Fetching from the wire…
01
02
03
04
05
06
07
08
Infra2026-07-12 · source-backed
Posted July 11 (293 points on HN), it distributes inference across a network built on iroh, using hole-punching to route requests directly between nodes with no central server (iroh). Models get partitioned by layer ranges, layers 0-15 on one machine, 16-31 on the next, so several modest GPUs run a model none could hold alone. For home labs and small teams hitting VRAM ceilings, this is the most interesting local-inference idea I've seen in a while. Whether the network latency stays acceptable across the hops is the open question.
Each link below shares sources, entities, or timing with this story.
Shared entities / Shared topic / Earlier coverage / Downstream implication
Both cover GPUs, VRAM; overlapping topics (gpus, model); earlier GPUs coverage from 2026-04-10.
Shared entity: GPUs / Shared topic / Earlier coverage / Tension
Both cover GPUs; overlapping topics (could, gpus); earlier GPUs coverage from 2026-06-26.
Both cover GPUs; overlapping topics (gpus, model); earlier GPUs coverage from 2026-05-09.
Shared entity: GPUs / Shared topic / Earlier coverage
Both cover GPUs; overlapping topics (directly, gpus, hitting); earlier GPUs coverage from 2026-05-11.
Both cover GPUs; overlapping topics (consumer, gpus, layer); earlier GPUs coverage from 2026-03-18.
Both cover GPUs; overlapping topics (alone, gpus, model); earlier GPUs coverage from 2026-03-08.
Both cover GPUs; overlapping topics (consumer, directly, gpus); earlier GPUs coverage from 2026-02-28.
Shared entity: VRAM / Shared topic / Earlier coverage
Both cover VRAM; overlapping topics (alone, layer, model); earlier VRAM coverage from 2026-02-23.