2 ‘AI Bottleneck Stocks' That Could Continue Their Heroic Run For Investors – AAOI + MU

Micron reported FY26 Q1 revenue of $13.6B, with revenue, gross margin and EPS all beating guidance, while management said memory and storage markets should remain supply constrained through 2028. Applied Optoelectronics delivered FY25 revenue growth of 82.8% to $455.7M and signaled break-even EPS in Q1 2026, with potential FY27 revenue of $4B+ and major capacity expansion plans. Both companies are positioned to benefit from AI-driven demand for memory, storage and optical interconnects.

Analysis

The biggest second-order effect is that the AI capex cycle is becoming a memory-and-interconnect bottleneck story, not just a GPU story. If hyperscalers keep shipping racks faster than the supply chain can add HBM, SSDs, and optical links, margin leverage should migrate upstream to the few vendors with qualified capacity and vertical integration. That favors MU more than the broader semiconductor complex, while AAOI becomes a higher-beta proxy for the same buildout with much less earnings quality.

For MU, the market may still be underappreciating how supply discipline can extend the cycle well beyond the current AI deployment phase. When memory tightness persists, the usual cyclical oversupply trigger gets delayed because customers will over-order to secure allocation, which inflates revenue visibility for several quarters and can keep pricing firm even if end-demand normalizes. The risk is that this also raises the probability of a later air pocket if capacity comes on line faster than token growth, especially into 2027-2028 when new fabs and industry capacity additions converge.

AAOI is more interesting as a transition story than as a clean fundamental compounder. The company’s upside is not in current profitability but in whether it can convert capacity announcements into multi-quarter design wins before larger optics players crowd the same demand pool. The market is likely to overreact to any evidence of hyperscaler qualification, but the downside is equally severe if customer concentration or execution slips; this is the kind of name where a single delayed ramp can reset the multiple quickly.

The contrarian take is that the consensus is probably too linear on AI token growth translating into immediate hardware demand. A lot of token consumption can be absorbed by software optimization, model routing, and lower-precision inference, which can slow the rate of unit growth in memory and optics relative to headline usage. So the setup is bullish, but the cleaner trade is to own the constrained, irreplaceable layers of the stack and avoid extrapolating AOI’s ramp too far until the orders become recurring rather than aspirational.

Key Decisions for Investors

• Long MU on pullbacks over the next 1-3 months; target a 15-20% upside with downside limited to a modest multiple de-rating unless HBM pricing or capex guidance cracks.
• Use AAOI only as a tactical momentum trade: buy post-earnings or on confirmed hyperscaler qualification news, but size small because the risk/reward is asymmetric to the downside if any ramp slips.
• Pair trade: long MU / short a basket of legacy memory or weaker capex-exposed semis over 3-6 months to isolate allocation power and reduce exposure to broad semiconductor beta.
• For higher-risk expression, buy AAOI call spreads 3-6 months out; this gives upside if the data center ramp continues while capping the risk of execution disappointment.
• Avoid chasing NVDA here; the cleaner expression of the theme is the bottleneck suppliers. NVDA remains the demand driver, but the relative value is less compelling than MU's supply-constrained positioning.