THE MINING RIG EVERYONE WROTE OFF IS RUNNING LOCAL AI MODELS THE CLOUD CHARGES $340 A MONTH TO ACCESS.
12 RTX 3060s. Each with 12GB of VRAM. A rig built for one era that just found its second.
The article says buy a used RTX 3090 for $700. He already owned something bigger.
The technical reality: 12 GPUs on PCIe don't behave like one GPU. The cards can't pool memory the way NVLink allows. But they can run parallel inference. Multiple models. Multiple workloads.
Cloud is convenient. Local is control.
He stopped paying for compute he could run himself. The invoice stopped coming.
EVERYONE IS PAYING $20/MONTH FOR AI. HE PAID $70 ONCE. THE MATH STOPPED MAKING SENSE AFTER MONTH FOUR.
Not a discount. Not a trial. A Raspberry Pi Zero 2W, a PiSugar 2 battery, an E-Ink display and one afternoon.
It sits in his pocket now. Smaller than a wallet. Runs Ollama locally. Connects to his phone through a private web interface at inkling.local. No internet required.
Task manager. File system. Custom themes. E-Ink display showing responses in real time. Battery life: 45+ minutes under load, hours on standby.
His conversations never leave the device. His prompts never hit a server. His monthly AI bill went from $20 to $0.
Cancelling the $20/month subscription saves him $240 a year. The code is on GitHub. Under $70 and one afternoon is all it takes.
HE SPENT $249 ON A BOX THAT REPLACED HIS $200/MONTH AI BILL. IT HEARS HIM AND ANSWERS OUT LOUD.
Smaller than a book. NVIDIA Jetson Orin Nano. Sits on his desk. Runs all day for $2/month in electricity.
He speaks to it. It listens. Processes everything locally. Answers out loud in seconds.
Everything stays in the room. His conversations. His data. His hardware.
He asked it who his favorite robot was. It answered correctly. Instantly. Without touching the internet.
That level of AI used to mean a $200/month subscription. Now it means a $249 box and one afternoon of setup.