Analysis-Investors wary of slowdown in US corporate profit boom

S&P 500 full-year earnings are expected to rise 35%, helping drive the index up about 12% year-to-date, but growth is forecast to slow to 15% in 2027 as AI capital-spending growth moderates. Investors are questioning the durability of AI infrastructure profits, with the median AI infrastructure stock's forward P/E falling to 22x from 32x in April and the S&P 500 technology sector multiple declining to about 21x from 26x. Five AI hyperscalers are projected to increase spending from just over $800 billion this year to $1.1 trillion next year, though the growth rate would decelerate from nearly 100% to 37%, while higher rates, oil prices and a potential consumer slowdown remain risks.

Analysis

The investable distinction is not whether AI spending remains high, but whether incremental spending continues to surprise upward. Infrastructure equities are priced off the second derivative of hyperscaler capex and earnings revisions; a deceleration in growth can compress multiples even while revenue and absolute capex remain robust. This favors suppliers with constrained, differentiated content and contractual visibility—MU’s HBM and high-end DRAM exposure—over system assemblers such as SMCI, whose revenue is more exposed to order timing, component availability, customer concentration, and a normalization in server build intensity.

The next 1-3 months hinge on hyperscaler guidance and whether management teams can demonstrate monetization rather than merely announce capacity additions. A broad reduction in AI-infrastructure valuation premiums creates a lower bar for upside surprises, but it also makes estimate cuts particularly damaging where earnings expectations still embed rapid unit growth. Regulatory delays on power, grid interconnection, and local data-center approvals would be a delayed demand shock: chips may remain supply constrained near term, while rack, networking, cooling, and construction demand bears the first order deferral risk.

Contrarianly, slowing aggregate AI capex growth need not imply a semiconductor downcycle. Spending can rotate from initial cluster deployment toward memory-rich inference systems, raising memory content per server even if server-unit growth cools. The larger macro risk is that higher real rates force a simultaneous de-rating across long-duration technology and weaken consumer-sensitive earnings, leaving banks and cyclicals unable to provide the usual breadth offset if credit costs begin rising.

Key Decisions for Investors

• Initiate a 3-6 month long MU / short SMCI pair, sized beta-neutral. MU offers exposure to memory-content growth and potentially tighter HBM supply, while SMCI is more vulnerable to a pause or deferral in data-center deployment. Reassess if MU fails to guide HBM supply or gross margin higher, or if SMCI demonstrates sustained backlog conversion without working-capital pressure.
• Use upcoming hyperscaler earnings as an event gate: add AI-infrastructure exposure only if aggregate capex guidance is maintained or raised and management quantifies capacity utilization or AI revenue traction. A vague reaffirmation of spend without deployment milestones is a sell signal for high-beta infrastructure names rather than confirmation of the cycle.
• Favor APP over a basket of capex-dependent AI names on a 6-12 month horizon. Its AI exposure is tied to advertising yield and software monetization rather than data-center construction; the thesis fails if incremental AI-driven revenue does not translate into durable margin expansion or retention trends weaken.
• Maintain downside hedges on broad technology exposure through 3-6 month QQQ puts or a QQQ/SPY relative short if real yields rise further. The key falsifier is renewed upward earnings revisions outside mega-cap technology, which would support market breadth and reduce the valuation-duration concentration risk.