AI-rewritten: This is a summary of an article from Tom’s Hardware, rewritten by AI (Qwen, running locally) to make it easier to read. The facts come from the original article – read it for the full story.

---

Futurum Group CEO Daniel Newman stated that AI agents currently use five times more tokens than humans do. His data, based on an Andreessen Horowitz chart of OpenRouter figures, shows agents consumed 7.3 trillion tokens compared to humans’ 1.4 trillion as of August. This marks six months after agent usage first surpassed human usage. However, the article notes that these agents are mostly rereading content they have already seen. More than 85% of agent tokens come from cached prompts, according to a16z citing OpenRouter.

OpenRouter is described as a leading AI model gateway and routing platform. Its head of insights, Peter Walker, provided a chart showing that since the February crossover point, agents are using 14x more tokens while human usage has increased 2.8x. The platform sorts API keys into categories like agentic, mixed, or human using a weighted composite score based on factors such as tool call rate and turn count. While agent use is growing elsewhere, with McKinsey reporting that 40% of large organization respondents scaled AI agents in their 2026 survey, the data comes from only one platform and shows some dips in April and July.

The article explains that cached tokens account for nearly all of the relative growth in token usage because they cost far less than processing a prompt from scratch. However, these tokens must still be held in memory within what is called the KV cache. This storage requirement ties into rising demand for high-bandwidth memory (HBM), where models keep stored context. Industry reports suggest RAM and storage shortages will worsen in 2027 and 2028, with customers paying more than this year as memory makers prioritize AI data centers.