AI capex supports U.S. growth as manufacturing softens

That matters because the AI capex cycle is no longer a niche theme — it is a macro variable. Hyperscalers and platform companies are still pouring money into servers, networking gear, data centers and power infrastructure, and that spending is rippling through the industrial base. Microsoft has warned in filings that demand for cloud and AI products is hard to forecast and that supply constraints in semiconductors, networking equipment and power systems can persist. Amazon has described continued investment in technology and infrastructure as essential to scale. Alphabet and Meta are making similarly large commitments to technical infrastructure. For suppliers, that is a multi-year demand engine; for the broader economy, it is one of the few sources of real incremental growth.

Investors should care because this is exactly the kind of environment where the market can misread the cycle. A technical recession — usually defined by two consecutive quarters of contraction — would not necessarily come with mass layoffs if AI spending remains intact and the unemployment rate stays near 4.1%. That makes the downside more about earnings dispersion than a broad market crash. The beneficiaries are the shovel sellers: semiconductor equipment, chipmakers, networking, power, cooling and data-center infrastructure. The laggards are the cyclicals that depend on broad-based factory demand, where industrial production at just 102.64 still looks more like recovery than acceleration.

The recent move in chip stocks reinforces that point. The VanEck Semiconductor ETF, SMH, has traded as high as 655.89 and closed most recently at 587.82, while its 200-day moving average sits far lower at 461.82. That tells you the AI trade has been powerful even after a sharp pullback from overbought levels earlier this year. The iShares Semiconductor ETF, SOXX, has been equally volatile, dropping to 465.00 on July 29 before rebounding to 550.42, with its 200-day moving average at 414.87. The Nasdaq-heavy QQQ closed at 731.07, holding well above its 200-day average of 649.53. In other words, the market is still paying up for AI infrastructure exposure even as it braces for macro noise.

The smarter thesis is not that recession is impossible. It is that this may be the first cycle in which AI-driven capital intensity supports GDP and industrial demand without translating into a classic labor-market breakdown. That creates an asymmetric setup: stay overweight the picks-and-shovels beneficiaries of AI buildout, and be selective on businesses that need a broad consumer or factory rebound to justify their valuations. If the next leg of growth comes from compute, power and data-center buildout, the market will reward the toll roads into that spending long before it rewards the rest of the economy.