3 Stocks Bound to Ignore The September Effect

The article identifies Nvidia, Broadcom and Micron as attractive AI-linked investments heading into year-end, citing projected AI-hyperscaler capital expenditures of nearly $800 billion in 2026 and $1.3 trillion in 2027. Nvidia trades at roughly 14x next-year earnings, Broadcom at 18x projected 2027 earnings, and Micron at about 6x FY2027 earnings. Micron's upside thesis is supported by memory-market supply tightness that management expects to persist until 2028, while Nvidia and Broadcom are positioned to benefit from sustained demand for GPUs and custom AI chips.

Analysis

The relevant dispersion is not “AI exposure” but monetization per dollar of hyperscaler capex. NVDA remains the highest-quality earnings stream but is most exposed to a shift from training clusters toward inference optimization and internally designed accelerators; AVGO is the cleaner beneficiary if large customers prioritize lower total cost of ownership and reduce merchant-GPU concentration. A rising ASIC mix can support AVGO’s growth while capping NVDA’s terminal multiple even if aggregate AI spending remains strong.

MU has the greatest operating leverage, but its low headline multiple should be treated as a cycle-adjusted warning rather than a free option: memory equities typically de-rate before supply visibly loosens. The key variable over the next 1-3 months is not management’s capacity commentary but HBM contract pricing, DRAM spot-price direction, inventory days at cloud customers, and whether Korean competitors accelerate qualified supply. If pricing holds through the next earnings cycle, MU has the largest upside to estimates; if it rolls, earnings revisions can reverse faster than for logic suppliers.

Consensus is likely underestimating the financing constraint on AI capex. The marginal dollar of spend increasingly requires demonstrable enterprise monetization, so an unchanged capex headline can still produce lower component intensity or a mix shift toward networking, custom silicon and memory. Near term, seasonal positioning is a weak standalone signal; use any broad semiconductor drawdown to add only where estimate revisions remain positive. Over 6-18 months, the more durable trade is AVGO relative to NVDA, not an indiscriminate long-AI basket.

Falsifiers: NVDA regains accelerating data-center revenue growth without gross-margin pressure from new platforms; AVGO fails to convert disclosed custom-silicon programs into backlog/revenue; or MU sees HBM/DRAM pricing and utilization sustain above expectations through two reporting periods. A material hyperscaler capex-guide cut or a sharp rise in semiconductor inventory would invalidate the sector-beta long thesis immediately.

Key Decisions for Investors

• Initiate a 6-12 month pair: long AVGO / short NVDA in equal dollar beta-adjusted notional. Thesis is ASIC share gains and lower concentration risk; target 15-20% relative return. Stop if NVDA’s next two data-center reports show re-acceleration with stable gross margin, or if AVGO’s AI revenue/backlog conversion disappoints.
• Maintain MU as a tactical long only into confirmed memory-price strength, preferably after monthly DRAM/HBM pricing and inventory data validate tightness; use a 3-6 month horizon and size below AVGO/NVDA because cycle downside is asymmetric. Exit on sequential contract-price declines or inventory-days expansion; upside is 25-35% if FY estimates continue rising, versus 15-20% downside in a pricing rollover.
• For broad risk-off entry, buy SMH only after a 8-12% sector pullback while hyperscaler capex guidance remains intact; avoid chasing a calendar-based rally. This is a 1-3 month tactical trade with a stop on a capex-guide reset from a major cloud platform.
• Avoid treating quoted forward P/Es as valuation support without reconciling fiscal-year definitions and peak-cycle earnings assumptions. Set an earnings-revision alert: reduce exposure if consensus 2027 EPS falls more than 5% for NVDA/AVGO or more than 10% for MU.