I was plugging a USB microphone into the back of my desktop PC when the screen went black, the fans stopped, and every light on the machine died. No warning. No error. No graceful shutdown. Just instant, total death.
The port was tucked against the case faceplate at an angle, so I had to push and finagle the connector to get it seated. I didn't think much of it. Then the machine turned off like someone pulled the plug from the wall.
It wouldn't turn back on.
What actually went offline
The hardware was annoying. But the real loss was what was running on it.
Over the past few months, I'd been building a personal assistant system on top of Claude Code. Not a chatbot. An actual operating layer for my life. It managed my tasks, triaged my email, drafted blog posts, filed maintenance requests for my apartment, searched for jobs, tracked my projects, ran SEO audits on my sites, and generated content for my church's social media. Every morning at 5 AM, a cron job would pull my calendar, scan my inbox, review my task list, and leave me a briefing before I woke up.
All of it ran from this one Arch Linux desktop. My terminal was Claude Code. My vault was Obsidian. My cron jobs, my MCP integrations, my custom skills, my Tailscale network that pushed files to my phone -- all of it lived on this machine.
One piece of metal debris in a USB port, and the whole system went dark.
The money to fix it didn't bother me. What bothered me was the silence. No morning briefing the next day. No task triage. No "here's what you have today." I'd gotten used to having an assistant that actually worked, and now it was just gone.
The diagnosis
I did what any sane person would do in 2026: I opened a conversation with Claude on my phone and started describing symptoms.
Symptom one: No motherboard standby LEDs. These should always be on when the power supply is plugged in, even when the PC is completely off. They indicate the 5V standby rail is alive. Mine were dark. That meant either the PSU's standby rail was dead or the motherboard's power circuitry was fried.
Symptom two: When I flipped the PSU switch on and pressed the power button, the CPU fan and one case fan would spin for about one second, then immediately cut out. This is a textbook power-on self-test failure. The motherboard tries to start, detects something wrong (bad voltage, overcurrent protection tripping), and kills power to protect downstream components.
The root cause: That USB port had residual metal fragments inside it, likely left behind from a previously broken USB device. When I pushed the microphone connector in at an angle, the debris bridged the 5V power pin to ground. Instantaneous short circuit. The overcurrent ripped through the motherboard's USB power delivery circuit and fried it before the PSU's protection could latch fast enough.
A tiny piece of metal in a port I use every day. That's all it took.
The red herring
While inspecting the board, I noticed a red spot near the USB header area. My stomach dropped. A burn mark from the short circuit, visible to the naked eye? That would mean serious damage to surrounding components.
It was paint. Just red paint on the PCB. Classic misdirect.
What survived
The good news: short circuits on USB power lines are relatively contained. The 5V USB rail is isolated from most other board functions.
- PSU (Corsair RM850x): High-end unit with overcurrent and short circuit protection. It had latched into protection mode, which is why it initially appeared completely dead. After clearing the latch (switching off, unplugging, waiting, plugging back in), the PSU was confirmed alive.
- CPU: Downstream of the failed power regulation circuit. Untouched.
- RAM: Electrically isolated from the USB rail. Fine.
- GPU (GTX 970): PCIe-powered with its own voltage regulation. Completely unaffected.
- Storage (T-FORCE 1TB NVMe SSD): Running LUKS full-disk encrypted Omarchy (Arch Linux). Data intact behind encryption. This was the critical one -- my entire assistant system, all my projects, all my configs, encrypted and safe on this drive.
The only casualty was the motherboard: a Machinist X99-RS9 on the LGA 2011-3 platform.
The dead platform problem
The X99-RS9 is built for LGA 2011-3, a socket from 2014. Replacing it with the same board meant buying into a dead-end ecosystem with no upgrade path. The CPU wasn't dead, but it was trapped on a platform that had nowhere left to go.
I didn't want to patch this. I wanted to fix it properly. So I jumped platforms entirely -- from Intel LGA 2011-3 to AMD AM4. New Ryzen 5 5500 processor, new Gigabyte B550M K motherboard. A meaningful performance upgrade with a modern chipset and an actual path forward. Everything else carried over: the RAM, GPU, NVMe drive, PSU, and WiFi card all dropped right in.
I ordered the parts and had them the next day.
The build
Nothing about the assembly was smooth.
I only had about half the standoffs needed for the Micro-ATX board. The motherboard sagged toward the front of the case where it wasn't supported. I prioritized the four corners and the area under the CPU cooler mount, where the most mechanical stress lands. Not ideal, but stable.
The AM4 socket has retention brackets that need to be oriented with hooks facing inward toward the CPU. I initially pulled them off thinking they were packaging material. They were not packaging material.
I also managed to plant a fingerprint directly in the pre-applied thermal paste on the stock cooler. Turns out this doesn't matter. The paste still spreads and transfers heat fine.
After assembly, the GTX 970's fans would spin up for a second, stop, spin up again, stop, in a repeating cycle. Immediate dread -- was the GPU damaged too? No. It's normal semi-passive fan behavior. The card only spins its fans under thermal load. At idle, it cycles them off. I'd just never noticed because the old build ran warmer.
The software question
Here's the part I was actually nervous about. My NVMe drive has a full Omarchy installation with LUKS full-disk encryption, originally set up on an Intel platform. Now it was sitting in a completely different AMD system.
Windows would almost certainly blue-screen on a hardware change this dramatic. Linux, though, detects hardware at boot and loads drivers dynamically. There was a real chance Omarchy would just boot on the new platform with zero changes.
I had a three-phase plan:
Phase 1: Just try it. Plug in the drive and power on. Let Linux figure it out.
Phase 2: Repair via USB. If boot fails, boot from an Omarchy USB stick, decrypt the drive with cryptsetup luksOpen, chroot in, and rebuild the initramfs with mkinitcpio -P to pull in AMD chipset modules.
Phase 3: Fresh install. Nuclear option. Wipe and reinstall. All my code lives in GitHub repos. The assistant system configs could be rebuilt.
Back online
The machine I'm typing this on right now is the rebuilt one. Faster than what I had before. The Ryzen 5 5500 crushes the old Intel in single-thread performance. The B550 board gives me PCIe 4.0 and room to upgrade further if I want to.
But the part that actually mattered: my assistant is back. The cron jobs are running. The morning briefings are generating. My Obsidian vault, my Claude Code skills, my MCP integrations -- all of it came back because the encrypted NVMe drive survived and Linux handled the platform swap.
I've been building PCs since I was a teenager, but this is the first time I diagnosed a short circuit, selected replacement parts, and walked through every step of a platform migration entirely through conversation with an AI on my phone. No forums. No YouTube teardown videos. No calling a friend who "knows computers." Just a back-and-forth with Claude, describing symptoms and getting answers in real time.
The whole experience was more annoying than anything else. Hardware fails. That's what hardware does. What caught me off guard was how much I'd come to depend on a system I built myself. When the morning briefing didn't show up the next day, I felt it. Not because the computer was expensive to fix, but because the thing I'd been building on top of it had become genuinely useful. It had become part of how I operate.
That USB port is still on the dead motherboard, sitting in a box in my closet. The metal fragment is probably still in there. I'm not checking.
Check your USB ports. Blow them out with compressed air. And maybe don't force connectors into tight spots next to the faceplate. A grain of metal can take everything offline in a blink.