Open-weight AI: choosing control as well as a model

Open AI is not just an ideological choice. For people building business software, it is also about control over the infrastructure their product depends on.

In the Open Weights and American AI Leadership letter dated 24 July 2026, companies including NVIDIA, Microsoft, Meta, IBM, Mistral, and Hugging Face support open weights and call for avoiding premature restrictions on their distribution.

The original letter on open weights

Operational convenience and dependency

With a model available only through an API, the provider handles execution. This is convenient for the team integrating the service: there is no model infrastructure to administer. In exchange, the product depends on that service’s terms, pricing, and availability.

With downloadable weights, where the license permits it, a business can run the model on its own infrastructure and adapt it to its context. This offers more choice over where data is processed and how the service is organized.

Having the weights does not remove the work

Hardware costs money, maintenance costs money, and someone has to handle updates, access, and monitoring. Running a model internally does not automatically make the whole system secure: the application, logs, connected tools, and operating procedures matter too.

Open weight and open source are not synonyms. The Open Source Initiative’s definition also considers code, data information, and freedoms to use and modify; being able to download the weights alone does not describe all these conditions.

The Open Source AI Definition

My expectation: hybrid architectures

I expect many applications to combine managed services with internally hosted models, choosing according to the task. This is an architectural hypothesis: it needs testing against total cost, required quality, and available expertise.

The useful question is not just which model is best today. It is how much control we want to retain over the product when conditions change tomorrow.

Adapted from one of my LinkedIn posts and added to the blog on 10 September 2026. The date shown refers to publication on this website.