Should You Buy Micron Before Sept. 30?

Micron, up 1,300% over five years but about 20% below its high, reports fiscal Q4 earnings on Sept. 30 as investors assess whether AI-driven memory demand can sustain the current cycle. The company trades at 22.9x trailing earnings and 6.6x estimated fiscal 2027 earnings, but planned industry capacity additions in 2028-29 and Nvidia's reported reduction of Rubin HBM specifications to 4-hi from 8-hi/12-hi create oversupply risk. Offsetting this, demand is estimated at 2-3x supply by some industry sources, Micron has no visibility on supply catching demand, and it had secured LTAs for roughly 20% of DRAM bits and 33% of NAND bits; management targets 50% of revenue under such agreements. Investors will also watch for buyback plans when Micron's CHIPS Act-related repurchase restriction expires Dec. 9.

Analysis

MU’s Sept. 30 setup is less about the reported quarter than whether management can demonstrate that contracted volume and pricing are becoming a meaningful buffer against spot-memory cyclicality. The key underwriting variable is not aggregate AI demand but the share of high-value HBM capacity that displaces commodity DRAM output: sustained mix tightness can keep blended DRAM pricing elevated even if Rubin reduces memory content per accelerator. A credible increase in contracted revenue coverage would justify some multiple expansion; vague language around commitments without pricing, duration, or cancellation protections should be treated as marketing rather than de-risking.

The more important second-order risk sits in 2028: simultaneous capacity additions by MU, Samsung Electronics and SK Hynix could turn today’s shortage into a utilization problem, particularly if inference efficiency reduces memory intensity faster than AI-server unit growth accelerates. MU is therefore a better 6-12 month earnings/margin trade than a clean multi-year structural compounder. Equipment suppliers with HBM exposure, notably AMAT, LRCX and KLAC, retain backlog support from the capacity build but face the same 2028 digestion risk with a lag.

Consensus may be too anchored to the historic two-year memory template, but the bullish alternative is not simply ‘AI demand remains strong.’ It requires tight HBM yields, disciplined competitor capex, and enforceable take-or-pay contracts. The near-term asymmetry is favorable only if guidance embeds further pricing/mix gains; a beat driven by prior-quarter pricing while forward HBM allocation or contract coverage stalls would likely compress the forward earnings multiple despite strong headline EPS.

NVDA’s potential memory-content optimization is a relative headwind to the memory complex, but it could improve system economics and expand accelerator deployments. That makes unit demand versus HBM content the critical read-through: lower content per GPU is not bearish for MU if total GPU shipments and HBM attach rates rise enough, but it raises the bar for the next leg of memory-price appreciation.

Key Decisions for Investors

• Maintain a tactical long MU into earnings only at reduced sizing; add on a post-report confirmation that contracted revenue coverage, HBM allocation and gross-margin guidance all improve. Target a 10-15% upside over 1-3 months versus an 8-10% downside if forward pricing or utilization commentary weakens.
• Use a defined-risk event structure rather than unhedged common: buy MU Oct/Nov call spreads financed partly with an out-of-the-money put spread, sized for post-earnings implied-volatility compression. Do not initiate until checking implied move versus the prior four earnings reactions.
• Pair long MU / short SOX or SMH through year-end for investors seeking AI-memory exposure while hedging broad semiconductor-beta and NVDA-led valuation risk. Close if MU fails to guide higher on gross margin or if HBM demand commentary shifts from allocation scarcity to customer inventory normalization.
• Set a 2028-capacity alert: reduce or hedge MU if Samsung/SK Hynix capex accelerates materially, MU’s Idaho ramp advances ahead of demand, or industry DRAM contract prices flatten for two consecutive months. Those signals would challenge the thesis that HBM-related supply diversion can offset new wafer capacity.
• Watch NVDA Rubin platform disclosures for actual HBM stack configuration and total system memory content. A lower per-GPU specification accompanied by unchanged or lower rack-level memory demand is a direct negative read-through for MU; higher rack shipments would instead support the bullish volume-offset case.