Crypto glossary
Decentralized compute (GPU networks)
Decentralized compute networks are marketplaces where independent providers rent out GPU and CPU capacity, coordinated and often paid through a blockchain.
Providers such as small data centres, former mining operations or individuals with spare graphics cards register their hardware with the network. Customers submit jobs, from 3D rendering to AI model training and inference, and the protocol matches them with providers, tracks the work and settles payment, usually in a native token or stablecoins. Token rewards often subsidise supply in the early stages.
These networks sit within the DePIN trend and grew with demand for AI hardware, when access to high-end GPUs through big cloud providers became expensive and scarce. Some former bitcoin and ether miners repurposed their equipment or facilities toward this market, and some AI startups use decentralized capacity for batch jobs that tolerate variable performance.
The model faces practical limits. Training large models requires tightly connected clusters with fast interconnects, which scattered consumer GPUs cannot match, so much real demand is for inference and smaller jobs. Verifying that a provider ran the job correctly is hard, privacy for sensitive data is limited without secure hardware, and usage figures inflated by token incentives can hide weak paying demand.
Example
An AI startup rents 64 GPUs from a decentralized compute marketplace to generate image embeddings overnight, paying in USDC at a lower hourly rate than a large cloud provider would charge.