Open-source AI's profits are shifting to hosting, deployment and support around free models

An analysis argues that the revenue in open-source AI lies not in the freely downloadable model weights but in the services that make them usable, such as hosting, enterprise deployment, fine-tuning and consulting. It notes that most "open" releases are really open-weight, often under custom licences with commercial thresholds, and that downloads are concentrated among a small set of models, increasingly from Chinese developers like DeepSeek and Alibaba's Qwen. The piece also points to regulatory risk, citing a US Department of Homeland Security ban on DeepSeek for federal agencies despite its permissive MIT licence. It recommends that companies test models against their own data rather than relying on benchmarks before committing to them.