A discoverable offering
Help builders understand the models and policies you provide.
For inference providers
Your inference service.Connected to more builders.
Make supported models and inference capacity available through the ShareAI marketplace, where applications discover providers and route AI requests.
Your infrastructure. Your role in the network.
Applications
Models & providers
Inference requests
ShareAIRoute · meter · bill
YOUR CONTRIBUTIONConnected usage & payments
01
ConnectBring eligible infrastructure
02
ConfigureSelect supported offerings
03
ServeContribute to routed inference
04
TrackReview your Provider rewards
The part worth solving
An endpoint still needs distribution.
Serving a model is one task. Reaching the applications that need it and managing ongoing usage is another.
A clearer model
Meet applications where they choose models.
Use the Provider workspace to configure your offering and explore how supported models appear in the public catalog.
Where it fits
A contribution that connects.
A common integration path
Participate in the same routing layer used by applications.
Provider economics
Your earnings come from eligible inference capacity, distinct from Builder margins.
Put your GPU to work
What could your capacity look like?
Choose your GPU, set your rates and find out what a fully booked schedule could be worth.
RTX 3080-based estimateAbout 140 output tokens per second, per GPU.
Up to 24
Model, speed & job load
Reference modelLlama 2 7B · Q4_0
Short requests: 512 input tokens and 128 output tokens. Each GPU handles one request at a time.
Have your own benchmark? Replace both speeds. Selecting another GPU restores its reference values.
100% means requests arrive continuously throughout your selected hours. Lower it to account for idle time.
Published speeds come from community llama.cpp tests with Flash Attention off. The RTX 3080 Ti preset uses an RTX 3080 reference; it is an approximation, not a measured 3080 Ti result.
Input processing and output generation share the same time budget. Real throughput also depends on context, software and request overhead. A month is 30 days, with your weekly schedule repeated.
View benchmark sources$54.28per 30-day month
At 100% job load
Gross estimate, before electricity, fees and taxes.
310.2M input tokens$31.02
77.5M output tokens$23.26
Your weekly schedule40 GPU-hours
Your planned availabilityRest of the day
5 days selected1 GPU × 8 hours × 5 days
This is a synthetic scenario, not a payout offer. Actual serving and Rewards depend on eligibility, demand and program terms. Exchange and Mission have their own incentives.
How is a Builder different from a Provider?
A Builder monetizes AI traffic from an application they own. A Provider contributes eligible compute to serve the network. Their earnings come from different contributions.
Where should I start?
Open the Provider workspace to review the requirements and configure your offering. Use the public catalog to understand how models and provider policies are presented.
Does ShareAI build or host my application?
Your application is built and hosted by your team. ShareAI supplies the routing, usage and payment layer for inference routed through the network.
Connect your contribution