Semiconductor AI capex boom hits dot-com bubble levels in valuations

The semiconductor sector has entered uncharted valuation territory this week. Technical analysis shows chip stocks now trading 147% above their 200-week moving averageAverage price over a defined period; smooths noise to show trend., with weekly RSIRelative Strength Index - momentum oscillator on a 0-100 scale. readings of 85.7 and monthly RSI at 84.9, matching the exact overbought extremes seen at the peak of the March 2000 dot-com bubble. The rally has been relentless, driven by AI data-center capex narratives and record ETFExchange-Traded Fund - a basket of securities trading like a single stock. inflows. Yet a striking divergence has emerged: while equity flows remain bullish into mega-cap names like Nvidia and Broadcom, the 3x leveraged inverse semiconductor ETF (SOXS) recorded its largest monthly inflow on record in April, pulling in $2.4 billion. This suggests sophisticated investors are hedging against a potential crash even as momentumThe empirical fact that winners keep winning over the medium term. traders pile in.

The underlying bull case rests on a genuine supercycle narrative. Memory chip manufacturers are guiding to record margins through 2027 as AI inference demands drive insatiable demand for DRAM and NAND. Samsung, Micron, and SK Hynix all flagged supply constraints stretching into next year, which should support pricing discipline. Intel and AMD have both benefited from data-center CPU demand, as the CPU-to-GPU ratio in AI clusters is normalising from 1:8 toward 1:1, meaning agentic AI workloads require more CPU capacity. Goldman Sachs and others note that TSM (Taiwan Semiconductor Manufacturing Company) is ramping its 3nm production hard, targeting 180,000 wafers per month by year-end, a clear sign of conviction in demand.

Yet the market structure is dangerous. Valuations of premium semiconductor names have disconnected sharply from earnings visibility beyond 2026. Chip stocks that make leveraged bets on AI infrastructure, such as Broadcom, which supplies networking gear to data centers, trade at multiples that assume perpetual capex acceleration. Meanwhile, downstream users of chips (PCs, smartphones, automotive) show no equivalent demand surge, suggesting AI chip demand may be concentrated in a narrow buyer set. If hyperscaler capex slows, or if supply constraints ease faster than expected, the rapid multiple compression could be severe. The fact that bearish hedges are hitting record flows while bullish flows continue suggests the market is pricing a high probability of a significant correction, but momentumThe empirical fact that winners keep winning over the medium term. is still carrying prices higher. This is the classic setup for a crash.