These 5 AI Stocks Have Returned 300%-Plus Over the Past 3 Years -- and 1 Still Looks Dirt Cheap

The article highlights AI-driven semiconductor momentum, citing Micron shares up ~1,320% over three years (and ~85% gross margins) alongside Nvidia’s total revenue up 91% to $81.6B and non-GAAP EPS up 140% to $1.87. It notes memory demand is still constrained by lagging supply, but investors worry new capacity could eventually pressure pricing. Overall, it frames Micron as “cheap” with P/E ~19 vs the sector ~35, while emphasizing that any drop in data-center capex could hit Micron’s margins and earnings.

Analysis

The cleanest tradeable implication is that AI semis are no longer one homogeneous beta bucket. Memory is the most reflexive part of the stack: when supply is tight, incremental revenue drops straight to profit, but when new capacity hits, earnings can mean-revert much faster than design-led names. That makes MU the highest operating-leverage exposure to the current cycle, but also the first place to see an ugly multiple reset if pricing rolls over.

By contrast, AVGO and TSM have better duration. AVGO’s custom silicon and design-in relationships make it a wallet-share winner if hyperscalers keep optimizing for efficiency, while TSM remains the toll collector with the least direct pricing risk. NVDA still has the strongest ecosystem moat, but if customers push harder into ASICs to lower unit economics, the next leg of growth could be more about product mix defense than outright share gains.

The market’s blind spot is timing: consensus is extrapolating AI capex too smoothly into 2027, while the real risk is a spending air pocket over the next 1-3 quarters if return-on-capex scrutiny rises. Falsifiers are simple: weakening hyperscaler capex guides, inventory builds, or a drop in memory pricing. If those appear, MU is the first stock to cut, then AMD; if not, the rally likely has more room but should be expressed selectively, not as a blanket long semiconductor basket.

Key Decisions for Investors

• Prefer AVGO as the highest-quality AI exposure; buy on 3-5% pullbacks for a 6-12 month hold. Risk/reward is attractive because contract visibility and custom silicon adoption should support multiple durability even if the broader AI trade cools.
• Pair trade: long AVGO / short AMD over the next 1-3 months. Thesis: AVGO monetizes AI wallet share more reliably, while AMD carries more execution and capex-ROI skepticism. Falsify if AMD order commentary or margins inflect materially higher.
• Tactical long MU only on confirmation that memory pricing remains tight into the next earnings cycle; otherwise treat as a watch item, not a core position. Upside is strong if the shortage persists, but downside is a sharp margin unwind once supply normalizes.
• Use TSM as the lower-beta semicap winner for investors wanting AI exposure with less earnings convexity. Favor TSM over NVDA if the goal is participation without paying peak-growth multiples for the most crowded name.
• Set an alert around the next hyperscaler capex round of guidance; if spending growth slows, reduce exposure to MU and AMD first and rotate toward AVGO/TSM.