What If AI Was Powered by Regular People?

Summary

• c0mpute launched on June 4 as a decentralized AI network built around idle GPU power, private inference, and the ZERO token.

• The project is developing Shard, a distributed inference engine designed to run large AI models across consumer GPUs in different locations rather than centralized data centers.

• The developer behind c0mpute recently demonstrated large scale AI inference across consumer GPUs in different locations, a key step toward the project's decentralized network vision.

Read the full story

c0mpute launched on June 4 as a decentralized AI network that lets users contribute idle GPU power to run AI inference tasks. The project focuses on private and censorship resistant AI access, while compute providers are rewarded through the ZERO token.

On June 15, c0mpute shared an article explaining how it plans to run large AI models across multiple consumer GPUs instead of depending on centralized data centers. The system splits models across distributed hardware and uses speculative decoding to improve performance, with the goal of making private and uncensored AI inference available through a decentralized network.

Leyten, the developer behind c0mpute, later shared that the project’s sharded inference engine had reached 40 tokens per second while running GPT OSS 120B across three RTX 4090 GPUs. The result was up from roughly 25 tokens per second the previous day, and the GPUs were running from different locations rather than one data center. Leyten said this will become the base for a permissionless version of c0mpute’s decentralized inference network.

On June 19, the team published another article describing c0mpute’s move from an AI app into a decentralized AI network. The post highlighted Shard as the core engine, with ZERO serving as the network token. Inference fees are intended to support buybacks, burns, rewards for GPU providers, and staking incentives as more developers build on top of the ecosystem.

Beaver, a well known crypto personality and winner of X's $1M Article Contest, also weighed in on c0mpute. He said that if the project achieves its goals, it could change the industry by enabling fully decentralized AI.

Most recently, Virtuals Protocol announced that c0mpute is coming to its ecosystem. The team said decentralized AI needs agents running on top of it, while Virtuals brings one of the largest agent ecosystems in crypto. The integration is intended to connect c0mpute's decentralized compute infrastructure with AI agents built within the Virtuals ecosystem.

As the AI industry becomes more dependent on large centralized data centers, c0mpute is exploring a decentralized model where inference is powered by distributed consumer hardware.