Micron Technology
Last reviewed
Jun 7, 2026
Sources
20 citations
Review status
Source-backed
Revision
v1 · 2,451 words
Improve this article
Add missing citations, update stale details, or suggest a clearer explanation.
Suggest edit
CtrlK
Last reviewed
Jun 7, 2026
Sources
20 citations
Review status
Source-backed
Revision
v1 · 2,451 words
Add missing citations, update stale details, or suggest a clearer explanation.
Micron Technology, Inc. (NASDAQ: MU) is an American memory and storage chipmaker headquartered in Boise, Idaho. Founded in 1978, it is the only major United States-based maker of memory chips and one of the three companies, alongside South Korea's Samsung Electronics and SK Hynix, that together dominate the global market for DRAM and NAND flash. Micron's two core products, dynamic random-access memory and NAND flash storage, sit underneath almost every computer, phone, and server. Since 2023 the company has moved to the center of the artificial intelligence buildout through high-bandwidth memory (HBM), the stacked DRAM that surrounds AI accelerators such as Nvidia data-center GPUs. By early 2026, surging demand for HBM and high-capacity server memory had pushed Micron to record revenue and profit, with its output sold out well into the year.
| Field | Detail |
|---|---|
| Type | Public company (NASDAQ: MU) |
| Founded | 1978, Boise, Idaho |
| Founders | Ward Parkinson, Joe Parkinson, Dennis Wilson, Doug Pitman |
| Headquarters | Boise, Idaho, United States |
| CEO | Sanjay Mehrotra (since 2017; board chair since 2024) |
| Employees | About 53,000 (2025) |
| Fiscal year | Ends in late August |
| Core products | DRAM, NAND flash, HBM, SSDs, low-power memory modules |
| Consumer brand | Crucial |
| FY2025 revenue | $37.38 billion |
| FY2025 GAAP net income | $8.54 billion |
Micron began in 1978 as a four-person semiconductor design firm working out of the basement of a Boise dental office, started by Ward and Joe Parkinson, Dennis Wilson, and Doug Pitman with backing from local Idaho investors. It shifted from design to manufacturing in the early 1980s, completing its first chip fabrication plant in Boise around 1981, and built its business on commodity DRAM through the brutal boom-and-bust cycles that have defined the memory industry for four decades.
Over time Micron grew largely by acquisition and survived a long consolidation that left only three large DRAM makers standing. It bought Lexar in 2006, the flash maker Numonyx in 2010 for roughly $1.27 billion in stock, Japan's Elpida Memory in 2013 for about $2 billion, and completed full ownership of Taiwan's Inotera Memories in 2016. The company sells semiconductor memory under its own name to manufacturers and cloud providers, and reaches consumers through the Crucial brand of memory modules and solid-state drives.
Sanjay Mehrotra, a co-founder and former chief executive of SanDisk, became Micron's chief executive in 2017 and added the role of board chair in 2024. He has steered the company toward higher-value products for data centers, a pivot that positioned it for the AI memory cycle. Micron's fiscal year ends in late August, so its quarterly reports run a calendar quarter ahead of most peers.
Micron sells two main classes of memory. DRAM is the fast, volatile working memory used by processors; it is the larger and more profitable side of the business, typically accounting for roughly three-quarters to four-fifths of revenue. NAND flash is the non-volatile storage used in SSDs and phones, and makes up most of the remainder. The company also produces NOR flash and a range of specialty and low-power parts, and it has pushed into newer module formats such as SOCAMM, a compact low-power DRAM module aimed at AI servers.
In fiscal 2026 Micron reorganized its reporting into four market-focused units that reflect where memory is now consumed: a Cloud Memory unit (HBM and high-capacity data-center modules), a Core Data Center unit, a Mobile and Client unit, and an Automotive and Embedded unit. The change underscored how central the data center had become. In the second fiscal quarter of 2026, the Cloud Memory unit alone generated $7.75 billion and Core Data Center another $5.69 billion, so data-center demand drove well over half of company revenue, with Mobile and Client at $7.71 billion and Automotive and Embedded at $2.71 billion.
High-bandwidth memory is the product that connects Micron most directly to AI. HBM stacks multiple DRAM dies vertically and links them to a processor through a very wide interface, delivering far more memory bandwidth than standard DRAM. That matters because large AI models are frequently limited not by raw compute but by how fast data can move between the accelerator and its memory, a constraint often called the memory wall. Each Nvidia data-center GPU ships with several HBM stacks, so HBM supply has become a gating factor for AI hardware production.
Micron entered the modern HBM race comparatively late, ramping its first high-volume product, HBM3E, rather than competing hard at earlier generations. Its HBM3E 8-high 24 GB part reached mass production in February 2024 and was qualified for Nvidia's H200 accelerator, with the same 24 GB stack designed into the HGX B200 and GB200 NVL72 Blackwell systems. A denser 12-high 36 GB version, offering about 50 percent more capacity at lower power per stack and more than 1.2 TB/s of bandwidth, began ramping in 2025 and was designed into the HGX B300 NVL16 and GB300 NVL72 platforms. As early as 2024 Micron said its HBM was sold out for that year and most of 2025.
The next generation, HBM4, arrived in 2026. On March 16, 2026, at Nvidia's GTC conference, Micron announced it was in high-volume production of HBM4 designed for Nvidia's Vera Rubin platform, the successor to Blackwell. Micron said its HBM4 36 GB 12-high part runs at pin speeds above 11 Gb/s, delivers more than 2.8 TB/s of bandwidth, and improves bandwidth by 2.3 times and power efficiency by more than 20 percent over HBM3E. It also disclosed samples of a 48 GB 16-high stack, using 16-die stacking to add 33 percent more capacity per placement, plus a 192 GB SOCAMM2 low-power module and a PCIe Gen6 SSD for AI systems.
| Product | Capacity (stack) | Key specifications | Target platform | Status |
|---|---|---|---|---|
| HBM3E 8-high | 24 GB | More than 1.2 TB/s per stack | Nvidia H200, HGX B200, GB200 NVL72 | Mass production from February 2024 |
| HBM3E 12-high | 36 GB | About 50% more capacity, lower power | Nvidia HGX B300 NVL16, GB300 NVL72 | Ramp from 2025 |
| HBM4 12-high | 36 GB | Over 11 Gb/s pins, more than 2.8 TB/s, 2.3x bandwidth and 20%+ power efficiency vs HBM3E | Nvidia Vera Rubin | Volume shipments from Q1 2026 |
| HBM4 16-high | 48 GB | 16-die stack, 33% more capacity than 36 GB part | Next-generation accelerators | Sampling (2026) |
HBM has gone from a rounding error to a profit engine. Micron has said HBM was less than 5 percent of its DRAM revenue in 2022 and that it expects HBM to exceed 30 percent of DRAM revenue in 2026. Independent analysis put Micron's HBM revenue on an annualized run rate of roughly $8 billion exiting fiscal 2025. On its fiscal Q2 2026 call, management said it expected to sell out HBM capacity for all of 2026 within months, leaving large buyers such as Nvidia, Microsoft, and Meta competing for allocation.
The AI cycle has lifted far more than HBM. Building HBM consumes a disproportionate amount of wafer capacity, which has tightened supply of ordinary DRAM at the same moment that AI servers demand record amounts of high-capacity DDR5. The result has been one of the sharpest memory shortages in years. Research firm TrendForce estimated that DRAM contract prices rose roughly 90 to 95 percent in the first calendar quarter of 2026 compared with the prior quarter, and analysts described the squeeze as structural rather than a normal cyclical swing, with data center buyers absorbing the bulk of new supply.
For Micron, the combination of HBM, high-capacity server modules, and rising prices produced an extraordinary swing in profitability. Gross margin, which had been deeply negative during the 2023 downturn, climbed back above 70 percent by early 2026 as scarce memory commanded premium pricing.
Memory is capital-intensive, and Micron has announced one of the largest industrial investment programs in the United States. In June 2025 it said it would expand its planned US spending to about $200 billion, split between roughly $150 billion in manufacturing and $50 billion in research and development, a program it projected would create around 90,000 direct and indirect jobs over time.
The plan centers on three states. In Boise, Idaho, Micron is building new high-volume fabs co-located with its R&D, with first DRAM output expected around 2027. In Clay, New York, near Syracuse, it has committed to a megafab complex of up to four plants built out over more than 20 years. In Manassas, Virginia, it is expanding production of longer-lived memory for automotive, industrial, and networking customers. Under the federal CHIPS Act, the Department of Commerce in December 2024 awarded Micron about $6.2 billion to support the Idaho and New York fabs, and finalized a further $275 million for the Virginia site in June 2025.
The schedule has shifted with demand. In late 2025 Micron said it would reallocate part of its CHIPS funding to accelerate construction in Boise while pushing the commissioning of its first two New York fabs out by roughly two to three years, prioritizing the capacity that could come online soonest. Capital expenditure rose accordingly: net capital spending was $5.0 billion in fiscal Q2 2026 alone, and the company guided full-year fiscal 2026 capex well above its earlier plan as it raced to add HBM and DRAM capacity.
Micron's results through fiscal 2025 and into fiscal 2026 set successive records. Fiscal 2025, which ended on August 28, 2025, brought revenue of $37.38 billion, up from $25.11 billion a year earlier, with GAAP net income of $8.54 billion and operating cash flow of $17.53 billion. Data center reached 56 percent of total revenue for the year. Momentum then accelerated: fiscal Q1 2026 revenue of $13.64 billion (reported December 17, 2025) gave way to a fiscal Q2 2026 result, for the quarter ended February 26, 2026, of $23.86 billion, up 196 percent year over year and 75 percent sequentially, with GAAP net income of $13.79 billion. For the following quarter the company guided to revenue of about $33.5 billion and a non-GAAP gross margin near 81 percent.
| Period | Revenue | GAAP net income | GAAP diluted EPS |
|---|---|---|---|
| FY2025 (ended Aug 28, 2025) | $37.38 billion | $8.54 billion | n/a |
| Q1 FY2026 (reported Dec 17, 2025) | $13.64 billion | $5.24 billion | $4.60 |
| Q2 FY2026 (ended Feb 26, 2026) | $23.86 billion | $13.79 billion | $12.07 |
| Q3 FY2026 (company guidance) | $33.5 billion (+/- $0.75B) | n/a | $19.15 non-GAAP (+/- $0.40) |
In the HBM market that now drives these numbers, Micron sits third behind a clear leader. Industry estimates put SK Hynix at around 60 percent of HBM bit shipments through 2025, with Micron near 21 percent, up from roughly 4 percent a year earlier, and Samsung in a similar range to Micron after losing ground. SK Hynix, which has been Nvidia's primary HBM supplier, also overtook Samsung as the world's largest memory supplier by revenue during the cycle. Micron's pitch is less about volume than positioning: it argues its HBM3E and HBM4 lead on power efficiency, a meaningful edge when data-center operators are constrained by electricity and cooling, and it has been adding customers as it expands capacity.
Micron's strengths are its status as the only US-headquartered large memory maker, its leading-edge DRAM process technology, and deep design-in relationships with Nvidia and the major cloud providers. Its risks are equally clear: memory remains a cyclical, capital-hungry commodity business exposed to a possible AI digestion phase, it trails SK Hynix in HBM share, and much of its profit now depends on a handful of large AI buyers and on prices that history says will not stay elevated forever. Its central role in AI chip supply, and the fact that memory bandwidth is now a primary bottleneck for AI compute, have nonetheless made Micron one of the most strategically important suppliers in the industry, closely watched alongside foundry partner TSMC and its Korean rivals.