Economies of Scale have been such a huge influence on the last ~by 300 years of industrial change. In 1725 people sat at home hand crafting things to sell to neighbors. In 1825, capitalists were able to open factories that hired people. By 1925, those products were built in a massive factory that employed the entire town and were shipped all over the country on rail roads. By 2025 factories might be partially automated while hiring tens of thousands of people, cost billions to build and the products get distributed globally. The same trend applies to knowledge work as well, despite the rise of personal computing.

Companies are spending hundreds of millions of dollars on training AI models, why wouldn’t they expect to own the reward from that investment? These models are best run on $100k+ fleets of power hungry, interconnected GPUs, just like factory equipment vs a hand loom.

Open weight models are a political and marketing tool. They’re not being opened up out of kindness, or because “data wants to be free”. AI firms open models to try and destabilize American companies by “dumping”, and AI firms open models as a way to incentivize companies who don’t like closed-source models to buy their hosted services.