When will the AI boom end? Nobody knows, and the market doesn’t want to think about it

The SOX semiconductor index surged 47% in April as investors increasingly bet that AI will drive sustained chip demand, with AI-linked stocks now representing about 45% of the U.S. market. The article highlights massive planned AI capex, including a projected US$725 billion this year from Alphabet, Amazon, Microsoft and Meta, but warns that profitability remains unclear. It also flags a possible supply-chain boost from Strait of Hormuz disruptions, which could tighten inputs like helium and naphtha and further support chip prices.

Analysis

The market is treating AI as a single trade, but the economics are likely to bifurcate sharply between infrastructure sellers and capital allocators. Chip exposure is the cleaner near-term expression because it monetizes every dollar of incremental AI capex regardless of eventual end-market winners; however, that also makes semis the most crowded and most reflexive part of the stack. If hyperscaler spending merely slows from “hypergrowth” to “high growth,” valuation compression can hit chips before any true demand collapse shows up in reported revenue.

The more interesting second-order beneficiary is not necessarily the GPU incumbent, but the cheapest incremental compute path. If inference shifts toward CPU-heavy architectures, the market may be underestimating the installed-base and software ecosystem advantages of legacy x86 vendors versus the headline AI leaders. That creates a potential relative-value setup where low-expectation CPU names outperform on mix shift, while premium AI hardware names remain vulnerable to any evidence that model efficiency is improving faster than demand is scaling.

A separate risk channel is supply-chain geopolitics. Even without direct manufacturing exposure to the Gulf, any extended disruption can squeeze inputs and logistics just as inventories are being normalized, which would amplify pricing power for a few large foundries and component suppliers while forcing downstream customers to ration builds. But the bigger medium-term risk is that AI economics disappoint: if monetization stays weak, hyperscalers may be forced into a capex pause, which would hit semis with a 1-2 quarter lag and likely trigger a sharper de-rating than the initial slowdown implies.

Consensus is still assuming that spend intensity equals durable profit pools; that is the missing link. In practice, the first phase of AI usually rewards the pick-and-shovel layer, while the second phase decides whether the ultimate winners are the platform owners, the application layer, or no one at all. The current move looks directionally justified, but the asymmetry has worsened: upside from continued capex acceleration is probably smaller than downside from even a modest budget reset.

Key Decisions for Investors

• Long TSM / short a basket of high-multiple AI infra names over 1-2 months: favor the foundry with direct volume linkage and stronger pricing power, while fading names whose multiples already discount flawless capex growth. Use a 1:1 notional pair and take profit if the spread narrows 10-15%.
• Initiate a tactical long GOOGL or AMZN vs short NVDA pair into the next hyperscaler capex commentary window: if capex growth decelerates, the market should punish the pure hardware multiple first. Risk/reward is favorable because the shorts carry higher sensitivity to spend normalization.
• Buy AVGO on a pullback, but hedge with short-dated SOX puts: AVGO has more diversified cash flow and can absorb a slowdown better than pure-play chip beta. The hedge limits drawdown if the market abruptly reprices AI enthusiasm.
• Watch Intel as a tactical relative-value long only if CPU adoption headlines persist for another 2-4 weeks; otherwise avoid chasing. The setup is better as a momentum trade than a long-term fundamental call, with defined stop-loss on any reversal in inference architecture commentary.
• For event risk around geopolitics, consider a short-dated long-vol structure on TSM or SOXX rather than outright directional longs. It captures supply-chain shock optionality while limiting exposure if AI sentiment cools before any real shortage emerges.