Grayscale Research Chief Says Cheaper AI May Increase Compute Demand

Zach Pandl, Grayscale’s research chief, said lower-cost artificial intelligence could expand usage enough to increase demand for the computing capacity and physical infrastructure needed to run it.

In an Oct. 8 column, Pandl said falling token prices and more capable software could make additional applications and automated workflows economically viable. Competition from open-weight models is also pushing down AI costs.

Pandl distinguished between chatbots, which primarily answer questions, and AI agents, which complete multiple steps to achieve a goal. Preparing for a customer meeting, for example, can consume five to 50 times as many tokens as a single chat interaction.

Always-on agents can extend processing beyond a single conversation, potentially increasing computing demand as software works continuously toward a user’s objectives.

Even when individual tasks become more efficient, total computing demand can still rise if usage expands faster.

AI activity remains dependent on electricity, data centers and GPUs. Data centers and GPUs require time to deploy, creating a potential gap between rising software demand and available infrastructure.

Lower-cost AI can coexist with rising demand for computing capacity as usage expands.

Pandl’s conclusion was that cheaper intelligence and more capable agents could expand usage enough to offset efficiency gains and support demand for computing capacity.