A company is building an editorial assistant application powered by generative AI. Current pilot usage is low, and performance is not an issue. The company cannot forecast future usage after full deployment and wants to minimize costs.

Which solution meets these requirements?