Goldman Sachs and Morgan Stanley Signal Shift in AI Infrastructure

Major financial institutions are identifying a structural shift in the artificial intelligence landscape, moving away from a narrow focus on model quality toward the physical constraints of computing power and system orchestration.

Goldman Sachs reported that compute has become the primary binding constraint limiting AI growth. The firm noted that demand for data centers and high-performance chips is outstripping supply as AI applications transition toward real-world deployment and inference. This shift is enabling AI-native companies to create systems of action for autonomous execution in logistics, defense, and labor automation, though full commercialization may take up to a decade.

Simultaneously, Morgan Stanley highlighted the rise of autonomous, multi-step agentic AI. The firm argues that these systems are shifting economic value from GPUs toward CPUs and memory, with CPUs serving as the control plane for orchestration. Morgan Stanley estimates that agentic AI could expand the CPU total addressable market by $32.5 billion to $60 billion by 2030, while driving up to 45 exabytes of additional DRAM demand. This diversification benefits a wider infrastructure stack, including advanced packaging, server manufacturers, and foundry capacity.