AIyield: Documentation
Welcome to AIyield
AIyield: Turning idle GPU’s into yield generating RWA’s on-chain.
AIyield is an execution and orchestration layer that turns idle, high-end GPU capacity into productive AI training infrastructure — and makes participation accessible to retail without running hardware or managing operations.
AIyield aggregates verifiable GPUs, deploys them into distributed training protocols, and settles rewards and penalties on-chain. Users interact only with the economic participation layer, abstracted as on-chain GPU shares. AIyield is an official GPU supplier to the NOUS network. Through GPU shares the platform allows retail participants to participate in training NOUS models on the AIyield 8xH100 Node.
The Problem
Frontier distributed AI training requires:
scarce, expensive GPUs (H100-class)
strict uptime and performance guarantees
NOUS & others want the community to participate, but they lack the hardware. The Issue:
Computational needs from protocols like NOUS & Gensyn are higher than what the average user can provide. To actually perform Training and Finetuning at scale these Model Companies need large clusters of NVIDIA H100 Machines and above. Users who want to contribute GPU’s usually lack the performant hardware. That’s where AIyield comes into play.
Idle Capacity:
Idle GPU capacity is estimated to be $120B in NVIDIA hardware, at any given time.
AIyield is bringing these three together, Distro Labs (NOUS), retail communities & global idle GPU supply.
The AIyield Solution
AIyield bridges this gap through orchestration.
At the infrastructure level
Aggregates idle GPUs from approved suppliers (reliable, cost-efficient, h100+ supply)
Normalizes heterogeneous hardware into predictable compute units
Forms reliable, protocol-ready GPU clusters (8xh100 nodes)
At the execution level
Commits clusters to live distributed training runs
Enforces uptime, performance, and participation requirements
Manages replacement, failover, and penalties
At the settlement level
Tracks “GPU shares” per user
Settles user rewards on-chain
Abstracts protocol complexity away from users
Future: Compute outputs to be tradeable onchain (turning GPU outputs into a RWA)
GPU Shares (User Abstraction Layer)
Instead of owning hardware or managing training jobs, users book GPU shares.
Each GPU share represents:
proportional participation in a live 8xH100 GPU Node
30-day active deployment into NOUS or others.
exposure to outputs and rewards generated by that training
Maximum of 400 shares per node.
AIyield handles everything operationally:
sourcing → deployment → monitoring → execution → settlement Read - Shares for more info.
The Core Idea: Distributed AI Traditional AI training happens inside large centralized datacenters (OpenAI, Google, Anthropic).
Distributed AI breaks that model by allowing many independent GPU providers—individuals or small operators—to collectively run training jobs.
The opportunity: NOUS
**The Opportunity:
Airdrops*** Expected airdrops from some of the best funded protocols in the space. Where GPU suppliers are the key beneficiaries. These protocols together raised close to 500M in funding, with a valuation of 1B+ for NOUS.
2. AIyield Improving our infrastructure and understanding user behaviour allows us to build the foundation for the future of verified compute for distributed or regular AI training.
It allows us a first step to solve the idle GPU asset problem.
The Future of AIyield
Stage one: Become the premier destination for Community Contributed, highly performant GPU supply for Model training companies. Stage two: Allow model training in an easy fashion through our API, where we supply the GEO optimized GPU’s, efficient training/finetuning & potential community participation. Model companies, researchers, Universities take care of the Datasets + Environments and Training loops.
Stage three: Build a new financial primitive where we bring idle GPU capacity on-chain as a yield generating RWA.
Risks: Although AIyield is fully focussed on cost of running vs rewards of these protocols, the size of airdrops & point conversion towards tokens is currently unknown. Taking part in our app & contributing GPU’s is a high risk financial decision that you need to evaluate for yourself. We can’t guarantee highly profitable airdrops.
Join our Community for Questions
Discord community
Join our Discord community to post questions, get help, and share resources with over 3,000 like-minded developers.
GitHub
Our product is 100% open source and built by developers just like you. Head to our GitHub repository to learn how to submit your first PR.
Last updated