This Guy Built a DIY AI Supercomputer at Home
His homemade AI rig now has 36 computers working together — and it’s cheaper than the big guys
One tech enthusiast just took his backyard AI experiment to the next level. Earlier this year, he built a cluster of 16 small AI computers called DGX Sparks in his home server rack. Now, he’s expanded it to 36 Sparks, giving him 4.6 terabytes of unified memory—enough to run multiple AI tasks at the same time without slowing down.
He didn’t just stack more computers—he turned it into a flexible AI powerhouse. Instead of using a single big machine, he split the cluster into “inference modules” that work together like a team of experts. One group runs the latest AI models, like Kimi K3, while others handle tasks like reranking information, generating images, creating videos, or processing audio. It’s like having a mini data center in his living room.
So why not just buy a single, massive AI computer? For one, cost and cooling. Big systems like Nvidia’s B200 or B300 are powerful but need serious cooling and electricity—something his home setup isn’t designed for. Plus, he wants full control over his data and zero reliance on outside servers. The DGX Sparks give him just enough power without the headaches, and he can upgrade parts easily over time.
The real win? Flexibility. He can mix and match hardware, optimize power use, and even sell parts later when he upgrades. And with new tech like Apple’s M5 Ultras on the horizon, he sees even more opportunities to push his DIY AI cluster further—all while keeping his data safe at home.
- Built a home AI supercomputer using 36 small AI computers (DGX Sparks) instead of one big machine
- System runs multiple AI tasks at once—like image generation, video processing, and advanced AI models—faster than most office setups
- Costs less than buying a single high-end AI computer and keeps all data private, with the ability to upgrade parts over time
Why It Matters
Shows how regular people can build powerful AI tools at home for less money and more privacy than big cloud services