Memory crunch splits chip stocks into haves and have-nots

The artificial intelligence buildout has created a structural bottleneck in advanced memory (HBM, GDDR6X) that is carving a sharp line between winners and losers in the semiconductor ecosystem. NVIDIA, Broadcom, and AMD continue to dominate allocation flows as they sit at the center of GPU and high-bandwidth-memory supply chains, while memory generalists and legacy players face margin pressure and inventory obsolescence. This gap is now showing up in corporate earnings and stock performance divergences that are accelerating, not moderating.

Broadcom, a key supplier of networking and memory interconnect solutions for AI data centers, saw volatility on May 13 after a bearish kicker on its chart, signaling profit-taking after an extended rally. However, the underlying demand for AI-grade connectivity and switching gear remains robust. NVIDIA benefited from headlines that CEO Jensen Huang was flying to China for talks, a tangible signal of demand continuity despite geopolitical uncertainty. AMD is caught in a transitional phase; its positioning in CPU-to-GPU migration is solid, but execution risk on high-end GPU SKUs remains.

The memory crunch has secondary effects. Lam Research and Micron Technology, whose equipment and DRAM/NAND supply support chipmakers, are seeing uneven demand: fabs upgrading to advanced nodes benefit, while commodity memory producers lag. Some analysts note that Western Digital's outperformance versus NVIDIA over the past month signals that storage and secondary infrastructure players may be capturing more value than primary chip designers if AI capex plateaus. Copper and energy costs are rising alongside chip equipment orders, pressuring margin expansion.

The key debate is whether AI capex growth is accelerating into 2026-27 or peaking. Skeptics point to the sharp underperformance of chip stocks on May 13 as a sign of momentumThe empirical fact that winners keep winning over the medium term. exhaustion rather than CPI-driven selling. They cite slowing enterprise adoption timelines and concerns that cloud hyperscalers are consolidating capex into fewer, more efficient hubs. However, insider buying at major chipmakers and announced infrastructure expansions suggest executives still see multi-year runway.