AI costs set to rise as DRAM, licensing and token use increase

Users and developers will likely pay more for AI this year as rising hardware costs, new content licensing and growing token consumption push prices up, according to the report. AI runs on a token economy in which both input and output are billed as tokens, and the highest immediate cost for providers is DRAM memory used to hold tokens.

A supply crunch is pushing DRAM prices about 20% year over year, and high-bandwidth memory (HBM) and NAND flash are also rising, Micron CEO Sanjay Mehrotra told Wall Street analysts, saying the gap between demand and supply is "really the highest that we have ever seen." Providers also face pressure to monetize long-term investments.

The piece notes OpenAI raised a per-token developer price from $1.25 to $1.75 between models, and content-licensing deals are emerging—OpenAI recently announced a deal with Disney to license more than 200 characters, a pact that included a reported billion-dollar Disney stake in OpenAI while royalty terms were not disclosed.

The report warns token counts are increasing as models become more verbose, users submit longer inputs, and enterprises deploy inference and agentic workflows. A ByteDance paper cited in the article says agentic interactions can drive token consumption up faster than linearly, potentially escalating computational and financial costs as conversations and tool calls are fed back as context.

Key Topics

Tech, Token Economy, Dram, Hbm, Nand Flash, Openai