Micron CEO pushes back on memory bear case. 3 key takeaways from Cramer's big interview

Micron’s CEO Sanjay Mehrotra argues the AI-driven memory cycle is likely to be durable, citing broad-based demand beyond just data centers (“no AI without memory”) and constrained supply due to long factory lead times. Micron plans $250B of U.S. investment through 2035 (up to $6.2B from the CHIPS Act), with Boise fab first wafers in 2027 and production ramps in 2028–late 2028 and Clay, NY in 2029–2030; it also reported strategic customer agreements with 16 customers (including “four very large” and “three medium”). On capital returns, Mehrotra signaled future buybacks after Micron’s CHIPS Act-related 2-year buyback ban ends in early December; free cash flow was ~$17.5B in the fiscal Q3 (June), expected to be ~$50B in FY2026, implying potentially ~10% of market cap repurchases in FY2027 if excess cash is returned.

Analysis

The market is still pricing MU like a normal memory cyclical, but the mechanism here is a longer-duration supply lockup: HBM is becoming a bottleneck input to AI capex, not a commodity with easy substitution. That changes bargaining power toward the suppliers with the cleanest allocation, and it should keep margins and contract visibility elevated longer than the sell-side currently models. The first-order winner is MU; the second-order winners are the GPU/AI platforms that can secure memory allocation early, while buyers that cannot pre-commit may see deployment delays.

The bigger near-term catalyst is not just earnings, but the restart of capital returns once buyback constraints lift. If cash generation remains this strong into early 2027, the equity story can shift from "peak cycle" to "scarce cash compounder," which is enough to re-rate the multiple even if the absolute earnings peak is behind us. That matters because the stock is still being valued as if the next inflection is down, when the more likely inflection over the next 2-4 quarters is shareholder yield acceleration.

The main falsifier is demand digestion before the new fabs matter: if hyperscaler AI capex pauses, or if spot DRAM/HBM pricing rolls over materially for two quarters, the bull case loses its basis. Longer-term, the 2027-2029 capacity ramp is the obvious overhang, so the trade is not about owning MU forever; it is about owning the spread between today’s scarcity and the market’s underappreciation of how slowly that scarcity can be arbitraged away. Consensus is missing that the timing mismatch between demand growth and fab supply is now measured in years, not quarters.

Key Decisions for Investors

• Long MU on pullbacks over the next 1-3 months; target a re-rating as FY26/FY27 cash flow and buyback capacity become the focal point. Risk/reward is attractive if forward EBITDA remains stable, but cut if management flags pricing normalization or weaker SCA conversion.
• Use MU call spreads with 12-18 month tenor rather than outright calls to express the buyback/cash-return catalyst with defined risk. Best structure is a deferred upside bet into the post-December capital return window.
• Pair trade: long MU / short SSNLF or SKHYV for relative value. Thesis: MU has the cleaner U.S. shareholder-return catalyst and is less likely to be discounted for slower capital allocation; stop if Korean peers show materially better HBM pricing or faster buyback cadence.
• Watch NVDA and AMD only as second-order indicators, not primary longs here: if memory constraints start slowing GPU shipment growth, that supports MU pricing power but can pressure AI leader multiples. Falsify the bull case if hyperscaler capex guidance rolls over or memory lead times compress materially.