AI computing demand and processing cost trends

The artificial intelligence industry is experiencing a structural shift in computing demand and a rapid decline in processing costs.

Initially, the AI boom was characterized by the massive computational requirements of model training. However, demand is transitioning toward inference—the process of using trained models to perform tasks. Estimates suggest inference accounted for one-third of AI computing power in 2023 and is expected to reach two-thirds of total demand in 2024.

Simultaneously, the cost of achieving specific AI performance levels is dropping significantly. Research from Epoch AI indicates that the cost to reach equivalent performance levels is decreasing by nearly 50% per quarter. This rate of reduction reportedly outpaces cost declines in sectors such as lithium-ion batteries and DNA sequencing. Despite these falling costs, many major brands maintain high subscription prices to protect profit margins.

As the technology matures, the labor market is reacting to these shifts. In the banking sector, AI-related job postings are projected to rise by 49% in 2026 compared to 2025. Major institutions, including JPMorgan Chase, Citigroup, and Capital One, have seen a surge in demand for expertise, with total job advertisements reaching 139,819.

Growth is particularly concentrated in the orchestration of AI agents, a skill set that has increased by 1,721%. Other technical demands are rising for LangGraph (up 679%), LlamaIndex (up 291%), and Retrieval-Augmented Generation (RAG) (up 259%). Additionally, there is increased demand for expertise in AI governance (up 394%), risk management (up 359%), and responsible AI (up 657%). Vijay Swaminathan, CEO of Draup, noted that agent orchestration is likely the “most sought-after skill on Wall Street,” specifically for engineers integrating AI into trading desks, back-office operations, and human resources.