Memory Chip Shortage Widens AI Capex Gap

The artificial intelligence infrastructure race is driving a deepening memory-chip shortage, with global supply constraints now dictating corporate profitability and equity valuations. Companies that have secured long-term memory contracts and foundry capacity are pulling ahead in AI capex cycles, while competitors lacking these allocations are getting squeezed. Bloomberg reported that the worsening shortage is driving a widening gulf in corporate results and stock performance.

NVDA and Broadcom (AVGO) have benefited from their upstream positioning and strong supplier relationships. NVDA jumped on news that CEO Jensen Huang is traveling to China, signaling potential deals and validating the Street's thesis on China-US tech engagement. Western Digital, however, has outperformed NVDA by 3x over the past month, suggesting that memory-focused suppliers are gaining relative to AI accelerator leads. Arista Networks and Lam Research, which enable chip production and networking, are also in focus as memory capacity becomes the bottleneck.

The memory crunch is not uniform: HBM (high-bandwidth memory) for AI chips is the most constrained, while DRAM and NAND face milder pressure. This creates a two-tier market where HBM leaders (Samsung, SK Hynix, Micron) command premium multiples while legacy suppliers scramble. AMD, which depends on external memory suppliers, faces higher costs and longer lead times than NVDA, which has more vertical control.

The debate centers on whether memory shortages are temporary (resolved by mid-2026 as new fabs ramp) or structural (persist as AI demand keeps outpacing supply). If temporary, the memory-focused winners will underperform once supply normalizes. If structural, memory names stay outperformers. Recent earnings misses from some memory suppliers suggest supply is catching up, but the AI capex cycle's durationBond price sensitivity to interest rate changes. remains uncertain.