Generative AI's Sociotechnical Evolution: Scaling Limits, Governance Gaps, and Sustainable Pathways

This study provides a comprehensive sociotechnical analysis of the development of generative artificial intelligence (GenAI) by analysing 50 systems (2014–2023) and interviewing 25 global experts in the area. Three separate architectural epochs are identified by the research, and each is distinguished by unique scale patterns. Additionally, it demonstrates that performance peaks at 200B parameters, when a 1% increase in Fréchet Inception Distance (FID) scores corresponds to an 8× increase in processing power. There are non-linear trade-offs between increasing skills and conserving energy, according to quantitative studies. According to qualitative study, there are significant disparities in the speed at which different industries adopt new technologies. Global South nations are more affected than others (88% lack frameworks), with implementation delays of 2.3 years and governance delays of 4.2 years. A validated optimization matrix showing that new building designs can make things 3.8 times more efficient but are hard to put into practice, (1) extended scaling laws that include energy and adoption metrics, and (3) sector-specific policy tools to close the 72% policy gaps in education and the 92% accuracy-adoption paradox in healthcare. The results indicate that institutional readiness, rather than mere technical expertise, affects real-world outcomes, challenging deterministic narratives of progress. They also provide us helpful ways to develop artificial intelligence (AI) that follow the rules of Green AI.