3 Semiconductor Stocks to Buy Heading Into September

NVIDIA’s July quarter “blowout” set the tone for the AI hardware trade, with revenue of $96B (data center $89B) and Q3 guidance of $108B ±2%, alongside 75% non-GAAP gross margin and explicit commentary that supply remains a bottleneck through FY28. Broadcom is set as a near-term catalyst (Sept 2) with Q3 revenue guided to $29.4B and AI semiconductor revenue projected to reach $16B (+200% YoY), while Micron (Sept 30) is positioned for a record memory cycle with FY Q4 revenue of $50B ±$1B, EPS $31 ±$1, and ~86% gross margin. Overall tone is constructive given $30B+ AI bookings vs $10.8B shipped and management visibility into 2028, but risks remain around margin pressure (NVIDIA gross margin likely bottoming at 71–72% in Q4) and memory price normalization in a cyclical market.

Analysis

The edge here is not “AI” generically; it is the market’s transition from debating demand to rationing scarce inputs. That tends to favor the names sitting closest to the bottleneck—NVDA and AVGO first, then the memory stack and advanced-packaging tool vendors (AMAT, KLAC, LRCX) as customers are forced to pre-commit capital to secure supply. The second-order loser is the hyperscaler cohort (GOOGL, META), where capex may still be rewarded for growth optionality, but free-cash-flow quality and capital intensity become harder to ignore over the next 1-3 quarters.

Into the September prints, the risk is that “visibility” gets misread as immunity. For NVDA, the market is already paying for backlog conversion; any sign that China, memory cost inflation, or a Q4 margin trough is taking a bigger bite than expected would compress the multiple first, with spillover into SOXX over days. For MU, the setup is more fragile because cyclical pricing inflections usually peak before consensus admits it; a slower rate of price increases or higher-than-expected capex intensity would be the first warning that the scarcity trade is maturing.

The contrarian view is that the consensus is treating supply constraints as a perpetual moat, when in semis they often become a demand-destruction mechanism a few quarters later. The most interesting long is not the broad basket, but the supplier versus spender spread: when buyers are forced to lock in scarce compute, the suppliers capture the margin pool while the customers absorb the depreciation and balance-sheet drag. That relative trade can work even if the absolute AI theme merely grinds higher rather than exploding again.

Key Decisions for Investors

• Buy NVDA on any 2-4% pullback into the next 1-2 weeks; keep it as the highest-quality exposure to the bottleneck trade. Falsify if the next guide implies a Q4 gross margin floor below ~71% or if China commentary turns into a measurable drag.
• Go long AVGO into the September 2 earnings event, but size it smaller than NVDA given customer-concentration and mix-risk. Best expression is a limited-risk call spread or a tight post-earnings add only if AI revenue visibility is reaffirmed.
• Do not chase MU into the September 30 print; wait for the report/volatility. If the quarter confirms tight supply but the stock sells off on guidance moderation, buy the dip or use a MU/SOXX relative-value long if memory pricing remains firm.
• Use a tactical pair: long NVDA or AVGO vs short GOOGL or META over the next 4-8 weeks to express supplier-vs-spender outperformance. Exit if hyperscaler capex growth slows materially or monetization evidence starts offsetting the spend burden.
• Watch AMAT/KLAC/LRCX as second-order beneficiaries; if packaging and fab-tool demand accelerates on these reports, add them on weakness as a lower-beta way to own the capex cycle.