Size the factory before you build it.

Talk through your project or enter the numbers yourself. You get a first-pass design for GPUs, racks, power, cooling, three-year cost and risks, on NVIDIA and AMD platforms, with every input traced to a source. Planning estimates only.

Talk it through

Tell me what you want to build, or ask anything about AI factories. I’ll ask a few questions, and the design below updates as we go.

Your messages are sent to this site’s server and then to Anthropic’s API to be read, one turn at a time. This site does not store them, but don’t include confidential details. Every number stays editable below. Privacy and disclaimer.

What are you trying to do?

Pick a starting point, then adjust the details below.

Tune the details

Workload Model size (billion params) Concurrent fine-tune jobs GPU platform Rack power available (kW) Go-live target (months) Electricity ($/kWh) Facility PUE Your quote: all-in $ per GPU (optional) Your quote: cloud $ per GPU-hour (optional)

Your first-pass design

Updates as you change the details.

Where the numbers come from

Each input is tagged as verified against a vendor page, an estimate, or your own quote.