Micron Investment Thesis

Micron’s business in one minute

Micron designs and manufactures the memory and storage chips used in data centers, computers, phones, and vehicles. Its customers need chips that hold data while processors work and retain it afterward. Micron serves those needs through three main product groups: conventional DRAM, HBM, and NAND flash. [19]

Micron’s product mix

In the third quarter of fiscal 2026, ended May 28, Micron reported $41.46 billion in total revenue. Its revenue mix spans the three product groups above, although it reports conventional DRAM and HBM together. [2]

DRAM, including HBM, contributed $31.3 billion and NAND $9.9 billion. Data center SSD revenue exceeded $5 billion within NAND. [3]

Reported figures are rounded and do not exactly reconcile to total revenue.

A revenue share measures dollars, not the amount of memory shipped or its share of profit. Higher prices can change the mix even without an equivalent increase in capacity. For investors, the key is whether Micron can grow earnings across the whole portfolio.

How AI turns into memory demand and higher earnings

AI needs more memory to keep processors productive

AI models repeatedly move large amounts of data between memory and processors. Holding the model requires capacity, while feeding it to the processor requires bandwidth. Generating answers can become limited by memory delivery rather than calculation speed, increasing the importance of memory relative to compute. There is no fixed memory-to-compute ratio across workloads, but AI servers typically carry much more memory and storage than conventional servers. Micron’s infrastructure white paper estimated six times the DRAM and 2.6 times the SSD capacity per AI server. [23] [4]

HBM demand also tightens conventional DRAM supply

Building more AI accelerators raises demand for HBM. Producing HBM uses the same underlying DRAM manufacturing resources but consumes more wafer capacity per bit of memory. Wafers are the silicon discs on which chips are made. Micron’s earlier-generation comparison put HBM at roughly three times the wafer requirement of conventional server DRAM. Micron says this wafer requirement relative to conventional DRAM increases with each new HBM generation. Shifting production toward HBM therefore leaves less capacity available for conventional DRAM, with newer generations adding to that supply pressure. [12] [3]

AI servers also need conventional DRAM around their accelerators. Growing server demand and constrained supply therefore reinforce each other, pushing prices up. TrendForce’s June 2026 survey estimated that conventional DRAM contract prices rose roughly 93–98% from the preceding quarter in calendar Q1. [24]

AI creates a second demand surge in NAND

NAND serves two distinct needs in AI. First, training datasets, saved models, company documents, and generated media need persistent storage: they must survive when power is off. HBM and conventional DRAM cannot provide that on their own. SSDs built with NAND retain this data and supply it to working memory when needed. [25]

Second, NAND can hold temporary working data that does not need to survive a shutdown. An AI system might move cached information from an earlier conversation out of HBM or DRAM while it is not being used. NAND costs less per bit, so keeping that information on an SSD frees expensive memory for active work. The trade-off is higher latency: retrieving it takes longer. This works when data is accessed less often or can be loaded ahead of time, so the capacity savings outweigh the delay. NAND therefore gains demand both as permanent storage and as a cheaper extension of working memory. [26]

Profits have grown faster than revenue and the share price

Micron’s fiscal Q3 2026 revenue rose 346% year over year, from $9.30 billion to $41.46 billion. Net profit grew even faster: up 1,398%, from $1.89 billion to $28.24 billion, using generally accepted accounting principles (GAAP). That is nearly 15 times the profit on roughly 4.5 times the sales. [2]

Improving margins explain the gap. Gross margin, the share of sales left after production costs, rose from 37.7% to 84.6%. Stronger pricing lets more revenue flow through to profit. Operating leverage adds another boost: research, administration, and other operating expenses grew only 30% while revenue more than quadrupled. Spreading those expenses across much larger sales lifted operating margin from 23.3% to 80.4%. [2]

The stock’s rally has been extraordinary too. Between the two fiscal quarter-end dates, May 29, 2025 and May 28, 2026, Micron’s closing share price rose from $96.80 to $923.52, a gain of 854% before dividends. Yet even that increase lagged profit growth. Diluted earnings per share rose from $1.68 to $24.67, or 1,368%, so the earnings growth outpaced the stock even after accounting for the share count. [2] [27]

A much higher share price can still buy more earnings per dollar when profits rise faster. Whether that makes Micron attractive depends on how much of this earnings base lasts and what comes next. We address that in the valuation section.

Competition and Micron’s relative edge

Micron competes primarily with the Korean companies Samsung and SK hynix in DRAM and HBM, and with those companies and other flash suppliers in NAND. Customers choose on performance, power consumption, price, and the ability to deliver reliable products in volume. AI growth expands the opportunity, but Micron still has to win its share. [19]

We are generally bullish on memory competitors too, including Samsung and SK hynix. Expanding AI demand can benefit several suppliers. We favor Micron for its U.S. base and its combination of working memory and storage.

A U.S. base with a geopolitical advantage. Although Micron also manufactures extensively in Asia, it is headquartered in Boise, Idaho, and leads its major memory competitors in existing and planned U.S. memory manufacturing. Its domestic footprint includes production in Virginia and new fabs underway in Idaho and New York. [28] This aligns Micron with its customer base: U.S.-headquartered customers accounted for roughly 65% of fiscal 2025 revenue. [19] We see that alignment as a meaningful geopolitical advantage over Samsung and SK hynix. In a climate of shifting tariff policies and a U.S. push to bring semiconductor production home, Micron’s established domestic presence and expansion plans strengthen its position as a long-term supplier to American customers. [29]

A closer fit for investors who favor the dollar. Micron uses the U.S. dollar as the functional currency across all its operations, meaning the dollar is the economic basis on which it measures business performance. Its sales are predominantly in dollars, and it hedges foreign-currency cash and investments to protect their dollar value. [19] Samsung and SK hynix, by comparison, use the Korean won as their parent companies’ functional currency, tying their financial results to a different currency base. [30] [31] For investors who favor USD as a long-term store of value, we prefer Micron’s alignment with that base.

Both sides of the memory opportunity. Micron sells DRAM, including HBM, and NAND storage. We prefer that breadth to a NAND-focused investment such as Sandisk: Micron can benefit from the memory used to run AI and the storage used to retain its data. [22] This breadth is not unique to Micron. SK hynix also supplies DRAM and NAND, so our preference combines product exposure with Micron’s U.S. base and valuation. [21]

Why the valuation makes Micron attractive

A great business can still be a poor investment if its price already assumes exceptional results. We favor Micron because we believe its earnings can endure and grow beyond what the share price implies.

A price-to-earnings (P/E) ratio tells you how much investors pay for a dollar of annual profit. Trailing P/E uses the past 12 months. Forward P/E uses forecast earnings, generally based on analyst estimates. A low forward multiple can signal an opportunity, or a market that expects those forecasts to fall apart.

Rounded published ratios, with differing dates and methods. *SOXX forward P/E uses the index it tracks as an approximation, while trailing P/E uses the fund’s holdings. These are illustrative comparisons, not uniform calculations. Read the valuation methodology.

The striking discount is in the forward numbers. Micron’s trailing multiple is roughly in line with the S&P 500 in this snapshot. On forecast earnings, it trades at less than half the multiple of every comparator. The gap between Micron’s own trailing and forward ratios also shows how much stronger the forecast earnings base is. Our thesis requires that stronger earnings base to endure and grow. A cheap multiple on a temporary profit peak would not necessarily justify the investment.

Why the discount exists

The bear case deserves a fair hearing. Memory has a history of brutal economic feedback loops. Shortages lift prices and profits. Producers build factories. New supply eventually arrives, customers stop stockpiling, and prices fall. Because manufacturing costs do not fall as quickly as revenue, profits can collapse. Near the top of that cycle, a stock can look cheapest precisely when its earnings are least sustainable.

Micron’s low forward multiple reflects substantial skepticism about how long its current level of earnings can last. Analysts can forecast a strong coming year while investors refuse to pay much for it because they expect weaker years afterward.

Where we disagree

We think the market is underpricing the duration of demand. AI adds a recurring need for memory across an expanding range of activities. The supply coming online must serve a market that is still growing when those factories finally deliver. Announcing capacity is easy compared with producing large volumes of memory that meet customers’ requirements, and demand does not stand still during construction.

We expect expanding demand and profitable capacity additions to support earnings growth over time, even if pricing moderates. At a modest starting multiple, that growth can reward shareholders without requiring a higher valuation multiple. Greater confidence in the durability of those earnings could provide additional upside.

Why we expect earnings to keep growing

More memory shipped, a larger share of sales from premium products, and attractive pricing can lift earnings.

Three overlapping themes support our growth view. They explain how demand could grow and should not be added together as separate forecasts.

Supply constraints and long-term contracts support durability

More specialized memory is harder to replicate

Memory has historically been viewed as a commodity: standardized products that customers can substitute for one another, leaving price and production cost to drive much of the competition. As advanced memory becomes more specialized, matching a competitor’s capacity alone is increasingly insufficient.

High-bandwidth memory illustrates the shift. It stacks multiple memory chips vertically, requiring precise manufacturing and packaging while meeting demanding bandwidth and power targets. [41] Customer approval is another hurdle: Micron’s June 2026 results described HBM4 in high-volume shipments for its lead customer’s platform while qualification samples were being sent to other customers. [2] For the subsequent HBM4E generation, Micron has outlined an option to customize the logic chip beneath the memory stack for individual customers. [42]

We believe this raises the difficulty of reproducing a competitive product and replacing an approved supplier. Performance, power efficiency, manufacturing yields, and integration with the customer’s processor all matter. While standard memory remains price-sensitive, and rivals can still qualify competing products, we believe growing specialization can strengthen Micron’s customer relationships and support more durable margins as these products become a larger part of its business.

Supply takes time to deliver

The strongest bearish objection is also the obvious one: extraordinary profits will attract extraordinary investment. We agree. Where we disagree is the assumption that new capacity will arrive into a market that has stopped expanding.

A factory announcement does not translate directly into saleable memory. Between the announcement and commercial output come construction, equipment installation, yield ramping (raising the share of chips that meet specifications), advanced packaging, and customer qualification. Each step can constrain how quickly new capacity turns into chips customers can actually use and buy. Expanding floor space may increase capacity, but it does not instantly remove those bottlenecks.

While supply works through those constraints, we expect demand to keep expanding. New capacity must catch up with the market’s future needs, making today’s shortage harder to resolve than current demand alone would suggest.

Micron’s June 2026 outlook called for tight DRAM and NAND conditions beyond calendar 2027. [3] That outlook supports our view that supply will struggle to keep pace with demand for an extended period, making a quick return to chronic oversupply less likely. Memory prices can soften along the way while the longer-term supply constraints persist.

The bullish case does not require supply to stop growing. It requires demand to absorb that growth while Micron earns attractive returns on the capital needed to serve it.

Customer commitments support capacity investment

Micron’s new long-term contracts exchange dependable memory supply for binding customer purchases over several years. Customers secure memory for their future products, while Micron gains advance commitments on how much they will buy and, in many agreements, limits on how far prices can fall or rise. Micron calls these strategic customer agreements (SCAs). [3]

The business model shift is in the strength of those commitments. Micron already had long-term customer agreements, but its fiscal 2025 annual report explained that prices, quantities, and other terms were typically renegotiated periodically as market conditions and customer demand changed. Customers were generally reluctant to lock in long-term fixed prices. [19] That left Micron committing capital to factories whose eventual sales volumes and prices remained exposed to the market. Binding purchases and agreed pricing limits give it a firmer basis for those investments.

In June 2026, Micron disclosed 16 signed SCAs, typically running from 2026 through 2030, with automotive agreements generally lasting three years. They covered roughly 20% of DRAM volume and one-third of NAND volume over their terms. Their “take-or-pay” provisions require customers to purchase specified volumes or pay under the contract. The largest agreements generally include minimum and maximum prices, called floors and ceilings. Other agreements have fixed prices or remain exposed to market pricing. [3]

That shifts more of the risk of expansion onto customers seeking assured supply. Purchase commitments make production planning more predictable, and price floors provide protection when market prices weaken. Customers benefit because dependable memory access helps them plan their own product launches. Micron’s July announcement of automotive agreements with Qualcomm, DENSO, and Hyundai Mobis illustrates that demand for greater supply and pricing visibility. [18]

Much of Micron’s output remains outside these commitments, and price ceilings limit some upside. Even so, we see this shift toward committed sales as a meaningful improvement in the business model: greater confidence in future volumes and pricing can support better investment decisions, more durable earnings, and a stronger case for a higher valuation.

What could break the thesis?

We would first reconsider our bullish view if Micron’s valuation discount narrows substantially, even with an unchanged business outlook. We also monitor regulation, AI economics and capabilities, Micron’s execution, and competing technologies.

Conclusion & next steps