Grant-funded AI hardware includes servers, graphics cards, storage and network components needed to run models directly at the recipient's premises. Calls allow this kind of purchase, but they require justification of why in-house infrastructure is needed instead of the cloud – typically sensitive data that must not leave the company, the volume of processing, or a response-time requirement in production. Without this justification, evaluators often strike the expenditure out as disproportionate. Hardware is a long-term asset, so it is subject to the obligation to keep it within the project and at the place of implementation for the entire sustainability period, and it must not be sold or leased out. The rapid obsolescence of this equipment therefore needs to be factored in already at the budget-planning stage.
See also: Cloud services in a project, Depreciation of assets in the project, Project implementation location.
