AI chip shortage deepens; memory crisis splits tech winners and losers

The memory chip shortage is not easing; instead, it is accelerating a bifurcation in the semiconductor sector. Advanced AI accelerators and high-bandwidth memory (HBM) components are in acute shortage, with foundries like TSMC rationing supply to the highest-bidder customers. Meanwhile, commodity DRAM and NAND suppliers are operating under severe margin compression as demand cools for legacy devices. This divergence is feeding directly into earnings surprises and stock performance gaps across the industry.

Companies positioned in AI inference and training infrastructure are winning market share and pricing power. NVIDIA remains the linchpin of this ecosystem, but peers like Broadcom and custom-silicon specialists are also capturing upside. Conversely, traditional PC and smartphone chipmakers are losing relevance as OEMs shift R&D budgets toward AI-ready architectures. Intel and other legacy players are struggling to regain process leadership and compete for AI workloads.

The shortage is also creating second-order effects. Utilities and data-center operators are booking massive capex commitments to expand power and cooling infrastructure for AI clusters, driving orders to companies like American Electric Power and infrastructure suppliers. However, this capex acceleration is itself constrained by chip lead times and supply uncertainty, creating a feedback loop of demand volatility and pricing power concentration among leaders.

Bears argue that memory utilization rates will eventually normalize as supply catches up to demand and competition intensifies. If that occurs, margin compression could sweep across the sector, and overvalued chip stocks could face significant repricing. However, the current momentumThe empirical fact that winners keep winning over the medium term. favors AI-native architectures and suppliers with locked-in customer relationships and long-term contracts.