Biography
Dr. Pallavi Singh Yadav is a faculty member of Shaheed Bhagat Singh College, University of Delhi, was previously taught in Shivaji College, University of Delhi, and research scholar worked in the field of Non- Performing Assets of the Indian Banking Industry. Simultaneously working on other research work relating to the field of accountancy, technology and business statistics. She is appointed as Global Assistant General Secretary of International Council of Management Research (ICMR) by the international Research Council of Eudoxia Research Centre. She is also an Editorial Board Member of New Swertia Publications, Advisory Board Member of Seminars, Conferences & Listed Journals in IGNITED MINDS PROFESSIONAL & ACADEMIC RESEARCH CONSORTIUM (IMPACT). She earned her UG (B.Com.), PG (M.Com. - accountancy and business statistics), M.Phil (Commerce) from Maharaja Ganga Singh University, Bikaner (A State University). She is two times UGC-NET qualified (December 2015 and July 2018).
Conference contributions 1
THE ECONOMICS OF INFERENCE – HOW LARGE LANGUAGE MODEL COST STRUCTURES ARE RESHAPING SOFTWARE BUSINESS MODELS
This paper examines how the introduction of LLMs (Large Language Models) into commercial software is driving a structural transformation of work, management practices and economic value capture across the software industry, a transformation with consequences for professions, institutional adaptation and digital labour markets. For nearly two decades, SaaS (Software as a service) business models rested on a single load-bearing assumption that the marginal cost of serving an additional user was effectively zero. This assumption justified flat rate subscriptions, i.e. all you can eat pricing tiers and the 75-85 percent gross margins that defined cloud software economics through the 2010s. The integration of LLMs (Large Language Models) into commercial software products has broken that assumption. Every model inference now consumes real, metered compute and the cost scales with prompt length, response length, model size and request frequency. We document this structural shift, develop a quantitative framework, the inference Cost Propagation Framework for understanding how token-level prices set by foundation model providers cascade through application layer business models and analyze the strategic responses now visible in industry data. Drawing on May 2026 pricing from OpenAI and Anthropic, public company disclosures and the ICONIQ Capital 2026 State of AI survey, we show that average AI product gross margins have settled near 52 percent against a SaaS benchmark of 75-85 percent, with inference alone consuming roughly 23 percent of revenue at scaling stage firms. The paper closes by identifying which firms, professions and institutional arrangements are most exposed to these new economics and what governance and management responses are emerging