Next development / Inference network
In developmentCANUS is developing one inference service across independent AI clouds. AI companies would buy the service from CANUS; participating clouds would supply agreed capacity and keep their existing customers. As demand grows, the next build can enter the CANUS programme.
In development. Expressions of interest only; the inference service is not yet available.
In ordinary words
Training teaches a model. It happens in long runs, in one place, before anyone uses the product.
Inference is the model doing its job: every question answered, every reply written, every image read. It happens each time a customer uses the product, all day, wherever those customers are.
Training
Teaching the model.
Long runs on a large cluster, in one place.
Once, then again when the model changes
Inference
The model doing its job.
Short bursts on GPUs that are running, reachable and near the people using them.
Every use, everywhere, all day
Inference needs GPUs that are running, reachable and close to the people using them.That is the demand the network would connect to independent clouds.
How it would work
AI companies would deal with one service. Independent clouds would supply it. CANUS would sit between them, holding the agreement and keeping every responsibility on one record. Where demand outgrows the capacity supplied, the next build enters the CANUS programme.
Demand committed
Network · Demand committed
Hover a step to hold itTap a step to hold it
Demand, service delivery and expansion, connected. Every party keeps its own contracts and its own decisions.
Admission to the network
Our planned network would admit GPU capacity only after independent verification and testing for the agreed AI workload. CANUS would then bring that capacity into one inference service for AI companies.
Who supplies what
AI companies
What it keeps
Keeps its models, its data and its customers.
CANUS
What it keeps
Owns no GPUs and operates no cloud.
Independent clouds
What it keeps
Keeps its infrastructure, its existing customers and its own commercial terms.
Why both sides
For AI companies
Buy inference once.Run it in many places.
For independent clouds
Serve new demand.Keep your cloud.
Participation would be agreed cloud by cloud, with capacity, locations and terms set out in writing before anything is supplied.
From demand to the next build
Today an independent cloud brings its next project into the CANUS programme with the demand it has found itself. The network would add a second source: inference demand CANUS has already agreed with customers. Where that demand outgrows the capacity in the network, the next build has a customer case before it starts.
Commitments accumulate
The gap becomes visible
The next build enters the programme
New capacity joins the network
Customer commitments can support plans for more capacity. Any financing remains subject to the lender’s approval and terms.
One service. Clear responsibility.
We’re developing one CANUS inference service across participating AI clouds. Instead of managing several cloud relationships, an AI company would come to CANUS for the service, support and billing.
The clouds supply the agreed GPU capacity. CANUS manages the service. A qualified external firm would independently check the supplied capacity against agreed requirements.
In development. External verification arrangements are not yet in place.
Where this stands
Questions
For the proposed inference network, a qualified external firm would carry out the independent checks and issue its own findings. CANUS would remain responsible for its service. This keeps the seller and the independent checker separate. The external firm has not yet been appointed.
Expressions of interest
Two short paths. Choose the one that fits and say what you would bring. A named person replies; nothing is committed by writing.
Choose your path
Tell us roughly what you would run, where your users are and when you would need capacity.
Expressions of interest only. The inference network is in development; submitting does not reserve capacity.
The inference network
One service for AI companies. Agreed capacity from independent clouds. A route for the next build through the CANUS programme. In development, and taking expressions of interest from both sides.
In development. Expressions of interest only; the inference service is not yet available.