Semiconductor Valuations Hit Dot-Com Bubble Levels on AI Capex

Semiconductor stocks, particularly memory makers, have entered rare air. The SOX index (Semiconductors vs InflationThe rate at which prices rise across an economy. ratio) has reached the same elevated territory as the 1999-2000 internet bubble peak. Memory chip prices are soaring on AI data-center demand, with Micron, Samsung, and other producers guiding to margin expansion through 2027. Goldman Sachs data shows that South Korea's EWY ETFExchange-Traded Fund - a basket of securities trading like a single stock. now accounts for roughly 14 percent of global equity ETF volume, an unprecedented concentration that mirrors past bubble episodes.

The narrative rests on genuine supply constraints. AI memory demand is still in early innings, and production bottlenecks are expected to stretch for years. Major foundries like TSMC are ramping 3-nanometre capacity aggressively, targeting 180,000 wafers per month by year-end. NVDA, AMD, SNDK, and MU have all benefited from enthusiastic earnings and guidanceCompany-issued forecasts of future financial performance. revisions. Yet the speed and magnitude of the move have bred a secondary risk: retail and institutional flows chasing the trend rather than anchored fundamentals.

The risk skews toward mean reversion. While the AI supercycle thesis holds water, valuations have compressed all uncertainty premium. A single miss on memory pricing, or evidence that customers are over-building inventory, could trigger violent deleveraging. Inverse ETFs like SOXS have seen renewed interest from traders betting on a pullback. Hardware makers face circular investment risk: NVDA, CoreWeave, Iren, and others are all building supply to feed the AI boom, but excess capacity could trigger a race to the bottom on chip pricing.

Contrarians point to broader market breadth concerns. Mega-cap software and premium SaaS names like EPAM, CSGP, PODD, and BSX have been repriced hard year-to-date, suggesting institutional capital is rotating, not expanding. If semis are truly a secular bet on AI infrastructure, why are software and enterprise tools rolling over? The divergence hints at exhaustion, not healthy rebalancing.