“Commoditization” is the thesis that the base layer of AI - the large foundation models themselves - is becoming an interchangeable commodity rather than a durable source of advantage. The argument rests on two trends documented in primary sources: capable open-weight models catching up to the closed leaders, and the per-token price of using a model collapsing.
The catch-up showed clearly in July 2024, when Meta released Llama 3.1. Meta described its 405-billion-parameter model as “the first openly available model that rivals the top AI models” and as having capabilities that “rival the best closed source models.” When a freely downloadable model approaches the frontier, the premium a closed lab can charge for comparable quality erodes. Mark Zuckerberg framed it explicitly as a strategy to keep AI from being “concentrated in the hands of a small few.”
The price collapse is the second force. The a16z “LLMflation” analysis found that the cost of inference for a model of equivalent quality was falling roughly 10x per year - a thousand-fold over three years for the GPT-3 quality tier. When the same capability gets dramatically cheaper every year and is available from multiple providers, the model itself behaves like a commodity input, and value migrates to whatever sits around it: data, distribution, workflow integration, and the application layer.
Why a business reader should care: if foundation models commoditize, the economics of the AI industry change shape. Spending hundreds of billions to train a frontier model is hard to justify if a comparable open model arrives months later for free and prices fall every year. The thesis is contested - the frontier keeps moving, and the best models still command a premium - but it is one of the strongest counterarguments to the assumption that scale alone guarantees a defensible business.