The Profit Margins for These 3 AI Infrastructure Stocks Are Off the Charts

Sandisk, Broadcom, and Micron are showing exceptionally strong gross margins of 78.4%, 77.0%, and 74.4%, respectively, driven by surging AI infrastructure demand and tight supply. The article argues these margins are likely cyclical rather than permanent, but says AI spending remains elevated, with hyperscalers and Tesla expected to spend a collective $750 billion on AI infrastructure this year and Alphabet planning up to $190 billion. Micron is highlighted as the cheapest of the group at about 35x earnings versus 83x for Broadcom and 53x for Sandisk.

Analysis

The near-term winner set is more nuanced than the headline implies: the real beneficiary is not just the memory/ASIC vendors, but the entire AI capex stack that can lock in pricing power before incremental supply normalizes. The second-order effect is that hyperscalers’ spend is effectively subsidizing upstream vendors’ margins today, while compressing the bargaining power of smaller OEMs, legacy storage suppliers, and any late-cycle capacity entrants that are underwriting fab/storage expansion at peak profitability.

The key risk is duration mismatch. These businesses can look like software on gross margin for a few quarters, but the assets behind them are still cyclical manufacturing assets with long lead times; once capacity catches up, margin mean reversion can be sharp and nonlinear. The market is likely underpricing how quickly consensus can flip from "scarcity premium" to "inventory digestion" if AI capex growth merely slows rather than contracts, especially over the next 6-18 months.

Contrarian view: the strongest signal here may be not that demand is strong, but that investors are extrapolating scarcity into valuation multiples already discounting an extended boom. That is most dangerous for the higher-multiple name in the group, where upside from continued strength may be smaller than downside if spend pauses. The cleaner expression is to own the most fundamentally levered beneficiary on a pullback, and fade the most duration-sensitive multiple expansion where good news is already priced.

For portfolio construction, this argues for selective exposure rather than an indiscriminate AI-infrastructure basket. The thesis is strongest while hyperscaler capex guidance continues to ratchet higher; the first material sign of deceleration will likely hit these names 1-2 quarters before it shows up in end-market data, because orders and channel inventory will lead reported fundamentals.

Key Decisions for Investors

• Long MU on weakness over the next 1-3 months; use the lower relative multiple as the best risk/reward expression of continued AI capex, with a favorable setup if hyperscaler guidance remains upward revisions-only.
• Reduce or hedge AVGO exposure via covered calls or a modest short against a basket; the valuation already prices in extended scarcity, so upside is less convex and drawdown risk is larger if AI spend growth normalizes in the next 2-4 quarters.
• Pair trade: long MU / short a legacy storage or non-AI semiconductor proxy to isolate the pricing-power cycle from broader chip beta; target a 3-6 month horizon as inventory and order momentum diverge.
• Use call spreads rather than outright longs in SNDK for a 3-6 month window; the operating leverage is attractive, but the cyclical profile makes single-name multiple expansion fragile if supply additions start to show up.
• Set a catalyst watch on hyperscaler 2026 capex commentary; if guidance decelerates, rotate out of the basket quickly because the gross-margin peak is likely forward-looking by at least 1-2 quarters.