Mic Neale

Every AI application today depends on an API call to a handful of centralized providers. Your code can't run without it, your costs scale with usage, and your privacy is governed by someone else's terms. What if that architecture were optional?

Mic Neale is prototyping a decentralized mesh LLM: idle GPU capacity across neighbourhoods and organisations, automatically pooled into shared inference infrastructure. When you need inference, the mesh provides it; when your machine is free, it serves others. A neighbourhood of 100 households likely has $50K–$100K of GPU capacity sitting idle—enough to dwarf the cost of API calls for the whole community.

The technical challenges are real: model sharding across heterogeneous hardware, latency, fault tolerance, fair scheduling, and defending against bad actors. But Neale's work shows they're solvable. Drawing on two decades building distributed systems and AI infrastructure at CloudBees and Red Hat—and his work on Goose, Block's open source AI coding agent—he makes the case for a mesh that's more resilient, more equitable, and harder to shut down than any centralized service.

This isn't blockchain fantasy or anti-cloud ideology. It's practical mutual aid: my spare GPU serves your inference, yours serves mine—because we're using capacity that's otherwise wasted. A glimpse of what becomes possible when we shift from "data centre required" to "your neighbourhood has enough."