2 Trillion-Dollar AI Infrastructure Stocks With Up to 136% Upside, According to Select Wall Street Analysts

Raymond James analyst Simon Leopold set a $515 Nvidia price target, implying 136% upside from its Sept. 11 close, citing persistent AI-data-center hardware shortages, potential added manufacturing capacity, and demand beyond hyperscalers. Melius Research's Ben Reitzes targets $2,200 for Micron, or 126% upside, after the company disclosed at least $100 billion of strategic customer agreements through 2030. The bullish case rests on sustained supply constraints in HBM, DRAM and NAND supporting Micron's pricing power and elevated margins, though both forecasts are analyst opinions rather than company guidance.

Analysis

The actionable issue is not incremental AI demand but whether the bottleneck migrates from GPU availability to advanced packaging, HBM qualification and power/networking. NVDA’s revenue can remain supply-constrained even as end-customer demand broadens; this favors upstream capacity owners such as TSMC (TSM) and, indirectly, HBM leaders SK Hynix and MU over lower-value server assemblers. A supply release is initially earnings-positive for NVDA but may also expose its forward multiple to normalization if delivery lead times compress faster than enterprise AI utilization improves.

MU’s structural re-rating requires its contracted backlog to be economically binding, priced above future spot-market downside, and backed by credible customer volume commitments. Memory investors should not treat nominal multi-year agreements as equivalent to take-or-pay revenue: cancellation clauses, product-mix flexibility, qualification timing and capex discipline remain the critical diligence items. The key 1-3 month catalyst is earnings disclosure on HBM bit shipments, HBM3E/HBM4 yields and gross-margin bridge; the 6-18 month risk is that Samsung’s HBM qualification progress or aggressive DRAM capacity additions restore the historical commodity cycle.

Consensus is likely underweighting the asymmetric downside from a capex digestion phase rather than a collapse in AI spending. Hyperscalers can slow accelerator deployment while maintaining headline AI capex by redirecting spend toward networking, power and internally designed silicon; that outcome hurts NVDA unit growth and disproportionately challenges memory pricing. Conversely, sustained non-hyperscaler adoption would broaden the demand base and reduce dependence on a small set of buyers, supporting both companies’ pricing durability.

Key Decisions for Investors

• Prefer a 3-6 month long MU / short SOXX pair rather than outright semiconductor beta: MU offers the cleaner upside if HBM contract economics and margin durability are confirmed, while the hedge reduces exposure to a broad AI-capex reset. Exit if MU guides HBM pricing or consolidated gross margin materially below prior-quarter expectations.
• Maintain NVDA exposure only on post-results confirmation of supply conversion: add if Data Center revenue upside is accompanied by stable gross margin and evidence that lead times remain extended; reduce if revenue rises primarily from backlog release while forward gross-margin or networking attach-rate guidance weakens.
• Establish a watchlist long TSM against short lower-margin AI server OEM exposure (for example, SMCI) over 6-12 months. Advanced-packaging scarcity captures value upstream, whereas OEM margins are vulnerable when GPU supply normalizes; invalidate if CoWoS capacity additions exceed demand growth and packaging lead times contract sharply.
• For MU, require disclosure that strategic agreements include enforceable volume/pricing protections before underwriting a multi-year premium multiple. If the next earnings release shows rising NAND/commodity DRAM exposure rather than HBM-led mix expansion, treat the long thesis as a cyclical trade rather than a structural re-rating.