Micron Technology Investment Deep Dive

Written October 1, 2026, using data through September 30, 2026. MU refers to Nasdaq-listed Micron Technology. Financial periods are Micron fiscal periods and dollar amounts are U.S. dollars.

Micron’s investment case rests on a difficult judgment: how much of the extraordinary profitability created by today’s memory shortage will survive when supply catches up? AI gives the company a stronger demand foundation, more demanding products, and closer relationships with customers. Those changes could raise its long-term earnings power. They also coexist with the pricing volatility and capital requirements that have defined memory manufacturing for decades.

My assessment is constructive on the business, with valuation dependent on the durability of margins. Treating Micron as an unchanged commodity producer understates its opportunity. Treating its current earnings as permanently repeatable understates its risk. The strongest analysis separates technological differentiation, contractual protection, and temporary scarcity, then asks what each contributes to earnings through a full cycle.

Micron manufactures memory and storage. DRAM provides fast working memory and loses its contents without power; NAND retains data and underpins solid-state storage. High-bandwidth memory, or HBM, is a form of stacked DRAM used in demanding computing applications. Micron therefore participates in both the rapid movement of information and its longer-term storage. ^[1]

That distinction matters economically. Demand can expand because customers buy more devices, because each device contains more memory, or because memory prices rise. All three increase revenue, but they say different things about sustainable growth. Higher memory content can support recurring demand. Higher selling prices can generate spectacular profits while encouraging customers and competitors to change their behavior.

The latest results show the scale of the change. Fiscal 2026 ended September 3, 2026. The figures below use GAAP gross margins and diluted earnings per share; adjusted free cash flow is Micron’s non-GAAP measure. ^[2]

• Revenue: fiscal 2025 $37.38 billion; fiscal 2026 $133.19 billion; Q4 fiscal 2026 $54.23 billion. ^[2]
• GAAP gross margin: fiscal 2025 39.8%; fiscal 2026 80.7%; Q4 fiscal 2026 86.8%. ^[2]
• GAAP diluted EPS: fiscal 2025 $7.59; fiscal 2026 $74.33; Q4 fiscal 2026 $32.87. ^[2]
• Adjusted free cash flow: fiscal 2025 $3.72 billion; fiscal 2026 $62.31 billion; Q4 fiscal 2026 $33.20 billion. ^[2]

Annual revenue increased approximately 256%, calculated from reported results. ^[2] Yet the most revealing feature is the margin expansion. Revenue growth accompanied by moderate margins would suggest an increasingly large manufacturing business. Gross margins above 80% suggest unusually favorable pricing, product economics, or both. Investors need to determine which portion can remain when shortages ease.

For fiscal Q1 2027, Micron guided to revenue of $61.5 billion, plus or minus $1.5 billion, and non-GAAP gross margin of approximately 86.25%. This outlook implies continuing strength into the next quarter; it remains management guidance rather than a realized result. ^[3]

The Q4 pricing data helps answer that question. DRAM bit shipments rose by a mid-single-digit percentage sequentially, while average selling prices increased in the high teens. NAND bit shipments rose approximately 10%, while prices rose approximately 30%. Revenue growth was therefore substantially faster than physical shipment growth. ^[3]

Pricing is central to the current earnings surge. More demanding products can justify higher prices, but the shipment figures caution against interpreting every dollar of revenue growth as equivalent growth in underlying consumption. If price increases slow, earnings can remain high while revenue growth decelerates. If prices fall materially, the same manufacturing footprint can produce a sharply different profit outcome.

AI improves the demand side because powerful processors need fast access to large quantities of data. Buying additional compute delivers less value when processors spend too much time waiting for memory. That creates an economic reason for customers to pay for bandwidth, capacity, and efficiency: better memory can improve the productivity of expensive infrastructure.

Micron’s HBM4 product specifications illustrate the technical demands. The company reports bandwidth above 2.8 terabytes per second per stack, pin speeds above 11 gigabits per second, and improved power efficiency relative to its HBM3E generation. These are manufacturer specifications, rather than an independent comparison of all competing products. ^[4]

The commercial advantage depends on translating those specifications into qualified products, reliable supply, and acceptable manufacturing yields. A component that performs well in isolation still has to work within the customer’s complete platform. Qualification and joint engineering can make customer relationships more durable, although every subsequent product generation creates another competitive test.

Inference—the process of running a trained model—also makes memory capacity consequential. In Micron’s agentic AI study, heavy concurrency exhausted the available key-value cache, the memory holding intermediate attention data, and sharply worsened inference performance despite available computing resources. This is evidence from one vendor-run experiment, rather than a universal forecast for every AI workload. ^[5]

The investment inference is that AI demand can extend beyond training clusters into systems serving many simultaneous users. Nevertheless, better software, smaller models, compression, and different hardware architectures can alter memory requirements per request. Long-term demand depends on whether growth in usage exceeds improvements in efficiency. Rising AI adoption alone does not establish a precise memory revenue forecast.

HBM has an additional supply effect. Micron explains that producing a given number of HBM bits requires more wafers and cleanroom space than conventional DRAM at the same technology node. Allocating manufacturing resources to HBM can therefore restrict conventional memory output. The company also warns that a reversal of that allocation could increase conventional DRAM supply. ^[1]

This mechanism links the AI opportunity to the wider memory cycle. AI demand can improve economics across products that are not themselves HBM. The reverse is equally relevant: weaker HBM demand could release manufacturing resources into ordinary DRAM, putting pressure on prices. HBM’s influence reaches beyond its direct contribution to sales.

The segment results reinforce that broader interpretation. Q4 Cloud Memory revenue was $16.28 billion at an 83% reported segment gross margin. Core Data Center revenue was $18.00 billion at 90%, and Mobile and Client revenue was $13.11 billion at 90%. These are segment measures, distinct from consolidated GAAP margins. ^[3]

Cloud Memory includes conventional memory alongside HBM, so its revenue cannot be used as a proxy for HBM sales. Management also said higher HBM mix offset pricing benefits in that segment and that 2027 HBM price increases would narrow the margin gap with conventional DRAM. ^[1], ^[3], ^[6]

That is a significant qualification to the popular HBM narrative. A technologically advanced product can provide strategic value and excellent growth without having the highest current percentage margin. Conventional DRAM scarcity can temporarily produce even better economics. HBM’s long-term appeal lies in customer integration and differentiation as well as profitability.

Micron’s strategic customer agreements offer a second possible source of durability. Management reported 26 agreements covering an estimated more than 35% of revenue through 2030. They are multi-year take-or-pay arrangements. Three-quarters of that estimated agreement revenue has a defined pricing framework; most of this portion uses floors and ceilings. The remainder uses periodically negotiated market pricing. ^[6]

The contracts could improve the floor beneath earnings, but they do not fix the whole income statement. A price floor can limit downside on covered business; a ceiling can surrender upside during a shortage. Contract coverage is substantial but partial, and management’s percentage is an estimate of future revenue. The effect on profits also depends on manufacturing costs, product obligations, and the terms governing customer commitments.

This is why a backlog or commitment figure should not be treated as immediately earned revenue. Long-term contracts can help finance and plan expansion while still leaving substantial execution risk. The useful question is whether covered volumes and pricing support attractive returns on the capacity Micron must build to fulfill them.

Customer funding needs equally careful treatment. Management reported $32 billion in financial commitments under the agreements and extensions, predominantly deposits. It said $12.7 billion of deposits were on the balance sheet, are unrestricted, and are returned later in the agreements if minimum purchase requirements are met. Deposits are reported as financing cash flows and do not inflate reported free cash flow. ^[6]

That accounting distinction is favorable: the strong free cash flow does not simply result from classifying customer deposits as operating cash. But deposits still carry future obligations. An investor assessing excess cash should consider those obligations before assuming all cash can be distributed to shareholders. Funding support and distributable wealth are different concepts.

Micron reported $89.68 billion of fiscal 2026 operating cash flow. Subtracting $30.71 billion of gross property and equipment expenditure gives approximately $58.96 billion. Its $62.31 billion adjusted free cash flow additionally reflects government incentives and small equipment-sale proceeds in net capital spending. Both figures show substantial cash generation, with different treatment of investment offsets. ^[2]

Free cash flow deserves more attention than earnings alone because memory manufacturing requires continuing investment. Depreciation spreads previous spending across accounting periods; the cash needed for new equipment and facilities arrives on its own timetable. A company can report strong profits while a large expansion program consumes much of the cash those profits generate.

The investment program is growing. Micron’s July announcement raised planned U.S. investment to more than $250 billion through 2035, supporting a goal of producing 40% of its DRAM domestically. That is a multi-year plan covering fab and technology investment, rather than a single year’s capital expenditure budget. ^[7]

Management expects Idaho’s first new fab to start wafer output in mid-2027, a second in late 2028, and initial New York output in 2030. It also expects roughly $25 billion of net capital spending in the first half of fiscal 2027, with greater spending in the second half. These are forward-looking plans. ^[3], ^[6]

Long lead times can sustain shortages before new capacity arrives. They also create a forecasting challenge: investment decisions made during exceptionally favorable conditions produce output years later. The relevant test is whether demand and competitive supply still justify the planned capacity when that capacity becomes productive. Government support can reduce the cost of investment, but it cannot guarantee profitable demand.

Competition remains formidable. SK hynix announced readiness for HBM4 mass production in September 2025. Samsung announced commercial HBM4 shipments in February 2026. Those announcements demonstrate active competition in the same product generation; each company’s performance and leadership claims should be understood as its own statements. ^[8], ^[9]

Micron’s competitive position therefore needs repeated proof. Customer qualifications, production yields, supply reliability, and progress into subsequent HBM generations matter more than an isolated claim to the fastest product. Customers also have reasons to maintain multiple qualified suppliers. Micron can win meaningful business while buyers preserve alternatives and bargaining power.

The wider risk is that competitors add capacity aggressively enough to weaken the pricing environment. Even firms that lag at the most advanced HBM generation can affect conventional memory pricing. Investors should distinguish competition for premium customer platforms from competition across the entire memory market; both can influence Micron’s earnings.

Micron’s regulatory filing identifies Samsung, SK hynix, Kioxia, Sandisk, CXMT, and YMTC among its competitors. It warns that government-supported investment can contribute to oversupply. It also discloses that most fiscal 2025 DRAM output came from Taiwan, and describes Chinese restrictions affecting sales to critical information infrastructure operators. These are concrete supply and market-access exposures. ^[10]

The customer side presents another potential interruption. If AI infrastructure spending slows, the impact can travel through the supply chain before overall AI usage declines. Customers may defer platform deployment, rebalance inventories, or use existing hardware more intensively. Take-or-pay arrangements provide protection on covered commitments, while uncovered business and later negotiations remain exposed.

History provides a useful check on confidence. Micron recorded a $5.83 billion GAAP net loss in fiscal 2023. That does not prove a comparable downturn is imminent, but it shows why recent profitability should be evaluated across several operating conditions. ^[1]

Valuation is where these uncertainties become consequential. A low earnings multiple can mean investors are missing a durable improvement. It can also mean the denominator contains profits that will decline. For Micron, choosing an earnings base is at least as important as choosing a multiple.

For illustration, use a share-price reference of approximately $1,065. FinanceCharts reports a September 30 adjusted close of $1,065.11; ChartExchange shows $1,066.85 at its near-close timestamp. The rounded reference accommodates that small provider difference and is not an asserted live execution price. ^[11], ^[12]

At $1,065, fiscal 2026 GAAP EPS of $74.33 implies approximately 14.3 times earnings. Annualizing Q4 GAAP EPS of $32.87 produces $131.48 and a multiple of approximately 8.1 times. These are calculations using reported earnings, not forecasts. The second calculation assumes one quarter’s performance repeats four times; it is useful for understanding the pace of improvement but weak evidence of sustainable earning power. ^{[2] [11]}

An investor who believes margins remain exceptional for several years can reasonably see a modest multiple on future earnings. An investor who expects supply to catch up will use a lower earnings base and may see a much less attractive valuation. The disagreement often lies in the assumed durability of profits, even when both investors agree that AI memory demand will grow.

The following sensitivity comparison makes that disagreement explicit. Each value equals hypothetical sustainable annual EPS multiplied by a hypothetical price-to-earnings ratio. The EPS assumptions and multiples are illustrative; none is a company forecast or a price target.

• $60 hypothetical sustainable annual EPS: $600 at 10 times earnings; $840 at 14 times; $1,080 at 18 times.
• $90 hypothetical sustainable annual EPS: $900 at 10 times earnings; $1,260 at 14 times; $1,620 at 18 times.
• $120 hypothetical sustainable annual EPS: $1,200 at 10 times earnings; $1,680 at 14 times; $2,160 at 18 times.

At the approximately $1,065 reference, the $60-and-14-times combination implies about 21% downside, while $90 at the same multiple implies about 18% upside. Those outcomes use identical valuation multiples. They differ only in earnings durability. More severe earnings normalization could create greater downside than these examples show. ^[11]

There is also a reason earnings and multiples might move together. If investors gain confidence that contractual protection and product differentiation have raised the earnings floor, they may accept a higher multiple. If they conclude that peak profits mainly reflect temporary scarcity, both the earnings estimate and the multiple can fall. This interaction creates more risk than a model that changes only one input.

To judge which path is developing, I would follow a small set of operating measures consistently. Average selling prices versus bit shipment growth reveal how much growth depends on scarcity. Conventional DRAM and HBM economics reveal whether the premium product portfolio is becoming more resilient. Contract coverage and pricing terms indicate how much protection extends into future periods. Capital expenditure, productive capacity, and free cash flow show whether growth is earning an adequate cash return.

Inventory and receivables deserve attention alongside those measures. Inventory can rise for sensible production reasons or because demand is weakening. Receivables can grow rapidly with revenue or because collections are deteriorating. The interpretation requires turnover measures, management explanations, and subsequent cash conversion rather than a conclusion from one balance-sheet number.

The bullish case would strengthen if Micron sustains attractive cash generation as capacity expands, converts advanced products into durable customer relationships, and keeps contract-protected pricing above the full cost of supply. The bearish case would strengthen if customer deployment slows, uncovered prices weaken, or investment produces excess capacity. Either outcome could unfold while the broader AI market continues to grow.

Micron offers compelling exposure to a critical constraint in AI infrastructure, with returns that depend on what happens after that constraint eases. The investment discipline is to estimate the earnings floor as carefully as the earnings ceiling. A valuation supported by conservative cash generation can survive disappointment; one requiring prolonged scarcity leaves much less room for error.