The archive · AI & Models · Product decision · 2024–2026
Exo Labs: pool your phones and Macs into a local AI cluster — 47k stars
Exo turned everyday devices into a peer-to-peer GPU for local LLMs, hit GitHub Trending #1 and 47k stars — with no disclosed venture round.
Exo Labs
What the business is
Open-source distributed inference engine that pools everyday devices (phones, laptops, Raspberry Pis, Macs) into one local cluster for running LLMs, with an OpenAI/Claude/Ollama-compatible API.
How it started
Exo Labs was founded in March 2024 by Alex Cheema (Oxford physics, ex-trading infrastructure) and Mohamed Baioumy (Oxford-trained AI and robotics researcher) to 'democratize access to AI'. The runtime grew out of wanting to run research models on devices they already had, and hit the HN front page on 2024-07-16 with 439 points.
What happened
Exo kept landing on GitHub Trending — all-languages #1 in Jan 2025 and again 2025-12-21–23 — and passed 47k stars by Sep 2026. Cheema demonstrated a ~$5,000 cluster of four Mac Mini M4s plus a MacBook Pro M4 Max running Qwen2.5Coder-32B at 18 tokens/sec and Nemotron-70B at 8 tokens/sec, versus a $25–30k H100. The project added a macOS app, RDMA over Thunderbolt 5, and OpenAI/Claude/Ollama API compatibility, with several companies already using it for local inference.
How it ended up
Still live and pre-revenue as of 2026-09-01: under active development (last commit 2026-08-25) with 47.2k stars and no disclosed venture round.
Background
Exo is an open-source distributed inference engine that pools everyday devices — phones, laptops, Raspberry Pis and Macs — into one local cluster for running LLMs, with an OpenAI/Claude/Ollama-compatible API. Exo Labs was founded in March 2024 by Alex Cheema and Mohamed Baioumy to 'democratize access to AI', growing out of the founders' desire to run research models on hardware they already owned.
The bet was that people would run frontier-class open models on clusters of devices they already had instead of renting cloud GPUs. The project hit the HN front page on 2024-07-16 (439 points, 151 comments), repeatedly topped GitHub Trending — all-languages #1 in January 2025 and again on 2025-12-21–23 — and passed 47k stars by September 2026.
The cost math made the pitch concrete: Cheema demoed a ~$5,000 cluster of four Mac Mini M4s plus a MacBook Pro M4 Max running Qwen2.5Coder-32B at 18 tokens/sec, against a single H100 priced at $25–30k. Exo remains open source and pre-revenue, with active commits through August 2026, several companies using it for local inference, and no disclosed venture round.
What has to be true
- The bet is contrarian but grounded: pooled devices can't beat datacenter GPUs on raw power, but they win on cost, privacy and ownership.
- Open source plus a strong demo loop did the distribution: HN front page and repeated Trending #1s replaced a sales team.
- The pricing contrast was the pitch: a ~$5,000 cluster versus a $25–30k H100 made the trade visible to every reader.
- No venture round keeps the question honest: the company is still proving whether local distributed inference is a business, not just a viral project.
What can be applied
Attack the cost wall with assets people own: Exo pooled idle phones and Macs instead of asking users to buy GPUs, and open source made every star come without a sales call.
Aftermath
As of 2026-09-01, Exo remains open source with 47.2k GitHub stars (global rank #541) and active development through August 2026; the roadmap centers on RDMA-over-Thunderbolt-5 clusters for multi-Mac setups and a dashboard/API compatible with OpenAI, Claude and Ollama. The company has not disclosed funding or revenue; in November 2024 it said several companies were already using Exo for local inference and that it planned a free benchmarking website comparing hardware setups.
Sources
- Exo: Run your own AI cluster at home with everyday devices
- Exo Labs is Here! Enabling Powerful Open Source AI Models to Run Locally on Mac M4 Computers
- exo-explore/exo star history
spotted an error? The archive wants to know.
Your turn
You just read one. Describe what you are building, and see who is betting on the same thing.
Free account · 3 free questions · no card