Starting August 23, DeepSeek adjusted its peak/off-peak billing rules, making weekends uniformly off-peak all day; Nvidia has told some of its largest customers that AI server prices will rise by more than 15%. With API pricing using market levers to guide off-peak scheduling while hardware costs keep climbing, access to computing power is shifting from simple on-demand purchasing toward more refined cost and supply management.

On August 23, DeepSeek once again adjusted its API peak/off-peak billing rules: weekends will no longer distinguish between peak and off-peak periods, and calls will be charged at the off-peak rate all day. Off-peak pricing is roughly half of peak pricing. This is an important optimization after the company introduced peak/off-peak pricing on August 17. Its core purpose is to use market-based price leverage to encourage enterprise developers to schedule large language model tasks during off-peak windows, thereby diverting pressure on peak computing capacity.

lmost simultaneously, some of Nvidia's largest customers were told that servers equipped with its AI chips will rise in price by more than 15%, with the increase taking effect on systems shipping early next year and covering the flagship Vera Rubin and Grace Blackwell series. The reason points directly to sharply higher memory chip costs: HBM and other key components remain in short supply, significantly raising the manufacturing cost of GPU servers.

Taken together, these two developments outline a new normal in the computing power market: tight supply