Open source models operating on the open internet will fundamentally devalue proprietary, closed models trained on open data, leading to a shift in economic value. This shift will create new opportunities in infrastructure, particularly in the development of a new kind of substrate focused on provisioning tokens per second to applications built on top of the model. Companies with their own proprietary hardware for token per second services, such as Amazon, Google, and startups like Grok, are in a prime position to capitalize on this shift. Furthermore, the combination of open source models and proprietary data will become the two main sources of economic value in the AI industry, leading to significant opportunities for the providers of infrastructure and those who have ownership of proprietary data.