These 3 Stocks Have Been the Hottest Buys on the Nasdaq-100 This Year, and Here's Why They Could Still Go Higher

Sandisk, Micron, and Marvell are all up more than 200% in 2026, with Sandisk up 594%, Micron up 228%, and Marvell up 214% as AI-driven memory and custom-chip demand remains strong. Sandisk reported revenue of just under $6 billion, up 251% year over year, Micron posted $23.9 billion in revenue, up 196%, and Marvell said net revenue rose 28% to $2.4 billion while forecasting 35% growth ahead. The article argues all three stocks still have upside, supported by strong fundamentals, elevated AI spending, and bullish commentary from Nvidia CEO Jensen Huang on Marvell.

Analysis

The market is treating AI memory as a one-way trade, but the more interesting second-order effect is that this is now a capital-allocation cycle disguised as a demand cycle. The near-term winners are the memory vendors, but the next beneficiaries are equipment, substrates, packaging, and power-management suppliers as customers race to secure supply and de-bottleneck advanced memory stacks. That tends to extend the cycle longer than fundamentals alone would justify, because every hyperscaler order turns into multi-quarter capex commitments rather than spot demand.

The risk is not that AI demand disappears; it is that supply catches up faster than investors expect once pricing visibility improves and every producer tries to scale simultaneously. Memory is historically the cleanest example of margin mean reversion after peak enthusiasm, and the current setup has all the ingredients: concentrated customer base, highly levered operating models, and valuation multiples that already assume several years of flawless execution. The most fragile part of the trade is the expectation that gross margin expansion persists even as competitors re-enter and customers push for price concessions in procurement rounds.

Among the three names, the most asymmetric opportunity is not the fastest mover but the one with the most room for estimate revisions if AI custom silicon demand keeps compounding. The key question for the next 1-2 quarters is whether bookings convert into sustained backlog visibility or simply pull-forward purchases. If management commentary turns from 'exceptional bookings' to 'tight supply,' the trade remains intact; if it shifts to 'normalizing lead times,' the multiple compression could be abrupt.

Consensus is underweight the probability that this becomes a broader semiconductor sentiment trade, not just a memory trade. Once the market believes AI infrastructure spending is durable, the rerating can spill into adjacent names with less obvious leverage, especially those with content in networking and custom compute. But because these stocks have already repriced aggressively, the better risk/reward is increasingly in expressing the theme via pairs rather than outright longs.

Key Decisions for Investors

• Stay long MU on a 3-6 month horizon, but fund it with a partial short in a slower-growth semiconductor peer basket; the setup is favorable as long as AI capex remains elevated, but the pair reduces exposure to the inevitable memory ASP reversal.
• Take a tactical long MRVL vs. short a broad semiconductor ETF for 1-2 quarters; MRVL has the clearest estimate-revision catalyst if AI bookings translate into raised guide, while the ETF leg offsets sector beta.
• Avoid chasing SNDK after the parabolic move; if already long, trim into strength and consider using trailing stops, because the stock is most vulnerable to any sign of lead-time normalization or customer inventory digestion.
• For investors seeking convexity, buy medium-dated call spreads on MU rather than outright stock; the implied upside from continued shortage economics is meaningful, but call spreads cap premium bleed if the cycle starts to stabilize sooner than expected.
• Watch for commentary on customer concentration and capacity expansion over the next earnings season; if capex plans broaden beyond current hyperscaler demand, rotate out of the most crowded longs before margin compression becomes visible in reported numbers.