Last price
$455.07
Memory chips (HBM, DRAM, NAND); essential for AI training and inference.
Live quote sourced from Yahoo Finance. Prices cited in narrative below reflect the original memo date and may be stale.
Micron manufactures memory chips: DRAM (volatile), NAND flash (storage), and emerging high-bandwidth memory (HBM). HBM is critical for AI training and inference (NVIDIA GPUs, AMD MI300, etc.). As AI demand scales, HBM capacity becomes a bottleneck; Micron is one of three HBM suppliers (NVIDIA, SK Hynix, Micron). DRAM and NAND are essential for data centers and edge devices.
MU is the HBM bottleneck beneficiary and the most directly thesis-aligned memory supplier. HBM (high-bandwidth memory) is essential for AI training and inference—it is as foundational as NVIDIA GPUs. Every NVIDIA H200, AMD MI300, and custom-silicon accelerator requires HBM. Capacity constraints mean MU is supply-constrained for 12–18 months, with pricing and margins that reflect this scarcity.
HBM is the memory tier that connects AI accelerators to the data pipeline. Training a 1T-parameter model requires 10–20 HBM stacks per GPU. NVIDIAs H200 and H300 roadmaps assume 12 HBM stacks per GPU. AMDs MI300 uses 12 HBM stacks. Metas MTIA uses HBM. If AI compute grows 10x by 2028, HBM demand grows 10x. Supply is constrained to three vendors (NVIDIA, SK Hynix, Micron). MU is ramping production (targeting 300K+ units in 2026). Even at full capacity, HBM will be supply-constrained through 2027. Pricing is premium (HBM margins ~40%+); demand is price-inelastic (customers pay because there is no alternative).
| Memory Type | HBM Role | AI Exposure | Margin Profile | Competitive Position |
|---|---|---|---|---|
| HBM (high-bandwidth) | Essential for AI training/inference | Core (critical bottleneck) | 40%+ GM | Capacity constrained; MU 2nd to SK Hynix |
| DRAM (data-center) | Essential for system memory | Supporting infrastructure | 30%+ GM | Commodity; intense competition |
| NAND flash (storage) | Essential for data-center storage | Supporting infrastructure | 25–30% GM | Commodity; competitive pressure |
| Emerging (CXL, HBM stacking) | Next-gen memory interfaces | Exploratory stage | TBD | Early-stage development |
Every accelerator (GPU, custom silicon) requires HBM. Supply is constrained to three vendors. MU is ramping, but SK Hynix is ahead. For the next 12–18 months, HBM is supply-constrained, which supports premium pricing (40%+ gross margin).
NVIDIA integrates HBM vertically; SK Hynix and MU are the only merchant suppliers. MU has pricing power by default—if there is HBM demand and NVIDIA cant supply it, MU wins.
Every data-center expansion requires DRAM. AI inference workloads are memory-bandwidth-sensitive. DRAM is the largest TAM (70%+ of MU memory revenue) and is growing 8–10% CAGR from cloud and AI adoption.
MU is trading at historical multiples despite 73–77% guidance for FY26 (vs. historical 5–10% growth). If HBM margins sustain, MU is undervalued.
SK Hynix will likely capture 40–50% of HBM supply by 2027. MU could end up 2nd-best positioned, with less pricing power. HBM margin compression is a risk if SK Hynix oversupplies.
If HBM supply catches up to demand by 2027–2028, pricing collapses to commodity levels (20–25% gross margin). MUs current valuation assumes sustained premium; a cycle downturn would be a shock.
These segments (70–80% of current revenue) are exposed to pricing cycles. Oversupply risk is real if capex is too aggressive. Blended margins could compress if DRAM/NAND soften.
MU is spending $15–20B annually on fabs to scale HBM. If demand disappoints or cycles trough, capex ROI could be negative. Taiwan geopolitical risk is an existential concern.
MU is the most directly thesis-aligned memory supplier. HBM is as foundational to AI infrastructure as NVIDIA GPUs. Supply is constrained, and MU is the only independent merchant supplier scaling to meet demand. For the next 12–18 months, HBM is MUs tailwind: 40%+ margins, supply-constrained pricing, high volume growth. After 2027, if SK Hynix outscales MU and memory markets cycle, MU faces margin compression and multiple compression. The verdict is Highly Positive for the next 18–24 months, with a structural headwind emerging in 2027–2028. Current guidance (73–77% growth, rising margins) supports aggressive positioning now, with a disciplined exit before the cycle turns.
Micron is essential for AI infrastructure (HBM bottleneck); strong business fundamentals, but indirect thesis benefit.