Hardware

Local AI hardware, compared

Every realistic way to run AI models locally, on three axes that actually decide the build: memory bandwidth, memory capacity, and what it costs you today, new and used. GPUs, Apple Silicon, unified-memory mini-PCs, and datacenter accelerators in one chart.

Prices updated 2026-06-04 · indicative US market figures · 59 devices

The mental model, borrowed from Ahmad Osman: local AI is capacity × bandwidth × software stack. Capacity tells you what fits, bandwidth tells you how hard the box can breathe during decode, and the stack decides how much of the spec sheet you actually cash out. Don't ask "which hardware is best?" Ask "which bottleneck am I buying?"

How to read this chart

Two questions decide a local-AI box: does the model fit, and how much speed do you get for the money. So memory runs up the side (higher = more fits) and bandwidth per dollar runs across (right = more speed for your money), with bubble size showing raw bandwidth. Up and to the right is the goal. Cheap gaming cards cluster bottom-right, great value but they run out of memory; big-memory Macs and datacenter cards sit higher yet drift left, more room but slow per dollar or priced for a server rack. Lucebox (amber) is the box you plug in at home that holds the upper right: 128 GB of unified headroom with real GPU bandwidth, tuned, for a fixed price. Tip: toggle Datacenter off to see only what you can actually buy. Click Lucebox for the build.

up = more memory (what fits)  ·  right = more bandwidth per dollar (speed for the money)  ·  bubble = raw bandwidth  ·  both axes log  ·  tap Lucebox for the build

Every device

Sort any column, or tick rows to compare a few head to head. Value is capability per dollar (memory × bandwidth ÷ price), so a box with lots of fast memory for the money scores high. Price is the one that matters: new while a part is still sold, used once it is discontinued. Apple, mini-PC and Lucebox memory is the full unified pool.

Method. Memory bandwidth is peak theoretical. Each device shows the price that actually applies: the current new price while it is still sold, the current used price once it is discontinued. Figures are indicative US-market numbers triangulated across retailers (Newegg, Amazon, B&H, Apple) and used marketplaces (eBay sold listings, resellers) as of 2026-06-04; the 2026 memory shortage has pushed new prices up and pulled several high-memory configs from sale. Value = memory × bandwidth ÷ price (capability per dollar). Datacenter parts are quote-based and sold in 8-GPU boards, so their per-unit figures are approximate. "u" marks unified memory. Numbers move fast; treat them as a snapshot, not a quote.