Nanya Technology Corporation (2408.TW)

Deep Dive Research Report

Prepared April 2026. All information sourced from public filings, earnings call transcripts, company disclosures, and third-party research. No investment recommendation implied.

1. What the Company Does

Nanya Technology makes DRAM chips. Not NAND flash, not SRAM, not logic chips - just DRAM, and nothing else. Dynamic Random Access Memory is the short-term working memory inside virtually every computing device: the RAM in your laptop, the buffer in your router, the temporary storage in your car's infotainment system. Every time a chip needs to hold data for fast, live access rather than permanent storage, DRAM is the answer.

The company was founded in March 1995 in New Taipei City, Taiwan, as a spin-out from the Formosa Plastics Group - one of Taiwan's largest industrial conglomerates, rooted in petrochemicals and plastics. The decision to enter DRAM was deliberate and arguably prescient: Nan Ya Plastics, the parent entity, saw the coming explosion of personal computing and bet that the chemical and manufacturing expertise it had built in plastics could transfer to the highly process-intensive world of semiconductor fabrication. In May 1994 - before Nanya Technology was even formally incorporated - the parent signed a technology licensing agreement with Japanese memory maker OKI, securing the process know-how to start producing 16Mb DRAM chips. The company broke ground on Fab 1 in Taoyuan in January 1995, produced its first pilot DRAM die in July 1996, and listed on the Taiwan Stock Exchange in August 2000.

That origin story is important because it explains Nanya's fundamental identity: a single-product industrial manufacturer, backed by one of Taiwan's most capital-rich conglomerates, competing in one of the world's most capital-intensive, technically demanding, and violently cyclical industries. The Formosa Plastics Group's backing has given Nanya access to patient capital that independent memory startups simply cannot access. But it has also meant the company operated for years as a commodity manufacturer, dependent on licensed process technologies from others, before it finally developed its own.

The pivot to independent process technology is the central drama of Nanya's recent history. For most of its existence, Nanya relied on technology transfer agreements - first from OKI, then from Infineon, then from Micron - to manufacture DRAM. The company's joint venture with Infineon (later absorbed by Qimonda) created Inotera in 2003; when Qimonda collapsed in 2009, Nanya partnered with Micron instead. By 2016, Micron bought out the entire Inotera venture for NT$47.62 billion, and Nanya walked away with both cash and - critically - the right to license two of Micron's 10nm-class DRAM process generations. This transaction gave Nanya the seed of its current technology independence. The company used those licenses as the foundation to build its own 1A-nm process technology in 2020-2021, then its proprietary 1B-nm node by 2023-2024.

The product in action looks like this: a module supplier building DDR4 SODIMM sticks for laptop manufacturers places a purchase order with Nanya for 16Gb DRAM dies in bulk. Nanya's Fab 5A in Taoyuan processes 300mm silicon wafers through hundreds of steps over several weeks, using photolithography, deposition, etching, and ion implantation to create memory cells at the 1B-nanometer process node. After wafer fabrication, dies are tested, packaged into individual chips, shipped to the module maker, and soldered onto a PCB alongside other chips to form the finished memory stick. The whole pipeline from wafer start to shipped die takes roughly six to eight weeks.

Chairman Chia-Chau Wu captured what this company aspires to be when announcing the new fab investment in 2021: "DRAM has become a critical component for the smart world." The statement is less grandiose than it sounds. In an era of AI servers, AI smartphones, AI cars, and AI cameras, every single device that processes data needs more memory than the previous generation. Nanya's bet is that its position as a pure-play conventional DRAM supplier - with no distraction from NAND flash or HBM GPU memory - allows it to capture that secular growth while the industry giants get drawn upmarket toward higher-margin but more technically demanding High Bandwidth Memory.

2. Business Segments

Nanya Technology operates as a single-product business. In its statutory filings, the company discloses four segments: a Manufacture segment (the Taiwan fabs, where chips are physically produced), a Japan segment (the Japanese sales subsidiary), a United States segment (the US sales subsidiary), and an Other segment (minor activities). But these are geographic and functional subdivisions of one unified DRAM business, not distinct product lines. There is no meaningful segmentation by end market, product type, or technology generation in the financial statements. The Manufacture segment generates essentially all revenue; the Japan and US segments are distribution arms.

Given this structure, the segment treatment in this report is handled differently. Rather than write separate sections for what are effectively sales offices, the discussion below organises Nanya's business along the two dimensions that actually matter for understanding it: product type and application market.

Standard DRAM (DDR3/DDR4/DDR5)

This is Nanya's largest business and the segment currently under the spotlight. Standard DDR (Double Data Rate) DRAM is the dominant memory type in desktop PCs, laptops, servers, networking equipment, and consumer electronics. Within this category:

DDR4 is Nanya's commercial centre of gravity. As of Q4 2025, DDR4 and Low-Power DDR4 combined accounted for roughly 70% of bit shipments. DDR4 is a product Nanya has produced since 2017, and the entire installed base of devices from roughly 2015 to 2022 uses DDR4. Because the major DRAM producers - Samsung, SK Hynix, and Micron - are aggressively converting capacity toward DDR5 and HBM, DDR4 supply is tightening structurally even as DDR4 devices remain the largest portion of the global installed base. Nanya's 1B-nm node produces DDR4 chips competitively.

DDR5 is Nanya's strategic growth vector. As of Q4 2025, DDR5 represented roughly 10% of bit shipments, and management has explicitly stated its intention to grow DDR5 at the expense of DDR3 rather than DDR4 - recognising that DDR4 pricing is currently stronger than DDR3 and should be preserved. Nanya's DDR5 product at 16Gb density running at 5600 Mbps speed is in volume production; the 6400 Mbps variant was in engineering sampling as of mid-2025 and achieved design specification targets. A 128GB RDIMM module using Nanya's DDR5 die achieved functional testing milestones by Q4 2025 at both 5600 and 6400 speeds, with mono-die performance reaching 7200 Mb/s, positioning the company for the server memory market.

DDR3 is the legacy tail. Older consumer devices, industrial controllers, and certain embedded applications still use DDR3. Nanya maintains a small DDR3 business on its older process nodes, but management has explicitly targeted this as the segment to shrink first as DDR5 ramps up. DDR3 represented roughly 20% of shipments as of Q4 2025.

Low Power DRAM (LPDDR)

LPDDR (Low Power DDR) is the mobile-optimised version of DRAM, designed for smartphones, tablets, laptops, and increasingly automotive and IoT devices. The power consumption characteristics are fundamentally different - LPDDR chips use sophisticated power gating and partial array refresh to minimise idle draw, which matters enormously in battery-powered devices.

Nanya's current LPDDR product line runs from LPDDR2 (legacy, minimal production) through LPDDR4 and LPDDR4X (the current volume products), with LPDDR5 under development on the 1B process. The introduction of LPDDR5 products was targeted for 2026 in statements made across the 2025 concalls. LPDDR chips at high density - particularly 16Gb die - are increasingly relevant for AI smartphones that need more working memory for on-device inference.

The LPDDR segment is strategically important for two reasons. First, it diversifies Nanya away from pure commodity PC/server DRAM. Second, the automotive qualification pathway for LPDDR (and standard DRAM) is genuinely differentiated: automotive-grade DRAM requires AEC-Q100 qualification, extended temperature range operation, and longevity commitments that span vehicle lifetimes of 10-15 years. Nanya has been building this capability and offers automotive-grade variants with longevity supply plans - a stickier customer relationship than typical spot market DRAM sales.

Specialty Products: KGD, MCP, eMCP, and Modules

Known Good Die (KGD) are individual DRAM chips that have been fully tested and binned for quality, sold to customers who want to integrate DRAM directly onto custom system-in-package designs. This is the form factor used in high-density AI edge devices, advanced mobile processors, and certain IoT applications where the entire memory needs to be embedded.

Multi-Chip Package (MCP) and embedded MCP (eMCP) products combine DRAM and NAND flash in a single package - Nanya typically supplies the DRAM die, while a separate NAND die is sourced externally. These products are used in feature phones, entry-level smartphones, and certain IoT devices.

Memory modules - SODIMM (for laptops), UDIMM (for desktop PCs), RDIMM (for servers), and the newer LPCAMM2 (a compact, swappable low-power module for ultra-thin laptops) - represent Nanya's downstream integration. The Elixir brand is Nanya's consumer-facing memory module label, sold through retail and distribution channels.

3. Products and Business Detail

The Full Product Catalogue

Standard DRAM - DDR2/DDR3/DDR3L: Legacy products maintained for industrial, embedded, and long-lifecycle applications. DDR3L (low-voltage variant) remains in production for certain networking and embedded markets. Nanya maintains these partly for the longevity supply agreements with industrial customers.

Standard DRAM - DDR4: The core business. Produced at the 1A and 1B process nodes. Available in 8Gb, 16Gb densities. Speeds from 2133 to 3200 Mbps. Used in virtually every PC, laptop, server, and networking device built from 2015 to present. This is the product in structural shortage as of 2025-2026 because Samsung, SK Hynix, and Micron are de-emphasising it in favour of DDR5 and HBM.

Standard DRAM - DDR5: Produced on the 1B process at 16Gb density. Speeds targeting 5600 and 6400 Mbps in volume. RDIMM and UDIMM module configurations being developed for server applications. This is Nanya's growth vector for the server market, where DDR5 is the standard going forward.

Low Power DRAM - LPDDR4/LPDDR4X: Current production volume for mobile and automotive applications. 8Gb, 16Gb densities. The LPDDR4X variant runs at lower voltage (1.1V vs 1.1V-1.2V), further reducing power consumption for premium mobile devices.

Low Power DRAM - LPDDR5/LPDDR5X: Under development on the 1B process, targeted for introduction in 2026. LPDDR5 is the current standard in flagship smartphones and will be required for AI-capable mobile devices.

KGD (Known Good Die): Fully tested bare dies for system-in-package integration. Supplied to module manufacturers, AI device makers, and automotive tier-1 suppliers.

MCP/eMCP: Combined DRAM+NAND packages for entry-level mobile devices and IoT. A commoditising product line with limited growth.

Elixir Brand Modules: Consumer memory modules (SODIMM, UDIMM) sold under the Elixir brand through distribution. This is primarily a volume channel for laptop and desktop upgrades, with modest differentiation.

RDIMM (Registered DIMM) for Servers: A strategic push product as of 2025-2026. Nanya's 128GB DDR5 RDIMM at 5600/6400 speeds achieved functional qualification milestones in Q4 2025. Server RDIMM requires not just the DRAM die but register chip integration, thermal packaging, and validation against server platform firmware - a more complex product than standard UDIMM.

The Manufacturing Process

DRAM manufacturing is among the most process-intensive activities in global industry. A modern DRAM wafer runs through roughly 700-900 individual process steps over 6-8 weeks. The key technical steps are:

Capacitor formation: DRAM cells store binary data as charge on a capacitor. As process nodes shrink, capacitors must still hold sufficient charge - requiring ever-taller, narrower capacitor structures etched into silicon with extraordinary precision. At the 1B-nm node, these capacitors are formed using HAR (High Aspect Ratio) etching, where structures have height-to-diameter ratios exceeding 50:1.

Transistor formation: Each memory cell also contains a transistor that controls access to the capacitor. Shrinking transistors while maintaining acceptable leakage is one of the key challenges in DRAM scaling. Nanya's 1B process uses recessed-gate transistors to reduce leakage.

Lithography: Nanya currently uses ArF immersion lithography (not EUV) for its production nodes. The new fab being constructed includes a separate EUV-compatible building, suggesting future nodes (1D and beyond) will adopt EUV for the most critical layers.

Testing and die sorting: After wafer fabrication, each die is tested electrically. Dies that pass are sold as KGD or packaged; dies with limited defects may be sold as lower-density parts through repair fusing.

Fabrication Capacity

Fab 5A (primary): Located in Taoyuan, Taiwan. Monthly capacity of approximately 60,000 wafers (300mm). This is Nanya's sole volume manufacturing facility. As of 2025, this fab has completed its transition to the 1B/1C-nm node mix, running DDR4, DDR5, LPDDR4, and specialty products in parallel. At full utilisation (which has been the case since mid-2025), this fab generates all of Nanya's revenue.

New Fab (P3) - Under Construction: Located in Taishan, Nanling Technology Park, New Taipei City. Total investment: NT$300 billion across three phases over approximately seven years. Target capacity: 45,000 wafers/month in phase one, scaling to the same total. Equipment move-in: early 2027. Initial production: H2 2027. The facility includes a dedicated EUV bay for future process nodes. When fully ramped, this fab will roughly double Nanya's total capacity and increase bit output by approximately 120% given assumed process technology improvements.

Research & Development

Nanya's technology independence is a recent and hard-won achievement. After the Micron/Inotera deal gave it access to licensed 10nm-class processes, the company has invested heavily in developing its own process generations:

1A nm: First-generation 10nm class. Used the Micron-licensed technology as a starting point. In production from 2021.
1B nm: Second-generation, proprietary. Fully in production with "reasonable yield" as of Q4 2025. DDR4 and DDR5 products running on this node.
1C nm: Third-generation. Design completed, trial production targeted for early 2026, qualification by end of 2026. Management confirmed this is on schedule.
1D nm: Fourth-generation. Approximately one year behind 1C in the development pipeline.

The Kioxia collaboration on OCTRAM (Oxide-Semiconductor Channel Transistor DRAM) announced at IEDM in December 2024 represents Nanya's bet on what comes after conventional silicon DRAM. The technology uses InGaZnO (IGZO) vertical channel transistors with a 4F² architecture - a cell geometry that is roughly half the area of current DRAM cells - and achieves ultra-low leakage current (below 1 attoampere per cell), enabling retention times orders of magnitude longer than silicon DRAM. This collaboration positions Nanya alongside Kioxia for a potential technology inflection in the late 2020s.

Nanya also invested in PieceMakers Technology, a startup working on customised ultra-high-bandwidth memory architectures - a small strategic option on the AI custom memory market.

Geographic Sales and Distribution

Nanya operates sales offices in Taiwan (headquarters), Japan, the United States, Germany (European coverage), and China. Based on the company's global exposure, Asia (particularly China, Japan, and the broader Asia-Pacific region) represents the majority of sales volume. The China exposure is significant: China is the world's largest consumer of DRAM, importing substantial volumes for PC assembly, smartphone manufacturing, and server buildout. The US and European offices primarily serve networking, industrial, and server customers. Germany represents the automotive-grade DRAM pathway into European OEMs and tier-1 automotive suppliers.

4. Customers

Who Buys Nanya's DRAM

Nanya does not publicly name its specific end customers, but the structure of its customer base can be reasonably inferred from the product mix and distribution channel.

Consumer electronics OEMs and ODMs: PC and laptop manufacturers (OEMs) and their contract manufacturers (ODMs) are likely the largest category. When a laptop manufacturer sources DDR4 SODIMM components, they may buy from Nanya directly or through a module manufacturer who buys Nanya's raw die. Consumer electronics - televisions, set-top boxes, gaming consoles, home networking gear - represents another significant slice. Management noted in the Q3 2025 concall that consumer applications represented roughly 60% of sales.

Module manufacturers and distributors: A significant portion of Nanya's output goes through module makers who combine Nanya dies with PCBs and passives to create finished DIMM sticks. These are sold under both the Elixir brand (Nanya's own) and third-party module brands. Distributors like WPG Holdings in Taiwan act as intermediaries, particularly for the industrial and specialty market.

Networking and communications equipment makers: Routers, switches, and base stations require substantial DRAM. Nanya's DDR4 and specialty products address this market.

Industrial and automotive customers: A smaller but strategically important segment. Automotive customers - through tier-1 suppliers - require AEC-Q100 qualified parts with longevity supply commitments. These customers run qualification processes lasting 12-18 months and then commit to multi-year purchase agreements. The switching cost is genuinely high: a car model that qualifies a specific DRAM part is locked into that part for the vehicle's production life. Nanya explicitly mentions automotive and industrial applications in its product roadmap.

AI edge and server customers: As of 2025, Nanya's server business remained in the "single digit percentage" of revenue (per the Q1 2025 concall), but management has targeted server RDIMM as a growth market. The 128GB DDR5 RDIMM qualification milestone in Q4 2025 was explicitly described as positioning the company for cloud and enterprise server customers. The custom AI projects mentioned in Q4 2025 - described as ahead of schedule with revenue expected in H1 2026 - suggest at least some cloud or AI hardware customers are pursuing Nanya as a supply source.

The $2.5 billion private placement cohort: In March 2026, SanDisk ($1B), Kioxia ($500M), Solidigm ($500M), and Cisco ($500M) each took equity stakes in Nanya and signed multi-year DRAM supply agreements. This fundamentally changes the customer concentration picture. These are not arm's-length spot buyers; they are strategic investors who have locked in supply at committed pricing. SanDisk needs DRAM to bundle with its NAND in SSDs. Kioxia, similarly, requires DRAM as a co-component in storage solutions. Solidigm (SK Hynix's Intel NAND business) has the same need. Cisco needs DRAM for its networking hardware. Each investor's equity stake aligns their interest with Nanya's success in a way that a pure supply contract does not.

Why Customers Buy from Nanya

The decision to source DRAM from Nanya rather than exclusively from Samsung, SK Hynix, or Micron comes down to three considerations:

Supply diversification: No major buyer wants to be entirely dependent on the three Korean/US giants. Nanya offers a fourth source of standard DRAM at competitive specifications. The 2025-2026 DRAM shortage made this particularly acute: as the big three diverted capacity to HBM, buyers who had not locked in alternative supply faced spot market premiums. The SanDisk/Kioxia/Solidigm investment was explicitly a supply diversification play.

Specialisation in conventional DRAM: Nanya is a pure-play conventional DRAM maker. It is not distracted by HBM development, NAND flash, or CMOS image sensors. For buyers who need large volumes of DDR4, DDR5, or LPDDR, Nanya's focused capacity is a feature, not a limitation.

Longevity and specialty support: For industrial and automotive customers who need 10-15 year supply commitments, Nanya is a more willing partner than Samsung or SK Hynix, whose commercial priorities may shift away from legacy products. The ability to commit to DDR3 supply "until further notice" for industrial controllers is genuinely differentiated.

Switching Costs

The switching cost landscape differs sharply by customer type.

For consumer and PC OEM customers buying through distributors, switching costs are low. A DDR4 chip from Nanya and a DDR4 chip from Micron are functionally interchangeable at the same specifications. Buyers choose on price, lead time, and relationship.

For industrial and automotive customers, switching costs are high and structural. Automotive qualification under AEC-Q100 requires stress testing (temperature cycling, humidity, ESD), lifetime testing, and documentation of the entire supply chain. Qualifying a new DRAM supplier can take 12-18 months. Once qualified and designed into a vehicle platform, that part number stays in the bill of materials for the vehicle's production life. This creates the kind of lock-in that makes automotive DRAM a strategically valuable, if smaller, market.

For the strategic investor cohort (SanDisk, Kioxia, etc.), the equity relationship and multi-year supply contracts create a hybrid lock-in: these customers are now financially motivated to maintain the relationship and have contractually committed volumes.

Contract Structure

Nanya has explicitly moved away from spot market sales toward monthly and quarterly contracts. This shift was described in multiple concalls through 2025. The Q3 2025 concall noted that customers were moving toward quarterly and even annual agreements, replacing spot market transactions. This is partly market-driven (customers securing supply amid shortages) and partly Nanya's intentional strategy to reduce revenue volatility. As of Q4 2025, spot sales were described as "minimal." The new supply agreements with the strategic investors are structured as multi-year arrangements.

5. Competitive Landscape

The Structure of DRAM Competition

DRAM is one of the most concentrated industries in global semiconductors. Three companies - Samsung Electronics, SK Hynix, and Micron Technology - control roughly 93-95% of global DRAM revenue. This oligopoly is not accidental; it is the result of 30 years of brutal consolidation driven by the economics of DRAM manufacturing. Building a world-class DRAM fab costs US$10-15 billion. The process technology takes 5-10 years to develop and yields improve only through continuous production learning. Any company that cannot sustain the capital cycle across down-cycles gets absorbed or exits. The last major DRAM player to exit was Qimonda (the rebranded Infineon memory division) in 2009.

Within this structure, Nanya occupies a small but distinct fourth-tier position.

The Big Three and Their Current Priorities

Samsung Electronics: Historically the world's largest DRAM maker and the technological pioneer of DRAM scaling. In 2025, SK Hynix surpassed Samsung in DRAM revenue for the first time in 33 years, largely because Samsung was slower to commercialise HBM3E for AI accelerators. Samsung holds approximately 32-33% of global DRAM market share. The company's technology roadmap includes bleeding-edge process nodes (1-gamma class), EUV deployment across multiple critical layers, and HBM4 development. Relative to Nanya's conventional DRAM focus, Samsung is simultaneously a massive competitor in DDR4/DDR5 and a company increasingly diverting capacity toward HBM - which indirectly tightens conventional DRAM supply and helps Nanya.

SK Hynix: The current market share leader in DRAM at approximately 33-36% globally. SK Hynix's dominance in HBM3E (it holds roughly 62% of the HBM market as of mid-2025) has made it the premier AI memory supplier. The company is investing enormous capital in HBM4 for the next generation. Like Samsung, its focus on HBM means its conventional DRAM capacity growth is limited - again creating tailwinds for Nanya.

Micron Technology: The US-based third player at approximately 26% global DRAM share. Micron has been aggressively closing the gap with SK Hynix in HBM. Importantly, Micron is also expanding its conventional DDR5 capacity for server applications and is the primary technical competitor to Nanya in the server DDR5 market. Micron's advantage over Nanya in DDR5 is substantial: Micron has multiple process generations ahead, larger production scale, and deeper customer relationships in the data center.

Taiwan-Based Competitors

Winbond Electronics (2344.TW): Winbond is a specialty DRAM maker focused on NOR flash and pseudo-SRAM as well as DDR3/DDR4. Its Kaohsiung fab is transitioning from 25nm to 20nm, adding approximately 15,000 wafers/month of DDR4 capacity. Winbond is smaller than Nanya (revenue roughly one-third of Nanya's) and less exposed to standard DRAM. The two companies do not directly compete on DDR5 server applications.

PSMC (Powerchip Semiconductor Manufacturing): PSMC is primarily a foundry but has a small self-produced consumer DRAM business. Its DRAM revenue is approximately one-twentieth of Nanya's, making it a marginal competitor.

Chinese DRAM Entrants - The Key Strategic Threat

CXMT (ChangXin Memory Technologies): CXMT is China's primary DRAM manufacturer and represents the most significant structural threat to Nanya's medium-term position. CXMT has been aggressively expanding DDR4 capacity and has reportedly been undercutting market prices with DDR4 at or near half the international price. Management at Nanya explicitly cited Chinese DRAM oversupply as "the main cause of the prolonged [2024] downturn" in the February 2025 earnings call. CXMT is currently on older process nodes (roughly equivalent to 1Xnm, one to two generations behind Nanya's 1B), but the company is catching up. If CXMT reaches volume production at 1B-equivalent process nodes with DDR5, it will be a direct competitor not just in legacy DDR4 but in Nanya's growth product. US export controls on advanced semiconductor equipment may constrain CXMT's pace, but this is a dynamic, politically sensitive situation.

YMTC: Primarily a NAND flash maker. Relevant for the module business where DRAM and NAND are combined.

Why Nanya Wins and Where It Loses

Nanya wins on supply commitment to conventional DRAM. In an environment where Samsung and SK Hynix are diverting capacity to HBM, Nanya's focus on DDR4 and DDR5 makes it a reliable supplier when others are not available. It wins with industrial and automotive customers who value longevity agreements and specialty qualifications. It wins on pricing in tight supply environments because it has no HBM upside to chase.

Nanya loses on technology leadership. DDR5 at 6400 Mbps is Nanya's leading product in 2025-2026. Micron and Samsung are already sampling DDR5 at 8400 Mbps and developing DDR6. Nanya is at least one to two generations behind in process node and two to three generations behind in server-class DDR5 performance. This technology gap limits Nanya's access to hyperscaler server procurement, where performance specification leadership matters.

Nanya also loses on HBM, where it has no product and no near-term plans. The fastest-growing and most profitable DRAM segment is one Nanya does not participate in at all.

Barriers to Entry

The barriers to entering DRAM manufacturing are among the highest in global industry:

Capital: A competitive 300mm DRAM fab costs US$10-15 billion to build and equip. Nanya's own new fab is budgeted at NT$300 billion (roughly US$9-10 billion at current rates).
Process technology: DRAM process technology takes 10+ years of continuous development. Even with licensed technology as a starting point, building an independent node takes 5-7 years of intensive R&D.
Yield learning: DRAM yield at a new process node starts below 50% and improves through production learning. This improvement happens only in high-volume manufacturing, creating a barrier for any entrant that doesn't commit to scale.
Customer qualification: Particularly in automotive and industrial applications, customer qualification cycles lock in existing suppliers.

These barriers explain why consolidation has proceeded so relentlessly. They also explain why Nanya's continued existence depends on the Formosa Plastics Group's willingness to fund the capital cycle through DRAM downturns.

6. Industry

What Drives DRAM Demand

DRAM demand is fundamentally driven by the number of devices that contain memory and the amount of memory per device. Both factors have a long secular growth trajectory:

AI infrastructure: The most powerful near-term driver. AI training servers require massive DRAM content - an NVIDIA H100 server has 6TB of DRAM per rack, compared to under 1TB for a conventional server. AI inference servers similarly require large working memory for KV-cache operations. TrendForce estimates AI-related applications consumed approximately 20% of global DRAM wafer capacity in 2026. This demand is not cyclical - it is structural and growing.

Content per device growth: Each new generation of PC, smartphone, and server typically includes more DRAM than its predecessor. PC DRAM content has grown from 4GB typical to 16GB typical over a decade. Server DRAM content is growing 15-20% annually. AI smartphones - devices capable of running local inference models - require 8-16GB of LPDDR5, versus 4-6GB for conventional smartphones. This "content per box" growth continues even when unit shipments are flat.

Consumer electronics and mobile: Smartphones, tablets, and consumer devices remain the largest unit volume driver for LPDDR DRAM. This segment is more cyclical and was in a pronounced correction in 2023-2024, but it recovered into 2025.

Automotive and industrial: Long-cycle but growing markets. Advanced driver assistance systems (ADAS), vehicle infotainment, and EV computing require increasing DRAM content per vehicle.

Industry Size

The global DRAM market generated approximately US$121-128 billion in 2025, up substantially from 2024 due to HBM demand and conventional DRAM price recovery. TrendForce projected 2025 DRAM industry revenue at US$165.7 billion (reflecting a 73% annual increase). Looking forward, AI Architecture Evolution is expected to drive memory market revenue to new peaks in 2026-2027, with TrendForce projecting total memory market revenue of US$551.6 billion in 2026 and US$842.7 billion in 2027. These figures are for the total memory industry (DRAM plus NAND); DRAM-specific estimates are roughly 40-50% of the total.

Nanya's share of this market is approximately 2-3%, reflecting its position as the fourth-largest DRAM maker.

The HBM Squeeze and Conventional DRAM

The single most important industry dynamic for Nanya today is the indirect benefit it derives from the HBM (High Bandwidth Memory) boom. HBM - the stacked memory inside NVIDIA and AMD GPUs - is dramatically more profitable per bit than conventional DDR4 or DDR5. Samsung, SK Hynix, and Micron have all been converting conventional DRAM capacity to HBM production. This conversion has tightened the supply of standard DDR4, DDR5, and LPDDR significantly.

The physics of this relationship are important: one HBM chip requires roughly four times the DRAM wafer capacity of a conventional DRAM chip of equivalent bit density. As the big three allocate more wafers to HBM, fewer wafers are available for conventional DRAM - even as demand for conventional DRAM continues growing. This is the mechanism behind the DDR4 shortage that drove Nanya's dramatic second-half 2025 recovery.

Supply Chain Position

Nanya sits in the upstream portion of the memory supply chain, producing raw DRAM dies that feed into module makers, system integrators, and device OEMs. The supply chain from silicon wafer to finished device runs: silicon wafer suppliers (Shin-Etsu, SUMCO) → DRAM fab (Nanya) → die testing → module manufacturing → device OEM → end market.

Import Substitution and China Dynamics

China is both a major customer for DRAM and an aspiring domestic producer. CXMT's growing production is partly an import substitution effort - China currently imports the vast majority of its DRAM consumption, and the government has invested heavily in CXMT as a domestic champion. If CXMT successfully scales to advanced process nodes and becomes cost-competitive with international suppliers, it will address a portion of China's domestic demand that currently flows to Nanya and others. However, US export controls on advanced semiconductor manufacturing equipment create a meaningful constraint on CXMT's ability to access EUV lithography and certain DRAM-specific deposition tools, which may slow its progress on post-1X nm nodes.

Regulatory Environment

DRAM is not subject to specific product-level regulation in most markets, but the industry is deeply affected by export control regimes. The US Bureau of Industry and Security's export controls on advanced semiconductor equipment have limited CXMT's ability to purchase tools from ASML (EUV), Applied Materials, Lam Research, and KLA. These controls benefit Nanya by constraining Chinese competition.

Taiwan's strategic position as a semiconductor hub means Nanya operates with strong government support for its fab construction, including infrastructure provision and talent development programs. The Nanling Technology Park site for the new fab was provided with government assistance.

Cyclicality

DRAM is one of the most cyclical industries in global manufacturing. The mechanism is structural: when prices rise, all DRAM makers simultaneously invest in new capacity; that capacity arrives 2-3 years later, often simultaneously, creating oversupply; prices collapse; weaker players exit or reduce investment; capacity tightens and prices recover. This cycle has repeated with remarkable regularity since the 1990s.

The 2023-2024 downturn was a classic manifestation: the post-pandemic inventory glut hit simultaneously as global consumer demand weakened, causing DRAM prices to collapse. Nanya reported losses in each quarter of 2024 and early 2025. The recovery began in mid-2025, driven by the HBM capacity diversion (reducing conventional DRAM supply), AI demand growth, and inventory normalisation.

The current cycle has an unusual feature: the emergence of HBM as a persistent capacity absorber may dampen the conventional DRAM cycle compared to historical patterns. If the big three continue allocating a structurally larger share of their capacity to HBM, the conventional DRAM market may remain in a tighter supply/demand balance than historical cycles would suggest.

7. Growth Triggers

All points sourced directly from the four most recent quarterly earnings calls: Q1 2025 (April 2025), Q2 2025 (July 2025), Q3 2025 (October 2025), and Q4 2025 (January 2026). Note that Nanya's fiscal year follows the calendar year, so "Q4 2025" refers to October-December 2025, with the earnings call held in January 2026.

1B process reaching one-third of total wafer capacity in Q2 2025, enabling unit cost reduction and product mix improvement. Management stated the 1B node would represent one-third of capacity by end of Q2 2025. The node offers better die size (more chips per wafer) than 1A, directly reducing per-bit cost. (Q1 2025 concall, April 2025)
DDR5 contribution ramping from "very small" in Q1 2025 to "teen-percentage" shipment mix in Q2, with further growth targeted. Management described DDR5 contribution as "very small" in Q1 but expected significant Q2 ramp. By Q2, DDR5 was in the "teen-percentage" range. By Q4, DDR5 was approximately 10% of bit shipments. (Q1 2025 concall, April 2025; Q2 2025 concall, July 2025) [Repeated and updated across multiple calls]
DDR4 shortage expected to persist for "several quarters" due to HBM capacity diversion by the major players.

"Robust AI demand and constrained non-AI supply fuel the DRAM upcycle. DDR4 shortage is expected to persist for several quarters."

(Q3 2025 concall, October 2025)
16GB DDR5 RDIMM module at 5600/6400 speeds reaching functional test milestones, positioning Nanya for the server DRAM market. Management confirmed functional testing achieved on 128GB RDIMM modules at 7200 Mb/s mono-die performance in Q4 2025. Revenue from server RDIMM represents a new market segment for Nanya. (Q4 2025 concall, January 2026)
Custom AI projects described as "ahead of schedule" with revenue expected in H1 2026, targeting both cloud and edge AI applications. Management stated cloud AI customer projects and edge AI customer projects are ahead of schedule. (Q4 2025 concall, January 2026)

"AI projects: customer make projects ahead of schedule; revenue expected H1 2026 for both cloud and edge AI."
1C process (third-generation 10nm-class) qualification targeted by end of 2026, enabling further cost reduction and density improvement. Management confirmed 1C is on schedule and qualification will be sought by end of 2026. The 1D node is approximately one year beyond 1C. (Q4 2025 concall, January 2026)
New fab equipment move-in beginning early 2027, with initial output from H2 2027. Management stated the new fab is "almost ready" for equipment installation and ramp to begin.

"Our new fab is almost ready, and we plan to start bringing in equipment at the beginning of next year. We aim to have some output by the end of the first half and gradually ramp up in the second half."

(Q4 2025 concall, January 2026) [Timeline mentioned across Q3 and Q4 2025 calls]
Customers shifting from spot market to quarterly and annual supply contracts, enabling greater ASP stability. Management described this shift as accelerating through H2 2025. Longer-term contracts reduce quarter-to-quarter revenue volatility. (Q3 2025 concall, October 2025; Q4 2025 concall, January 2026) [Repeated]
Depreciation expected to decline approximately 10% in 2026 versus 2025, as legacy equipment on older nodes completes its depreciation schedule. This structural cost reduction will flow directly to gross margin at equivalent ASP levels. (Q4 2025 concall, January 2026)
LPDDR5 products on the 1B process targeted for introduction in 2026, addressing the AI smartphone upgrade cycle where handset makers require higher-speed, lower-power DRAM for on-device inference. (Q1 2025 concall, April 2025; Q4 2025 concall, January 2026) [Repeated]
Q1 2026 pricing expected to increase "likely exceeding 20%", with management citing continued supply constraints across DDR5, DDR4, LPDDR, and DDR3.

"Supply shortage may persist for multiple DRAM products due to capacity constraint."

(Q4 2025 concall, January 2026)

Trigger	Timeline	Source	Status
1B process at 1/3 of capacity	Q2 2025 (complete)	Q1 2025 call	Delivered
DDR5 ramp to 10%+ shipment mix	Through 2025	Q1-Q4 2025 calls	Delivered
DDR4 shortage persists for "several quarters"	Through 2026	Q3 2025 call	Ongoing
Server RDIMM milestone achieved	Q4 2025 (complete)	Q4 2025 call	Delivered
Custom AI project revenue	H1 2026	Q4 2025 call	New
1C process qualification	End 2026	Q4 2025 call	New
New fab equipment move-in	Early 2027	Q3/Q4 2025 calls	Repeated
Depreciation decline 10%	2026 full year	Q4 2025 call	New
LPDDR5 1B introduction	2026	Q1, Q4 2025 calls	Repeated
Q1 2026 pricing >20% increase	Q1 2026	Q4 2025 call	Confirmed (Q1 2026 revenue +63% QoQ)

8. Key Risks

1. Chinese DRAM Competition (CXMT) - High Probability, Moderate-to-High Impact

CXMT has been the clearest near-term threat to Nanya. Management explicitly blamed Chinese DRAM oversupply for the extended 2024 downturn: the mechanism was straightforward - CXMT ramped DDR4 production and sold at aggressive discount prices (reportedly 40-50% below international market prices), flooding the spot market and suppressing contract prices for Nanya and others. If CXMT continues to scale and eventually achieves competitive yields on 1B-class process nodes, it will compete with Nanya not just on legacy DDR4 but on current-generation products. The key uncertainty is whether US export controls on semiconductor equipment (specifically EUV and certain DRAM-specific deposition tools) will meaningfully constrain CXMT's technology advancement. If those controls are relaxed or circumvented, CXMT's timeline to competitiveness on advanced nodes accelerates significantly.

2. Technology Gap Relative to Big Three - Structural, Chronic

Nanya is 1-2 process generations behind Micron, Samsung, and SK Hynix across all product categories. In a commodity market, process leadership means lower cost per bit, which means higher margins at equivalent prices - or the ability to undercut on price and still profit. If Nanya's cost structure remains structurally higher than the big three due to process lag, it will always be at risk of margin compression when prices are weak. The 1B and 1C development programs are the countermeasure, but process development has a history of delays. Any slippage in 1C or 1D timelines extends the period of structural cost disadvantage.

3. New Fab Execution Risk - Moderate Probability, Very High Impact

The NT$300 billion new fab is the most consequential decision in Nanya's history. The risk is not that the facility fails to be built, but that it is built and ramped in a period of DRAM oversupply, creating a massive fixed cost burden with insufficient revenue to cover it. The previous DRAM cycle had a similar dynamic for several companies: new fabs commissioned at cycle peaks came online into price collapses. The new fab's first output is expected in H2 2027. If the current DRAM upcycle peaks before then and corrections arrive by 2028, Nanya will face the simultaneous pressure of new depreciation loads (management acknowledged depreciation will rise again in 2027 with new fab equipment) and potentially weaker pricing. The capital financing also introduces risk: management mentioned that both debt and equity issuances are under consideration for the new fab funding, beyond the $2.5 billion private placement. Significant equity dilution would affect existing shareholders.

4. DRAM Cycle Reversal - High Probability, Periodic

DRAM cycles are structural and recurrent. The current upcycle, driven by HBM capacity diversion and AI demand, has been unusually sharp. But the conditions that will end it are already forming: the big three are adding DDR5 capacity even as they prioritise HBM; CXMT is adding DDR4 capacity; and demand in consumer electronics has been recovering, which could lead to inventory build. Nanya's relatively lean balance sheet (net cash was NT$38.5 billion at Q4 2025) provides some buffer, but the company generated losses in every quarter from 2023 through Q2 2025. If the next cycle downturn is as severe as 2023-2024, Nanya will again face a prolonged period of cash burn.

5. Currency Risk - Moderate, Ongoing

Nanya sells products primarily in USD (as is standard for global DRAM trade) but incurs costs primarily in TWD. A strengthening TWD reduces the TWD equivalent of USD revenues without proportionally reducing TWD-denominated costs. The Q2 2025 concall noted that exchange rate impacts totalled NT$5.2 billion on cash and equivalents in that quarter alone, contributing to the widening net loss despite revenue growth. Nanya holds significant USD cash as a natural hedge, but the underlying operational exposure to TWD/USD is persistent.

6. HBM Exclusion - Strategic, Long Duration

Nanya has no HBM product and no disclosed plan to develop one. HBM is the fastest-growing and most profitable DRAM segment. If AI GPU demand continues to be the dominant force in semiconductor capital allocation, HBM will claim an ever-larger share of DRAM industry revenue. A company that participates only in conventional DRAM will grow more slowly than one that participates in both. The Kioxia OCTRAM collaboration and PieceMakers investment suggest Nanya is exploring future memory architectures, but these are research-stage bets, not near-term revenue contributors. Nanya's willingness to remain outside HBM is a deliberate focus strategy, but it carries the risk of being left behind if HBM dominance is more durable than expected.

7. Customer Concentration Emerging from Private Placement - Moderate

The $2.5 billion private placement with SanDisk, Kioxia, Solidigm, and Cisco is strategically positive but creates new dependency. Multi-year supply agreements mean these four customers are now obligated buyers and Nanya is obligated supplier at terms negotiated under current conditions. If DRAM prices rise well beyond the contracted levels, Nanya will be underselling its market value on the contracted volume. If one of the strategic investors faces financial difficulty or changes strategy (as has happened in the memory industry before), the supply agreement dynamics could become complicated.

9. Walk the Talk

Across the four most recent concalls - Q1 through Q4 2025 - management's credibility has been solid overall, with one notable area of under-performance and several commitments that were delivered as or better than guided.

Q4 2024 Concall as Baseline (January 2025)

Before examining the 2025 calls, it is worth noting what management guided at the end of 2024. The Q4 2024 concall (from which we have data) provided two key commitments: bit shipment growth of "greater than 20%" for full-year 2025, and a recovery beginning in the "second half of 2025" driven by inventory normalisation and stimulus. Both of these were subsequently exceeded - full-year 2025 bit shipment growth came in at approximately 50% (against the 20%+ guidance), and the recovery did not just appear in H2 but accelerated dramatically through Q3 and especially Q4. The initial guidance was significantly conservative.

Q1 2025 Concall (April 2025): Raise and Deliver

By Q1 2025, management had revised its full-year shipment guidance sharply upward - from "greater than 20%" to "30%" growth year-over-year. This was itself a material upgrade that demonstrated management's ability to update guidance responsively. More specifically, management guided for operating breakeven potential by Q3-Q4 2025, dependent on depreciation improvements and ASP recovery. This commitment was delivered: Q3 2025 saw the company return to positive gross margin (18.5%) and positive net income (TWD 1.563 billion). The Q3 timing was precisely as telegraphed.

The 1B process reaching one-third of total wafer capacity by Q2 was also guided in Q1 - confirmed in subsequent calls.

Q2 2025 Concall (July 2025): Promised Q3 Gross Margin, Delivered It

The Q2 2025 call was the most difficult quarter in the reported period - gross loss widened despite revenue growth because ASP fell mid-single digits and exchange rates moved adversely. Management's response was direct:

"Gross loss increased by TWD 1.09 billion, mainly due to lower ASP mid-single digit and unfavorable exchange rate."

What matters is that management did not soften the outlook. Instead, they stated confidently that "Non-AI application bottomed out at the end of Q2" and that gross margin would "confidently" turn positive in Q3. Q3 2025 gross margin was 18.5% - not just positive, but substantively profitable. This statement-to-outcome match is the most important single data point on management credibility in this period.

The full-year shipment guidance was also raised again to "over 40%" YoY, from the prior 30% guidance. Actual full-year 2025 growth was approximately 50%. Again, the company delivered at or above guidance.

Q3 2025 Concall (October 2025): Accurate Framing of the Upcycle

By Q3 2025, the DRAM upcycle was clearly underway. Management's commentary was measured: they correctly identified DDR4 shortage as likely to persist "for several quarters," described AI demand as structural rather than cyclical, and guided Q4 shipments as "roughly flat with September levels." Q4 did see shipments grow modestly (low-teens percentage QoQ) while ASP surged 30%+ - the flat bit shipment guidance was approximately correct, and the revenue and margin surprise came from pricing rather than volume.

The "2027" timeline for new fab production was confirmed in Q3, consistent with what had been guided in prior quarters. No slippage was acknowledged.

Q4 2025 Concall (January 2026): Record Results, Disciplined Caution

Q4 2025 produced the company's best-ever quarterly results. Management's guidance for Q1 2026 - pricing "likely exceeding 20%" - turned out, if anything, to be conservative again. Q1 2026 revenue came in at NT$49.09 billion, up 63% QoQ (compared to Q4's NT$30.09 billion), implying pricing increases substantially above the 20% threshold when combined with volume.

Management's caution around dividend policy was also noteworthy:

"For 2026 earnings, we're looking for lower percentage [payout ratio] to preserve cash for future CapEx."

This is disciplined capital allocation language, prioritising long-term investment over short-term shareholder distribution when the company's biggest capital commitment (the new fab) is approaching its most intensive spend phase.

Overall Assessment

Management has demonstrated a consistent pattern of conservative guidance followed by delivery at or above that guidance. The single period of genuine underperformance - Q2 2025, where ASP weakness and currency headwinds widened losses relative to Q1 - was clearly explained and immediately followed by the accurately-predicted Q3 recovery. Across four calls, management did not make promises they could not keep. They were transparent about risks (CXMT competition, currency exposure, cycle timing) and accurate about operational milestones (1B ramp, DDR5 contribution, profitability return). The one area to watch is the new fab timeline and capex management: this is the largest execution commitment in the company's history and has not yet been tested by actual construction progress beyond groundbreaking.

10. Scenarios

Bull Case: The Conventional DRAM Decade

The AI era proves to be the most sustained DRAM upcycle in the industry's history - and Nanya's focused bet on conventional DRAM turns out to be exactly the right position. Samsung, SK Hynix, and Micron continue diverting ever-larger fractions of their capacity to HBM for GPU servers, leaving DDR4 and DDR5 markets perpetually undersupplied. CXMT's technology advancement is hampered by export controls that prevent access to EUV lithography, keeping Chinese competitors on older, less competitive nodes through the end of the decade.

In this scenario, Nanya completes the new fab on schedule in 2027, ramps to 45,000 wafers/month without the misfortune of commissioning into a down-cycle. The 1B and 1C process nodes deliver competitive costs, narrowing the gap with the big three. DDR5 and LPDDR5 products win qualification at major server and smartphone OEMs. The strategic relationships with SanDisk, Kioxia, Solidigm, and Cisco deepen: these partners route growing volumes of DRAM-adjacent demand to Nanya, and the equity relationships expand into joint development of next-generation products. Custom AI projects (cloud and edge) that management described as "ahead of schedule" grow into a genuine revenue line for AI edge memory. The Kioxia OCTRAM collaboration yields commercially viable technology that gives Nanya an early position in post-silicon DRAM. By 2028, Nanya has doubled its wafer capacity, held its market share in the 3-4% range despite the industry growing, and generated the cash to fund subsequent process generations independently.

Base Case: Steady Recovery, Measured Execution

The current upcycle continues through 2026 and into early 2027, with pricing remaining firm as HBM capacity absorption persists and AI demand grows. Nanya executes on its stated plans: 1C node qualification achieved by end of 2026, new fab equipment begins moving in as planned in early 2027 with first output in H2 2027. The private placement funds flow into new fab construction without requiring additional equity dilution. Server RDIMM qualifications at major hyperscalers take 18-24 months and begin generating revenue in 2027-2028, opening Nanya to a customer segment that was previously inaccessible.

CXMT remains a competitive threat but primarily in the DDR4 spot market and lower-tier consumer applications. Nanya's qualification advantages in automotive, industrial, and server markets protect the more defensible portions of its business. The cycle turns moderately downward in 2027-2028 as new DDR5 capacity comes online from multiple sources, but the new fab's ramp is gradual enough that Nanya avoids the worst of the timing risk. Over a full cycle, the company sustains moderate profitability through the investment period and emerges from the new fab ramp with a competitive cost structure.

Bear Case: The Timing Trap

The DRAM cycle peaks in 2026 and begins turning downward well before Nanya's new fab is productive. CXMT continues scaling aggressively, with Chinese domestic production of DDR4 suppressing international prices below what Nanya can profitably produce. The US export control regime is partially relaxed, or CXMT finds alternative routes to advanced equipment, accelerating its node transition toward 1B-class technology. Korean producers respond to any signs of oversupply by aggressively cutting prices to defend share, squeezing smaller players like Nanya first.

The new fab arrives in 2027-2028 precisely when DRAM prices are weakest. Nanya faces a double pressure: declining ASPs on existing production and the beginning of substantial new depreciation loads from the greenfield facility. The company's net cash position erodes as it funds the new fab capex (NT$50 billion in 2026 alone, with more to follow), and dividend commitments must be suspended. The strategic investors from the private placement have locked in supply at contractual pricing that, in a down-cycle, may be above spot market prices - giving them an incentive to minimise contract volumes while spot prices are lower. The 1C process suffers yield delays, pushing competitive cost parity with the big three further out. Nanya faces another multi-year period of losses while simultaneously managing the most complex operational challenge in its history.

The existential floor in this scenario is the Formosa Plastics Group's balance sheet: the parent has demonstrated willingness to fund Nanya through prior downturns, and the new fab is too strategically important to abandon mid-construction. But the cost - in shareholder value, in operational stress, and in competitive position - of a badly-timed cycle turn would be severe.

Sources: