Morgan Stanley: Open-weight models reduce AI costs, but the "Jevons Paradox" may drive continuous growth in computing power demand

According to Odaily, Morgan Stanley pointed out in its latest report “Open-Weight Models and Three Future Scenarios” that open-weight models do not necessarily reduce AI computing power demand. On the contrary, lower usage costs might accelerate AI adoption, resulting in the classic “Jevons paradox”: as the cost per inference decreases, businesses will apply AI to more tasks, ultimately driving higher demand for tokens, computing power, electricity, and infrastructure. The report emphasizes that open-weight models are not completely free—companies still bear costs related to GPUs, cloud services, maintenance, and security, and the actual economics depend on the application scenario. Morgan Stanley believes that regardless of changes in model openness, companies such as NVIDIA are likely to benefit.