Micron Technology (MU) research pages
- Micron Technology Financials: DRAM, NAND and HBM segment revenue, gross margin through cycle and capex intensity
- Micron Technology Earnings: quarterly results, guidance and memory pricing commentary
- Micron Revenue by Segment: compute and networking memory, mobile memory, embedded memory and storage breakdown
- Micron Gross Margin History: peak-to-trough margin cycle and HBM margin premium
- Micron R&D Spending: process node investment, HBM development and leading-edge research intensity
- Micron Capital Expenditure: fab investment cycle, clean room expansion and equipment intensity
- Micron Free Cash Flow: cyclical FCF profile and cycle-adjusted generation capacity
- Micron Valuation: through-cycle EV/EBITDA, price-to-book and HBM premium analysis
- Micron Technology History: 1978 founding, DRAM consolidation wave and HBM development
- Micron Drawdowns: historical memory cycle drawdowns and trough-to-peak recoveries
Quick answer
Micron Technology, Inc. (MU) manufactures DRAM (dynamic random-access memory), NAND flash storage and HBM (high-bandwidth memory) at scale. Micron is one of only three major DRAM producers globally, alongside Samsung Electronics and SK Hynix. Memory is among the most cyclically volatile segments of the semiconductor market: prices are set by supply-demand balance across a small number of global producers, and the resulting earnings swings can be severe. Micron is a member of the PHLX Semiconductor Sector Index (SOX), the benchmark index for the semiconductor industry.
The central investor question: Can Micron convert AI's structural demand for high-bandwidth memory into durably higher margins, while navigating the inherent volatility of the DRAM and NAND price cycles? HBM represents a qualitative shift in the memory business because it is differentiated, contract-priced and allocated, rather than spot-priced and commoditized like standard DRAM.
Investor takeaway: Micron's investment case depends on understanding two distinct dynamics at the same time. The first is the commodity memory cycle: DRAM and NAND prices are set by supply-demand balance across three dominant producers, creating earnings that can swing from deep losses to exceptional profits within 18 months. The second is the HBM structural opportunity: AI accelerators require stacked, high-bandwidth memory that only Samsung, SK Hynix and Micron can produce, and demand is growing faster than supply. Investors who treat Micron as a pure commodity play underestimate the HBM opportunity; investors who treat it as an AI growth stock underestimate the commodity cycle exposure.
Company at a glance
| Item | Overview |
|---|---|
| Company | Micron Technology, Inc. |
| Ticker | MU |
| Exchange | Nasdaq |
| Sector | Information Technology |
| Primary products | DRAM, NAND flash, HBM (high-bandwidth memory) |
| Core customers | Server manufacturers (hyperscalers, OEMs), PC makers, smartphone manufacturers, automotive and industrial customers |
| Primary economic drivers | DRAM bit demand and pricing, NAND bit demand and pricing, HBM allocation and contract pricing, server memory content per server |
| Key investor metrics | DRAM/NAND bit shipment growth, average selling prices, gross margin through cycle, HBM revenue share, capital expenditure intensity, net inventory days |
| Major competitors | Samsung Electronics (Memory), SK Hynix |
| Index membership | SOX, S&P 500 |
What Micron Technology actually makes
DRAM (dynamic random-access memory) is Micron's largest product by revenue, typically representing roughly two-thirds of total sales. DRAM provides short-term, high-speed memory that processors use to run active workloads. Server DRAM is the most valuable segment: each new generation of server packs more memory per socket as workloads scale, and AI training and inference servers require particularly large DRAM configurations. PC DRAM is larger in units but lower in average selling price per bit. Mobile DRAM for smartphones is a significant market but lower-margin than server DRAM. Automotive and industrial DRAM is smaller by volume but benefits from longer product qualification cycles and less commodity pricing pressure.
NAND flash accounts for approximately a quarter of Micron's revenue. Unlike DRAM (which loses data when power is removed), NAND retains data without power and is used in SSDs for data-center storage, consumer storage devices, smartphones and USB drives. The NAND market is more competitive than DRAM: in addition to Micron, Samsung, SK Hynix, Kioxia, Western Digital and Intel's former NAND business (now Solidigm, owned by SK Hynix) all produce NAND. Greater producer count and ongoing technology transitions create persistent pricing pressure and periodic oversupply.
High-bandwidth memory is a technology that stacks multiple DRAM dies vertically and connects them to a logic chip (typically an AI accelerator or GPU) via a silicon interposer, allowing dramatically higher memory bandwidth than standard DRAM. HBM is the memory type used in NVIDIA's H100 and Blackwell GPU families, AMD's Instinct AI accelerators, Google's TPUs and other AI-focused chips. Only Micron, Samsung and SK Hynix produce HBM at commercial scale, and AI demand has outpaced supply, keeping HBM pricing in a fundamentally different regime from commodity DRAM. Micron's HBM3E is qualified and in production for NVIDIA Blackwell systems.
The memory price cycle: how it works
The memory price cycle is the dominant financial driver for Micron and distinguishes memory semiconductor investing from most other technology analysis. DRAM and NAND are commodity products: a gigabyte from Micron is interchangeable with a gigabyte from Samsung or SK Hynix for most applications. Prices are set by supply-demand balance across the industry, not by individual company pricing power. When demand grows faster than supply, prices rise and producer margins expand dramatically. When supply grows faster than demand (often driven by competitors building new capacity), prices fall below cost and producers report losses.
The cycle is partially self-correcting. When prices fall below production cost, producers reduce capital expenditure on new capacity and sometimes cut wafer starts. As supply growth slows, the demand side (server refreshes, PC replacement cycles, smartphone upgrades) eventually catches up, and prices begin recovering. The cycle typically runs from trough to peak over two to three years, though the exact timing varies with macroeconomic conditions, technology transitions and the speed of capacity decisions. Key cycle metrics include: bit shipment growth (demand), wafer starts and capital expenditure commitments (supply), channel inventory days, and ASP (average selling price) trends compared to cost per bit.
Investors tracking Micron through the cycle should distinguish between cyclical earnings impairment (which reverses) and structural margin change. HBM introduces a complicating factor: a growing portion of Micron's DRAM production is going into HBM, which does not trade at commodity prices. This structural mix shift can support margins even during periods of commodity DRAM price weakness, but only to the extent that HBM allocation grows as a percentage of total DRAM bit production.
HBM and the AI memory opportunity
AI training and inference clusters require memory that delivers dramatically more bandwidth than standard DRAM. A single NVIDIA H100 GPU uses six HBM3 stacks delivering approximately 3.35 terabytes per second of memory bandwidth, compared to roughly 50 gigabytes per second for a typical DDR5 DRAM module. This bandwidth requirement cannot be met by conventional DIMM-based memory, which is why HBM is the only viable option for AI accelerators at current compute scales. As AI model sizes grow and inference workloads scale, HBM content per accelerator is likely to increase with each generation.
The economic significance for Micron is that HBM pricing is contract-based, allocated months in advance, and commands a meaningful premium over equivalent DRAM capacity in commodity form. Customers cannot easily substitute commodity DRAM for HBM in AI accelerator designs because the silicon interposer connection and the physical form factor are fixed at chip design time. This gives Micron (and its two HBM peers) pricing leverage that does not exist in commodity DRAM markets. The qualification process for HBM supply to a specific accelerator program is measured in months to years, creating switching costs on both sides of the supply relationship.
Key risks and failure modes
Micron's primary risks reflect both the structural characteristics of memory as an industry and the company-specific challenges of competing in a capital-intensive global market.
DRAM/NAND commodity cycle risk
The most immediate risk for most Micron investors is commodity memory price decline. A downturn can rapidly turn Micron's financials from significant profitability to operating losses. The triggers are typically oversupply (Samsung or other producers adding capacity aggressively) and demand weakness (enterprise IT spending cuts, slow PC cycles or smartphone unit declines). No pricing power exists to offset these moves.
Samsung and SK Hynix competitive response
Samsung and SK Hynix are substantially larger companies with memory revenues that dwarf Micron's. If either competitor pursues aggressive pricing to gain market share, or invests heavily in HBM to close any technology gap, Micron's pricing and margin position would be pressured. Samsung in particular has historically been willing to operate memory at a loss during downturns, which prolongs cycle bottoms and delays recovery.
NAND oversupply and Chinese producer expansion
The NAND market faces a structural oversupply risk from Chinese producers (YMTC and others) expanding production. Chinese NAND producers have benefited from government subsidies and domestic demand support, potentially enabling pricing below economic cost for extended periods. U.S. export controls have limited some Chinese access to leading-edge memory technology, but mid-range NAND segments remain exposed.
HBM allocation competition
Micron's HBM opportunity depends on maintaining a competitive product and receiving allocation from AI accelerator customers. Samsung and SK Hynix both produce HBM3E and will compete for the same NVIDIA, AMD and Google allocation. If a competitor achieves better performance, lower defect rates or higher yield, Micron's HBM share could decline even in a growing market.
Capital intensity and balance sheet risk
Memory manufacturing requires among the highest capital expenditures of any industry. Each new fab generation costs billions of dollars in clean room construction and lithography equipment. This creates fixed cost leverage: in good times, high utilization produces excellent margins; in bad times, fixed costs continue even as revenue falls. High capital intensity also means balance sheet stress can emerge quickly during a prolonged downturn.
How to think about valuation
Memory company valuation is notoriously difficult using current-period earnings because those earnings are a function of where the pricing cycle is at any given moment. At cycle peaks, Micron can earn $8-12 per share; at cycle troughs, it can report losses. A P/E multiple applied to peak earnings produces a number that looks cheap but may not be, while P/E applied to trough earnings produces a number that looks expensive but may also not be.
More useful frameworks include: (1) Through-cycle earnings: estimate normalized earnings at a mid-cycle DRAM price, weighting the proportion of revenue from HBM (which has less cyclical pricing). (2) Price-to-book: memory companies with strong balance sheets and good cycle management historically trade at or above book value at cycle bottoms. (3) EV/EBITDA through the cycle: compare current EV against peak, mid-cycle and trough EBITDA scenarios. (4) HBM mix trajectory: model the proportion of Micron's DRAM bits going to HBM in 12 and 24 months, estimate the margin premium, and calculate the impact on blended gross margin.
The key question for any given investment thesis is where the current cycle is. If commodity DRAM prices are at or near a cyclical bottom and HBM allocation is ramping, the forward earnings recovery can be substantial. If prices are near peak and HBM has already been priced in, future returns may be more limited. The best check on cycle position is a combination of industry inventory data, capital expenditure trend direction, producer commentary on wafer starts, and ASP trajectory across recent quarters.
Frequently asked questions
What products does Micron Technology make?
Micron Technology manufactures three main categories of memory products. DRAM (dynamic random-access memory), typically about two-thirds of revenue, is high-speed volatile memory used in servers, PCs, smartphones and automotive systems. NAND flash, roughly a quarter of revenue, is non-volatile storage used in solid-state drives for data centers, consumer electronics and mobile devices. HBM (high-bandwidth memory), a growing and high-margin category, is stacked DRAM connected to AI accelerators and GPUs via a silicon interposer, delivering very high memory bandwidth for AI workloads.
Is Micron in the SOX index?
Yes. Micron Technology, Inc. (MU) is a component of the PHLX Semiconductor Sector Index (SOX), the benchmark index tracking the semiconductor industry. Micron is one of the largest U.S. semiconductor companies by revenue and market capitalization, and its memory-cycle performance significantly influences the broader SOX index during periods of pronounced memory pricing swings.
What is HBM and why does it matter for Micron?
HBM stands for high-bandwidth memory. It is a technology that stacks multiple DRAM dies vertically and connects them to an AI accelerator or GPU chip via a silicon interposer, providing memory bandwidth many times higher than standard DRAM. AI training clusters require HBM because model sizes and inference throughput demands exceed what standard DRAM can deliver. HBM matters for Micron because it is priced on contract rather than commodity spot markets, commands a significant premium per bit over standard DRAM, and is only produced by three companies globally: Micron, Samsung and SK Hynix. Growing HBM content per AI accelerator represents a structural improvement in Micron's revenue mix relative to commodity memory cycles.
How does the memory price cycle affect Micron's earnings?
The memory price cycle has an outsized effect on Micron's financial results because DRAM and NAND are commodity products priced by supply-demand balance across a small number of global producers. When memory prices rise, Micron's gross margins can expand rapidly, producing significant earnings per share. When prices fall, as they do when supply additions outpace demand or macroeconomic conditions weaken, Micron can report operating losses within a few quarters. Earnings can swing from roughly $8-12 per share at cycle peaks to losses per share at troughs. This cyclicality makes traditional P/E analysis less useful for Micron than through-cycle frameworks that account for where pricing stands in the cycle.
How does Micron compare to Samsung and SK Hynix?
Micron, Samsung Electronics (Memory division) and SK Hynix are the three major DRAM producers globally and all three produce HBM. Samsung is the largest by revenue and has the broadest semiconductor portfolio outside memory. SK Hynix has been particularly aggressive in HBM development and was the first to supply HBM3 to NVIDIA at volume. Micron is the smallest of the three by memory revenue and the only major U.S.-headquartered memory manufacturer. All three face the same commodity pricing cycle, but differ in HBM qualification timing, technology leadership positions, geographic manufacturing exposure and balance sheet strength. For HBM specifically, SK Hynix entered the NVIDIA supply chain first, but Micron's HBM3E is qualified for Blackwell generation systems.
Related Swoopr research
Sources and research methodology
This profile is educational content, not investment advice. Company-specific facts should be refreshed against primary sources before publication and whenever a material corporate action changes the business model. Swoopr prefers SEC filings, official investor-relations materials and index-provider announcements over secondary summaries.
Primary source starting points: