Open-Source AI Could Shift Profits From Model Labs to Compute Providers

Investor Gavin Baker argues that open-source AI does not reduce the overall demand for compute—it simply shifts more of the economics from the model layer to the infrastructure layer.
Cheaper tokens with comparable intelligence should lead to significantly higher consumption. And whether a token comes from an open-source Chinese model or a closed frontier model like Anthropic, it still requires compute, energy, data-center capacity, capex, and operating expenses to generate.
That creates an attractive risk-reward opportunity for neoclouds, Cerebras, $NVDA, and other providers of AI infrastructure.
Baker expects frontier models to continue capturing a large share of the economic value, while cheaper open-source models process most of the actual token volume.
The likely future is a mix: enterprises use a powerful frontier model as the “conductor,” supported by numerous cheaper open-source models fine-tuned on their own private data.
Open source may pressure AI lab margins—but it could dramatically expand demand for the infrastructure underneath them.