Uses

At the end of 2025, I was in the market for something more current. The iMac was now about 12 years old, Apple had stopped supporting the model years ago, and I’d had to restore it from backups at least twice. I’d been talking to my cousins overseas about their own setups; they’re both gamers and one of them had been into custom builds for a while.

At this point I knew several people from work who had built their own custom PCs, most of them long ago… they had taken different paths to mine, and were already deep in the world of Cherry MX switches and Arduinos by the time I arrived. My dad used to build PCs, actually, when I was very young – I remember sitting at the back of his homelab playing Beyond Good and Evil, Etherena Beta, alongside a few incredible titles my brother and I found at the bottom of bargain bins in Camden.

I used Linux/Ubuntu at work and had gotten pretty comfortable with it over the years, but everything was managed – I’d never been sure if it was actually that approachable, or whether the sysadmins had hidden the sharp edges from me. I had some down time, so instead of buying another MacBook Pro, I saw December as an opportunity to build my own PC from scratch.

The way I use a computer has changed drastically since entering tech; today I typically run tmux via WezTerm with a vertical split: one side flickering between NeoVim and the command line, the other running Claude Code. For theming, I switch between Tokyonight (the LazyVim default), Rose Pine and Kanagawa Wave.

I learned Vim out of sheer curiosity a year before; those first few weeks learning Vim bindings were brutal, but I stayed consistent and it paid off big time. If you’ve ever thought about it, do it – I promise you will get faster, and even faster than that.

For the build, I wanted to craft something that would let me explore the areas of technology that most excited me moving into 2026. Like many others, I had started playing around with AI through chat interfaces, but soon found myself wanting to experiment with models via APIs. Working with these cloud-based frontier models isn’t like learning a new framework… I couldn’t pay my way into understanding with my time like I did with music, or even coding. Every API request cost real money, and those 10,000 hours weren’t going to pay for themselves.

I had heard about the Llama models, but had no idea what the difference was between those and something like GPT-4… I forget exactly when I discovered there were models you could run locally, but I knew that was going to be the key to figuring things out without resenting the process.

I saw a lot of recommendations to “build around the GPU” online – after a tonne of research on what exactly I’d need to run this stuff locally for the next year or two, I made the choice to “build around” an RTX 5090.

What else did I need… a CPU? Understood. Some RAM? I knew that one. A motherboard… okay? …thermal paste? Why do I need paste?

Eventually I had collated the full list, so I sought out a second opinion. My cousin brought it to my attention that I was building a cyborg.

I decided to go with Pop!_OS for this machine – I’ve become comfortable with Linux, and Pop! lets me focus on creating things. Putting all the pieces together was mentally and physically satisfying. It took about a day and a half for me to finish it; knowing the purpose of every part really changes how you engage with the whole. I decided on a Keychron K3 Max keyboard, and stuck with a Magic Trackpad for comfort.