In an Oct. 10, 2026, post linking to the analysis, Grayscale said lower-cost models and autonomous AI agents could amplify demand for computing power. Grayscale head of research Zach Pandl argued that expanding AI use could outweigh gains in model efficiency, while electricity, data centers and graphics processing units remain potential bottlenecks.

Grayscale also said agents, OpenAI’s "dots," and new lower-cost models could further increase compute demand. Pandl said that could create additional opportunities for owners of scarce computing resources.

AI agents increase token consumption

According to Pandl, competition from cheaper open-weight models is reducing the cost of tokens, the units of text that AI systems process and generate. Anthropic, for example, says its latest Sonnet model costs up to 30% less to complete a single task.

At the same time, autonomous agents that work continuously toward a user’s goals could sharply increase computing volumes. Unlike chatbots that answer individual questions, agents perform sequences of actions, including analyzing data, processing documents and preparing outputs.

Grayscale estimates that these workflows can consume five to 50 times more tokens than a typical chatbot interaction, depending on the task.

Data from OpenAI, OpenRouter and Similarweb, published in an analytical report by Andreessen Horowitz’s crypto division, a16z, also indicated higher consumption by agents. According to the report, AI agents use nearly five times more tokens than humans, while their token consumption has increased about 14-fold since February 2026.

The report said the most active companies generate more than eight times as many tokens as typical businesses. Rising agent use is also affecting demand for memory and traditional automation tools.

Compute demand could outpace efficiency gains

Pandl cited a Goldman Sachs forecast that token consumption could increase 24-fold by 2030, largely because of the spread of AI agents. He argued that overall demand for computing power would continue to rise if AI use expands faster than improvements in model efficiency.

Digital AI services depend on physical infrastructure, including electricity, data centers and GPUs, all of which take time to supply. Pandl pointed to CoreWeave, which reported signing new contracts at higher prices for computing capacity.

In Pandl’s view, cheaper AI and the development of autonomous agents will broaden the range of potential uses and could intensify competition for limited computing resources. He said this could benefit companies that own the relevant infrastructure.

Meta CEO Mark Zuckerberg has forecast that billions of people will use personal AI agents over the next five years.

Source: Incrypted