The hidden bill behind agentic AI — and where to run it to maintain control (www.businessinsider.com)

A recent analysis highlights the financial implications of agentic AI, particularly around token consumption. Agentic AI models require significantly more tokens than traditional systems, with a single complex task consuming up to a million tokens. This new "tokenomics" has raised alarms in both academic and financial circles, as Goldman Sachs predicts total token consumption could soar to 120 quadrillion tokens per month by 2030. The shift to a consumption-based model in enterprise AI signifies a departure from fixed software costs to variable infrastructure expenses, prompting organizations to rethink their budgeting and resource allocation strategies. Dell Technologies has responded to this evolving landscape with its Deskside Agentic AI, enabling local execution of AI models on existing hardware, thus minimizing costs associated with token consumption. This solution allows businesses to process sensitive workloads internally, ensuring compliance with privacy regulations while optimizing resource management. Analyses suggest that employing this system could lead to an impressive 87% reduction in token spending over two years compared to public-cloud alternatives. As the conversation around AI costs transitions from an IT focus to a broader company-wide concern, businesses are urged to implement governance early and strategically manage their agentic AI workloads to maintain control over expenses.