AI’s bottleneck economy is taking shape across Southeast Asia’s chip supply chain

The artificial intelligence investment story is transitioning from its initial focus on graphics processing units (GPUs) and high-bandwidth memory to a broader, more complex phase, according to Kenanga Research. This new phase, expected to begin in 2026, will be driven by inference, AI agents, and commercial applications, rather than solely large model training and GPU-heavy infrastructure.

Inference, the process of running AI models to serve real users and tasks, requires a different scale of computational power. While training a frontier model is expensive, inference occurs continuously as chatbots answer questions, AI assistants complete tasks, and enterprises automate workflows. As AI agents become more prevalent, the volume of activity could rise significantly, potentially disrupting the current market dynamics.

Kenanga's framework for the AI ecosystem, based on Jensen Huang's "five-layer cake" (energy, chips, infrastructure, models, and applications), suggests that initial opportunities may remain in the lower layers of the supply chain, particularly where supply constraints exist. These layers include compute, storage, optical connectivity, power, and equipment.

Southeast Asia, particularly Malaysia, can capitalize on this trend by focusing on the supply chain rather than competing directly with the US or China in developing advanced AI models. By supporting infrastructure and addressing key bottlenecks, companies in the region can position themselves as contributors to the AI boom.