These Experts See 'Screaming Buys' in AI Stocks After a Recent Pullback

AI chip leaders slid from highs—Nvidia down ~16% from its May high and Broadcom ~25% from its early-June record—while memory stocks surged, with Micron hitting record highs. Citi and other analysts frame the pullback as “healthy,” arguing AI compute demand remains undersupplied and reiterating top picks like Nvidia and Broadcom, plus Qualcomm (~30% off late-May highs) and Cerebras (~40% off May). Overall, the Street sees a dip-buying setup for AI infrastructure names based on demand-supply imbalance.

Analysis

The setup is less about a broken AI trade and more about a crowded factor rotation inside semis. Compute leaders still have the better earnings-quality profile: higher gross margin, more recurring software/stack leverage, and less direct exposure to spot pricing than memory, so a pullback in NVDA/AVGO is more likely a re-entry point than a thesis break unless hyperscaler capex rolls over.

Memory is the near-term momentum winner, but that can reverse fast if investors start discounting peak scarcity rather than peak demand. If HBM/DRAM pricing stays tight, MU can keep working for another quarter or two; if capacity additions from the supply chain show up, the stock can de-rate quickly because the market is already paying for scarcity. That makes memory the more fragile consensus trade over a 3-6 month window.

The second-order risk is that the market is lumping together very different AI exposures. QCOM and INTC can bounce on sympathy, but they are still mostly “AI adjacency” names; if capital rotates back to direct data-center beneficiaries, those names should lag, while NVDA/AVGO/MRVL/AMD regain leadership. The main falsifier for the dip-buy thesis is not tape action, but evidence that cloud capex growth decelerates for two consecutive quarters or that export/supply constraints translate into shipment delays rather than simply higher pricing.

Key Decisions for Investors

• Stagger into NVDA and AVGO on further 2-3% weakness over the next 1-2 weeks; 3-6 month risk/reward favors a rebound toward prior highs if hyperscaler capex guides remain intact. Invalidate if next earnings or capex commentary shows a meaningful deceleration in data-center demand.
• Pair trade: long NVDA/AVGO vs short QCOM/INTC for the next 1-3 months to isolate true AI infrastructure exposure from narrative AI beta. Target modest relative outperformance; cut if QCOM/INTC start printing accelerating AI-related backlog or design wins.
• Trim MU into strength or hedge with short-dated put spreads; 1-3 month upside can persist, but the stock is now the most crowded way to express AI scarcity and is vulnerable if HBM pricing or channel inventory normalizes. Falsifier: continued sequential price increases and no sign of inventory build.
• Keep MRVL on watch as a secondary beneficiary rather than a core long; it should outperform if the market refocuses on networking and interconnect bottlenecks. Best entry is after broad semi pullbacks, not after momentum spikes.