AI models are becoming outdated at record speed — so what does that mean for the tokens that companies are spending millions on?

New data from Vercel’s AI Gateway reveals that AI models are becoming obsolete faster than ever as newer, more capable versions hit the market. Most enterprise token spending is now directed toward models released within the last three months. A significant shift is occurring as open-weight models gain dominance over closed, black-box systems, capturing 56% of token volume while accounting for only 14% of total expenditure. This trend is driving down the average cost per token for the third consecutive month. As companies increasingly opt for self-hosted or affordable cloud-based open-weight alternatives, the industry faces a growing debate regarding the sustainability of rapid AI development. The constant churn of models forces businesses to reconsider their long-term infrastructure investments, as the rapid pace of innovation threatens to render current token commitments to legacy models redundant almost immediately after purchase.