake's Gem Set Rolex

A Rolex is a nice watch, but when you add factory-set gems onto, it takes it to a whole new level. Take for example, this latest offering from Wind Vintage: 

It's an incredible watch. What makes it extra special is that it wasn't locked up in a safe, but was actually worn. Watches are meant to be worn. It serves a purpose. If you can't manage to wear your watch, then it isn't for you. To me, Drake's watch is extra special because you can tell from the scratches and nicks that this was a watch that loved and worn.

Apple's new CEO

So the time has finally come: Tim Cook is retiring later this year and has named his successor, John Ternus, senior vice president of Hardware Engineering. I liked that someone in hardware is going to lead Apple in its next phase in this new AI paradigm. I've been trying to find a Mac Mini the past couple of weeks; the Apple Stores have a backlog until July and all the marketplaces (Craigslist, Facebook, Swappa, eBay) have Mac Minis or M1 Max Macbooks priced quite high relative to their pre-OpenClaw prices.

Given the successor announcement and my difficulty in locating an affordable Mac Mini, I decided to place a small buy order of AAPL today. I opened a small position of $10,000 to start tracking the stock. From my research, I've noted that the Apple hardware differs from Windows in that the memory is unified (or shared) across OS and GPU. What that means in practice is that the memory can be used flexibly across CPU and GPU which is a powerful combination. In the Windows world, which I am most familiar with, my computer RAM and Nvidia GPU RAM (aka Video RAM or VRAM) are separate. 

When loading up local AI models, I'm limited by my GPU's RAM. Since I have an Nvidia 5080 that means my max VRAM is 16GB which is not that much; I can accommodate a model that is has about 13B parameters or so. Whereas a Mac with 32GB can use a much more powerful model. 

In my experience as a consumer trying to run AI models, my first preference would be use a local model since my costs are fixed. I don't see a massive market for on-demand inference where consumers are paying for tokens. People like Netflix and YouTube because the costs are known upfront. Given the pace of open source AI models, I expect that Apple silicone is going to be able to power ever more powerful open source models within the next 3 years. That will create demand for Apple hardware. Of course, cloud inference will always be around but will mostly be used by enterprises.

I'm excited for Apple. Having a hardware guy at the helm is the exact type of person they need to lead the company now. 

Finally installed the deployant (after 5 years of owning it)

Back in 2021, during Covid, I purchased an Artem RM-style deployant hoping to pair it with my IWC Big Pilot. But, alas, I couldn't get the fitting just right and the deployant required a tiny flathead screwdriver which I didn't own at the time.

Fast forward ~5 years later I finally found a screwdriver that fits the screw. I had it all along and never thought of using it until now. I actually bought an eyeglass screwdriver kit off Amazon for $4.50 but the flat edge was too thick so it didn't work. Of course, I was 3 days past the return window so now I have an unused, brand new screwdriver kit for whenever someone's eyeglasses' screw comes loose. 

The deployant is actually quite easy to install once the pin screw is screwed in. If makes a secure fit, much better than simply using a spring bar.

Here is the final result, which I quite like. The buckle is quite thin so it doesn't feel bulky. One thing that was surprising to me was that the deployant adds a bit of length so I actually had to size down. I'm now using the last strap hole. Shoutout to Holben's Watch Strap  for selling me the Fluco Alcantara watch strap for $45. You can get a coupon for signing up as a new customer.