Canadians are unknowingly investing in the wrong segment of AI, and recent developments underscore this concern. The AI landscape is rapidly evolving, with a shift towards open-weight models, which publicly release the training data, as opposed to proprietary models, which are closed-weight and rented on the vendor's servers. This shift is driven by major players like Meta, Nvidia, and Cohere, who advocate for openness in AI models while maintaining proprietary control over their software and hardware. The commoditization of the model layer is evident, with open-weight models like Meta's Llama and Mistral, and China's DeepSeek and Qwen, scoring highly in benchmarks. In contrast, leading American labs like OpenAI, Anthropic, and Google DeepMind, which have not signed the open-weights letter, are under scrutiny for their closed-weight models. The U.S. government's recent actions, including export controls on Anthropic's models, further highlight the tension between open and proprietary AI. Canadians, through index funds and pension plans, are inadvertently investing in these proprietary models, which are at risk of commoditization. The key question is who owns the capacity, data, workflows, and permissions, as these aspects will determine the future of AI investments and the companies that dominate the industry.
