Closed-Weight Model
A proprietary AI model accessible only through a developer's private API, where the underlying weights are kept secret.
A closed-weight model is a proprietary artificial intelligence system where the creator retains exclusive control over the model's internal parameters, or weights. Users cannot download, modify, or run the model on their own infrastructure. Instead, they must interact with it through an application programming interface (API) or a web interface hosted and managed by the developer. This distribution method allows the developer to protect their intellectual property and maintain strict control over how the system is used. It also enables them to continuously update the model behind the scenes without requiring users to download new software versions. From an investment perspective, closed-weight models are evaluated to gauge a technology company's pricing power, software margins, and competitive moat. While these models can generate high-margin recurring revenue, they also require massive capital expenditure to host and run at scale. Investors must weigh the premium pricing these models command against the rising capabilities of open-weight alternatives, which can erode a developer's market share and pricing power.
Company A hosts a closed-weight model and charges enterprise clients $0.01 per 1,000 tokens processed. If a client processes 500,000,000 tokens in a month, the revenue generated for Company A is (500,000,000 / 1,000) $0.01 = $5,000, while the underlying model weights remain entirely secret.