How to read this page. Every maker in our Memory section uses the same 16 sections in the same order, the same shape as our AI Chips brand guides. Each section starts Simple, then goes Deeper, then Expert: stop wherever you have what you need. The small numbers are sources: click one to open the original document. Where a company has not published something, we say so rather than estimate.
1.At a glance
SimpleStart here
Micron is the third of the three companies that make most of the world's memory, and the only one headquartered in the United States 1. It makes DRAM, the fast working memory, NAND flash, the storage that survives a power cut, and HBM, the stacked memory AI accelerators need 2.
What it makes"the only company manufacturing today's major memory and storage technologies: DRAM, NAND, and NOR technology"
2
How big$41.46 billion of revenue in one quarter, Q3 FY2026
3
Who makes itIts own fabs in the US, Japan, Taiwan, Singapore and China
4
IndiaThe only big-three memory maker with a plant in India, at Sanand, Gujarat
5
DeeperThe detail
Read Micron's numbers carefully, because its calendar is unusual: "Our fiscal year is the 52- or 53-week period ending on the Thursday closest to August 31" 1. Its FY2026 therefore ran through most of calendar 2026. In FY2025 Micron reported "Revenue of $37.38 billion versus $25.11 billion for the prior year" and "GAAP net income of $8.54 billion" 6, split as "Total reported DRAM revenue was $28.58 billion in 2025" and "Total reported NAND revenue was $8.50 billion in 2025" 7.
ExpertFor specialists
Then FY2026 happened. Q3 alone, ended 28 May 2026, brought "Revenue of $41.46 billion versus $23.86 billion for the prior quarter and $9.30 billion for the same period last year", with "GAAP net income of $28.24 billion, or $24.67 per diluted share" 3 — one quarter larger than the whole of FY2025. Micron guided Q4 to "$50.0 billion ± $1.0 billion" at a gross margin around 86% 3. Those Q4 and full-year figures were due on 30 September 2026 8, four days after this check, so Q3 is the latest reported quarter here.
2.Company card
DeeperThe detail
Micron publishes an unusual amount about itself: "15 manufacturing sites and 13 customer labs", "21 design center locations", "30+ major cities worldwide" and "62,000 patents granted and growing" 2. It reports through "four business units, which are our reportable segments: Cloud Memory Business Unit (CMBU), Core Data Center Business Unit (CDBU), Mobile and Client Business Unit (MCBU), and Automotive and Embedded Business Unit (AEBU)" 7 — a structure that puts AI memory in a unit of its own.
ExpertFor specialists
Being US-listed is why this page can be more specific about Micron than about its Korean and Japanese rivals. Micron files 10-Ks and 10-Qs with the SEC, so DRAM and NAND revenue are broken out by line 7, quarterly figures are audited and public 1, and material commitments arrive as 8-Ks 10. Samsung reports memory profit only at division level and SK hynix names no customers; Micron's filings make the same kinds of questions answerable from primary documents.
3.The story
SimpleStart here
- 1978 "Micron Technology Inc., is founded on Oct. 5, 1978" — in a dental office basement 9.
- 1981 "Micron Ships its First 64K DRAM Product" 9.
- 1984 Becomes publicly traded on NASDAQ 9.
- 1999 "Micron Produces Industry's First DDR DRAM" 9.
- 2013 Acquires "Elpida Memory Inc. and Rexchip Electronics Corporation" 9.
- 2022 "Micron Ships World's First 232-Layer NAND" 9 11.
- 2026 Opens India's first semiconductor assembly and test plant 5.
DeeperThe detail
Micron grew as much by buying as by building. Its timeline records the Intel NAND joint venture IM Flash Technologies in 2005, Lexar Media in 2006, Numonyx from Intel and STMicroelectronics in 2010, Elpida and Rexchip in 2013, and the acquisition of Taiwan's Inotera Memories in 2016 9 — a company Micron had co-founded with Nanya as a joint venture eight years earlier. Micron's SEC filing dates the Inotera completion precisely, to 6 December 2016 12.
ExpertFor specialists
The timeline also shows what Micron chooses to mark as a first. It records HBM3E sampling in 2023 and "First production-capable HBM3E 12-high" in 2024 9 — but no entry for its first HBM product of any generation, and no official Micron statement found in this check dates one. Where Samsung's timeline omits HBM entirely, Micron's begins the HBM story at HBM3E, the generation where it became competitive. Neither company's history page is a neutral record of when the technology arrived.
4.The lineup
DeeperThe detail
The module end of the range is where Micron has been loudest. Its 512GB DDR5 RDIMM, announced September 2026, is described as the "world's first ultra-dense module", running "up to 9,200 MT/s" and enabling 12TB in a 24-slot dual-socket server, cutting operating power "by more than 60%" against four 128GB modules — 16.0W against 44.2W 14. SOCAMM2 is Micron's LPDRAM server module: 256GB per module, consuming "one-third of the power compared with equivalent RDIMMs" and improving "time to first token by more than 2.3 times" for long-context LLM inference 17.
ExpertFor specialists
Micron defines MRDIMM plainly, which is rarer than it should be: "a type of server memory module that uses an on-DIMM multiplexer to manage data across multiple memory ranks", reaching up to 8,800 MT/s for "39% increase in bandwidth" over a DDR5 RDIMM at 6,400 MT/s, with "up to 40% lower latency" 15. On the mobile and client side, Micron's LPCAMM2 module runs "Up to 9,600 MT/s with LPDDR5X" 18. On graphics memory, GDDR7 runs at a "32 Gb/s data rate" for "system bandwidth greater than 1.5 TB/s" on a 384-bit bus, built on 1β, using PAM3 signalling 19. Note that Micron's GDDR7 page does not state a production status 19, and this check found no Micron announcement of a shipping LPDDR6 part — only its participation in the JEDEC standard 20.
5.How it’s made
SimpleStart here
Micron names its DRAM processes with Greek letters and its NAND by generation number.
- DRAM: the ladder is "1α (1-alpha), 1β (1-beta), and 1γ (1-gamma)" 21.
- EUV: 1-gamma "utilizes extreme ultraviolet (EUV) lithography" 21 — Micron says so outright.
- NAND: the current generation is "9th-generation (G9) 3D NAND", in volume production since July 2024 22 16.
DeeperThe detail
1-gamma is the current leading DRAM node: Micron claims ">30% improvement in bit density per wafer versus 1β", speeds "up to 9200MT/s – 15% faster than relative 1β products" and DDR5 power "up to 20%" lower 21. It was first sampled in February 2025, when Micron said it was "the first in the industry to ship samples of its 1γ (1-gamma), sixth-generation (10nm-class) DRAM node-based DDR5 memory designed for next-generation CPUs to ecosystem partners and select customers" 23. The older 1α node is still in production and has a specific job: Micron calls it "the most advanced memory ever produced in the United States", used for DDR4 and LP4 in long-lifecycle automotive, defence, aerospace, industrial and medical parts 24.
ExpertFor specialists
On NAND layer counts, Micron is usually the most forthcoming of the three — it titled a 2022 release "Micron Ships World's First 232-Layer NAND" 11. That makes an omission conspicuous: neither the G9 volume-production release nor the G9 product page states a layer count 16 22. The widely-quoted "276-layer" figure for G9 appears only in press coverage, not in any Micron document found here, so this page does not state it. What Micron does claim for G9 is "the industry's fastest NAND I/O transfer rate, 3.6 gigabytes per second (GB/s)" 22 and density "up to 73% denser than competitive technologies" 16.
6.Where it’s made: fabs and addresses
SimpleStart here
Micron publishes full street addresses for its plants — the only big-three memory maker that does so this completely 4.
DeeperThe detail
The US build-out is the largest single commitment. In June 2025 Micron put it at "$150 billion in domestic memory manufacturing" plus "$50 billion in R&D", with Mehrotra saying "This approximately $200 billion investment will reinforce America's technological leadership, create tens of thousands of American jobs" 10. In Idaho, Micron plans "two leading-edge high-volume fabs", with "DRAM output scheduled to begin in 2027" at the first of them 25. In New York, it broke ground at Clay, Onondaga County in January 2026 on a "$100 billion complex with up to four semiconductor fabrication plants", where "Production is expected to start in 2030" 29.
Asia is expanding at the same time: Singapore has an HBM advanced packaging plant at "approximately US$7 billion (SG$9.5 billion)" 27 and a separate NAND wafer fab at "US $24 billion (SG $31 billion)" over ten years, with wafer output from the second half of 2028 30. Hiroshima was where Micron said it would be "the first semiconductor company to bring EUV technology to Japan for production", backed by "up to 500 billion yen" 26.
ExpertFor specialists
Two details reward attention. First, the CHIPS Act figure moves depending on who is stating it and when: the Department of Commerce announced "up to $6.165 billion in direct funding" for Idaho and New York in December 2024 31, Micron's own June 2025 filing says "up to $6.4 billion" 10, and the current NIST page gives "$6.44 billion" 32. This page cites each to its own source rather than picking one. Second, Micron has left a fab as well as building them: it sold its Lehi, Utah plant to Texas Instruments, completed 22 October 2021 33.
NIST describes what the Idaho money buys: "two high-volume manufacturing (HVM) fabs, each with approximately 600,000 square feet of cleanroom space focused on the production of leading-edge DRAM chips" 32. The spending rate to match is visible in Micron's quarterly reporting: "investments in capital expenditures, net were $7.1 billion" in the third quarter of fiscal 2026 alone 3.
7.How the memory works
SimpleStart here
DRAM holds each bit as a charge in a capacitor that leaks, so it is refreshed constantly and forgets everything when power is cut. NAND traps charge in a cell that holds it without power. Micron makes both, plus NOR, a third and older flash type used in smaller embedded roles 2.
DeeperThe detail
HBM stacks DRAM dies vertically and wires them straight through, next to the processor. Micron's HBM4 runs a 2048-bit interface for ">2.0 TB/s per memory stack" in its first samples 34, rising to bandwidth ">2.8 TB/s" and pin speed "over 11 Gb/s" in the production part 13. Micron's consistent argument for its own HBM is power rather than peak speed: its first HBM3E was pitched at "~30% lower power" than competing HBM3E 35, and HBM4 at "20% better power efficiency than HBM3E" 13.
ExpertFor specialists
The economically interesting point is which node the HBM sits on. Micron built HBM4 on 1β, not its newest node 3, reserving 1γ — the EUV node 21 — for HBM4E, where "volume production expected in calendar 2027" 3. That is a deliberate staging: prove the stack on a mature node first, then move it to the newer process for the generation after. It also explains why Micron could reach high-volume HBM4 production while its newest node was still ramping elsewhere 13.
8.Specs explained
DeeperThe detail
Micron's module claims are usually stated against a named baseline, which makes them checkable. The 512GB RDIMM's power figure is not a percentage in isolation but 16.0W against 44.2W for the four-module alternative it replaces 14; MRDIMM's "39% increase in bandwidth" is specifically against "DDR5 RDIMM at 6400 MT/s" 15; SOCAMM2's "more than 2.3 times" improvement names the workload as long-context LLM inference 17. Where a comparison has no named baseline, this page says so rather than implying one.
ExpertFor specialists
Micron does publish some third-party validation of its module work. The 512GB RDIMM release carries quotes from both AMD and Intel — Amit Goel, Corporate Vice President at AMD, and Karin Eibschitz Segal, General Manager of Platform and Systems Engineering at Intel's Data Center Group 14 — which is a stronger signal than a specification sheet, since those are the companies whose processors must accept the module. Note the timing, though: the 512GB part is announced with "volume production 2H 2027" 14, so it is a roadmap item, not something shipping at this check.
9.Official pricing
SimpleStart here
Micron publishes no price for any memory chip. Buying runs through "Micron authorized sales representatives and distributors", with partners using a quote system rather than a price list 36.
And as of 2026, Micron no longer sells to consumers at all — see Hidden in plain sight 37.
DeeperThe detail
Until December 2025, Micron did have a consumer-facing brand with published prices: Crucial. On 3 December 2025 it "announced its decision to exit the Crucial consumer business, including the sale of Crucial consumer-branded products at key retailers, e-tailers and distributors worldwide", with shipments continuing "until the end of fiscal Q2 (February 2026)" 37. That removed the one place Micron published retail memory prices.
ExpertFor specialists
What partly replaces a price list, for Micron, is contract disclosure. It now sells a large share of its output under multi-year Strategic Customer Agreements, which Mehrotra says "will significantly enhance the durability and predictability of Micron's strong financial performance" 3. Micron has named the partners in one segment: Qualcomm, Visteon, HARMAN, JOYNEXT, DENSO, Astemo and Hyundai Mobis, all automotive 38. It has not published the prices in those agreements, and this page found no official Micron document stating the contracted value, so none is given here.
10.The HBM position
SimpleStart here
Micron's HBM4 is in high-volume production and, like its two rivals, it is "designed for NVIDIA Vera Rubin" 13. Micron got to HBM3E first in one specific sense: its 24GB 8-high part was named as going into "NVIDIA H200 Tensor Core GPUs" back in February 2024 35.
DeeperThe detail
The HBM3E announcement is unusually specific for this industry — it names the customer platform outright: "Micron's 24GB 8H HBM3E will be part of NVIDIA H200 Tensor Core GPUs, which will begin shipping in the second calendar quarter of 2024" 35. HBM4 samples went to "multiple key customers" in June 2025, unnamed 34, and by March 2026 the 36GB 12-high part was in high-volume production for Vera Rubin, with a 48GB 16-high version sampling 13. As of the Q3 FY2026 release, HBM4 "is in high-volume shipments for our lead customer's platform, and qualification samples have been shipped to multiple end-customers" 3.
ExpertFor specialists
Two cautions about what Micron has not said. It names no HBM4 customer other than NVIDIA; "multiple end-customers" is as specific as the qualification language gets 3. And no Micron release checked here states the company's HBM market share — a figure widely attributed to Micron in coverage, and absent from its own documents.
The frequently-repeated claim that Micron's HBM is "sold out", through 2026 or 2027, also goes beyond anything this check could source to a Micron document. Statements to that effect trace to earnings-call commentary reported by third parties rather than to Micron's own published releases, so this page does not state one. What Micron's own release does say is narrower: HBM4 is "in high-volume shipments for our lead customer's platform" 3.
11.Who buys it and why
SimpleStart here
Micron sells to the companies building servers, PCs, phones and cars. The customer it names most often is NVIDIA, for HBM3E in H200 and HBM4 and SOCAMM2 in Vera Rubin 35 13.
DeeperThe detail
Micron's four reportable segments map directly onto its customer types: cloud memory, core data centre, mobile and client, and automotive and embedded 7. Automotive is the segment where it has named the most customers, disclosing strategic agreements with Qualcomm, Visteon, HARMAN, JOYNEXT, DENSO, Astemo and Hyundai Mobis 38. In India, its first shipment from the new Sanand plant went to Dell Technologies, for laptops manufactured in India 5.
ExpertFor specialists
The most revealing customer statement is about who Micron chose not to supply. It exited the consumer market explicitly to free up capacity, making the decision "in order to improve supply and support for our larger, strategic customers in faster-growing segments" 37. In a market this tight, Micron's customer list is as much a statement about allocation as about sales.
12.Hidden in plain sight
SimpleStart here
Micron stopped selling memory to consumers in the middle of the biggest memory boom in its history. On 3 December 2025 it announced its exit from the Crucial consumer business — the brand it had sold DRAM modules and SSDs under for decades 37.
DeeperThe detail
The reasoning is stated plainly, and it is about scarcity rather than weakness: "The AI-driven growth in the data center has led to a surge in demand for memory and storage. Micron has made the difficult decision to exit the Crucial consumer business in order to improve supply and support for our larger, strategic customers in faster-growing segments" 37. Shipments ran "until the end of fiscal Q2 (February 2026)", with warranty service continuing and Micron-branded enterprise products unaffected 37. Sumit Sadana's send-off was for a brand, not a business line: "Thanks to a passionate community of consumers, the Crucial brand has become synonymous with technical leadership, quality and reliability of leading-edge memory and storage products" 37.
ExpertFor specialists
It is worth sitting with what that means. A memory maker posting $41.46 billion in a quarter 3 chose to stop serving the retail market entirely, because retail was consuming supply it could sell at better margins to data-centre customers 37. Anyone who has noticed memory prices climbing on the shelf is looking at the consumer-facing edge of the same decision — and Micron is the one company that wrote down why.
The exit also quietly removed the only place Micron published a retail memory price. Component prices were never public — buying runs through "Micron authorized sales representatives and distributors" 36 — so with Crucial gone, Micron now publishes no memory price of any kind.
13.Weak spots
SimpleStart here
- It gave up a whole market to free capacity — the Crucial consumer exit 37.
- It is spending enormously — $7.1 billion of net capital expenditure in a single quarter 3.
- Its largest new fabs produce years out — Idaho from 2027, New York from 2030 25 29.
DeeperThe detail
The build-out is the risk that comes with the boom. Micron has committed to approximately $200 billion in the US 10, US$24 billion for one Singapore fab 30, US$7 billion for another 27, and up to 500 billion yen in Japan 26 — while its largest new US fabs do not produce until 2027 in Idaho and 2030 in New York 25 29. Memory is a cyclical industry, and that capacity arrives years after the demand that justified it.
ExpertFor specialists
Disclosure gaps are narrower here than at Samsung or SK hynix, but they exist and cluster around the same places. Micron publishes no layer count for G9 NAND despite having titled a release around 232 layers two years earlier 16 11; no HBM market share in any release checked here; no HBM4 customer beyond NVIDIA 3; and no production status on its GDDR7 page 19. This check also found no Micron release for its reported Xi'an expansion and none for a 2026 Hiroshima expansion, so neither is stated here, although the 2023 Hiroshima EUV commitment is officially documented 26.
14.India angle
SimpleStart here
Micron is the only one of the three big memory makers with a plant in India. Its assembly and test facility at Sanand, Gujarat was inaugurated on 28 February 2026 by Prime Minister Narendra Modi 39 5.
An important distinction: this is not a wafer fab. Micron's own description is that the plant "converts advanced DRAM and NAND wafers from Micron's global manufacturing network into finished memory and storage products" 5. The wafers are made elsewhere; India does the packaging and testing.
DeeperThe detail
The money is public and split three ways. Micron announced the project on 22 June 2023, stating that "Micron's investment will be up to $825 million over the two phases of the project and will create up to 5,000 new direct Micron jobs and 15,000 community jobs over the next several years" 40. The rest comes from government: "Under the government's 'Modified Assembly, Testing, Marking and Packaging (ATMP) Scheme,' Micron will receive 50% fiscal support for the total project cost from the Indian central government and incentives representing 20% of the total project cost from the state of Gujarat" 40. Total across both phases: "up to $2.75 billion" 40.
At the opening, Micron reported the plant already in commercial production, ISO 9001:2015 certified, with its "first shipment of made-in-India memory modules to Dell Technologies" for laptops built in India 5. Output is expected to scale from "tens of millions of chips in 2026" to "hundreds of millions in 2027" 5, from a phase-one cleanroom of "more than 500,000 square feet" 5.
ExpertFor specialists
Read the two official releases together and a schedule slip is visible. The 2023 announcement had phase one that "will start to become operational in late 2024" 40; commercial production actually began around February 2026 5 — roughly a year later than planned. India's own government frames the same period as fast, recording that "The Memorandum of Understanding was signed in June 2023, followed by the groundbreaking in Sanand in September 2023", leading "to the commencement of commercial production in February 2026" 39. Both are fair: slow against Micron's original target, quick for a first-of-its-kind plant in a country with no prior semiconductor assembly industry.
Two further cautions. The 5,000-direct-jobs figure is a 2023 forward-looking target; the 2026 opening release gives no actual headcount, so this page does not present it as current employment 40 5. And Micron's India presence is broader than the plant — it lists offices in Hyderabad and Bengaluru 4 — but this check found no Micron-hosted release detailing the Hyderabad development centre, so no investment figure or headcount for it is stated here.
15.What’s next
SimpleStart here
Only what Micron has announced:
- HBM4E: on 1-gamma, "volume production expected in calendar 2027" 3.
- Idaho: DRAM output scheduled to begin in 2027 25.
- New York: "Production is expected to start in 2030" 29.
- Singapore: NAND wafer output from 2H 2028 30.
- 512GB RDIMM: volume production 2H 2027 14.
DeeperThe detail
Micron's own framing of why it is spending at this scale is the durability of its contracts rather than today's price: Mehrotra points to "multi-year Strategic Customer Agreements" that "will significantly enhance the durability and predictability of Micron's strong financial performance" 3. That is an argument about visibility, not about the spot market — and it is the clearest statement Micron gives of why it is committing capital at this scale into a historically cyclical industry 10.
ExpertFor specialists
The nearest test is arithmetic. Micron guided Q4 FY2026 to "$50.0 billion ± $1.0 billion" with gross margin around 86% and GAAP diluted EPS of "$30.73 ± $1.00" 3, and was due to report on 30 September 2026 8 — four days after this page was checked. Whether that guidance held is knowable now and was not knowable here, which is exactly the kind of gap this page marks rather than papers over.