August 13, 2026
DIY AI or garage disaster?
We Have AI at Home Chapter 1: A Box of Scraps
Man tries to build home AI from junk, commenters bring the chaos
TLDR: A developer is trying to build a home artificial intelligence machine from cheap secondhand parts instead of relying on big tech services. Commenters are torn between calling it brave, hilariously chaotic, and possibly doomed — especially after the deafening fans, late-night drywall surgery, and risky chip choice.
A do-it-yourself developer decided the future of artificial intelligence shouldn’t live in some giant company warehouse, but in his garage. So instead of paying for a remote service, he went shopping in the internet’s bargain bin: old server parts, unloved graphics cards from a failed cloud gaming era, and a famously unpopular processor he basically roasts himself. The vibe is less “cutting-edge lab” and more Tony Stark after a yard sale — which is exactly why people in the comments were obsessed.
The biggest reaction? Equal parts admiration, concern, and laughter. One commenter called it “quite courageous” to trust AMD, a chip brand with a reputation for software headaches in this kind of work, basically setting up the community’s main suspense plot: will this thing actually run, or will Chapter 2 be a meltdown? Another commenter perfectly captured the home-project spiral when they quoted the moment the author realized there was no internet in the garage and started cutting holes in drywall at 11 PM: “Ahh, one of those projects.” Everyone knew exactly what that meant.
And then came the comedy gold. The author admitted he bought 10,000 RPM fans so loud he had to wear ear protection during setup, which turned the machine into less of a computer and more of a household aviation event. Meanwhile, one of the funniest dunks in the thread was simply: “Our definition of ‘Box of Scraps’ is wildly different.” Not everyone bought the dream, though. One skeptic argued that even if you build your own home setup, you’ll still depend on fresh model updates from outside sources. Translation: you can escape the big companies’ servers, but maybe not the big companies’ gravity.
Key Points
- •The article documents a plan to build a local home AI server instead of relying on externally hosted AI services.
- •The build centers on four used AMD V620 GPUs, originally produced for cloud gaming infrastructure and each equipped with 32GB of VRAM.
- •The article says V620 cards are relatively affordable on the used market because they were overproduced, not sold to consumers, and are considered weak for AI due to AMD software limitations.
- •The system uses an older X299 motherboard with four PCIe x16 slots and an Intel Core i9-10900X CPU selected largely for low used-market cost.
- •RAM and SSD storage were salvaged, and the article notes these parts are less critical after model loading because most computation occurs in GPU VRAM.