A small team. Macs across the world. Async AI jobs.
Acorn dispatches AI work across a vetted fleet of idle Apple Silicon. We started because per-token cloud pricing did not match the bursty shape of real inference workloads — and most of the world Apple Silicon was sitting unused at night.
Why Apple Silicon
M-series chips have the unified-memory architecture that makes Whisper-large fit in 16 GB. They are plentiful: tens of millions ship every year, mostly to people who do not use them at night. The hardware is already paid for; we just have to dispatch.
Why per-job
Async inference is bursty — a few thousand jobs at 9am, nothing at lunch, a backfill overnight. Renting an H100 by the hour to serve that shape is how the bill gets large. Per-job pricing maps cost to actual work.
What we are building
Transcription went GA in April. Embeddings, image generation, and async text generation are in private beta. The shape of every route is OpenAI-compatible by design — you can switch your existing client with two lines of diff.
Who is on the network
Hosts opt in to specific job categories, run a hardened sandbox, and get paid weekly. Developers get the API. No data-center lease, no quarterly GPU procurement, no idle bills.
Talk to us.
Email hello@acorncompute.com for partnerships, pilot access, or anything else. We read every email.