Taiwanese memory producer Nanya Technology once occupied the industry’s overlooked corner. Samsung, SK Hynix and Micron rarely glanced its way. Commodity chips for PCs and gadgets defined its output. Low margins. Cyclical pain. Losses that lingered.
Artificial intelligence changed the math. Demand for DRAM in servers, accelerators and edge systems exploded. Contract prices soared. Nanya’s first-quarter 2026 revenue jumped more than 580% from a year earlier. Gross margins hit 68%. By the second quarter, unaudited figures showed revenue of NT$82.55 billion, up 684%, with net income of NT$50.19 billion, a 1,324% surge. Gross margin reached 79.5%. Tom’s Hardware reported the stunning turnaround.
Executives see no quick relief. Chairman Tsou Ming-jen and President Pei-Ing Lee have warned the shortage could stretch into 2027 or even 2028. Major suppliers shifted capacity toward high-bandwidth memory for AI accelerators, starving conventional DRAM output. Cloud operators, smartphone makers and PC brands felt the pinch. Apple withdrew its cheapest Mac Mini from sale as costs climbed, according to reporting last year.
Nanya responded with ambition. The company plans capital spending exceeding NT$200 billion, or about $6.2 billion, in 2027. That would quadruple this year’s budget of more than NT$50 billion. The outlay, still preliminary and awaiting board approval, targets one project above all. A new 5A fabrication plant in New Taipei City’s Taishan District. Equipment installation begins in the first quarter of 2027. Initial output targets more than 30,000 wafers monthly by 2028, eventually scaling toward 45,000. Total investment in the facility could reach NT$480 billion at full capacity. President Lee shared the figures during an online briefing covered by Reuters.
The scale dwarfs Nanya’s recent history. Capital expenditure ran NT$13.2 billion in 2023, NT$16.1 billion in 2024 and NT$13.4 billion in 2025. The 2026 plan of NT$52 billion already tripled the prior year, with 70% directed at the new facility. A leap to $6.2 billion would mark the most aggressive expansion in the company’s existence. And yet it remains modest next to rivals. SK Hynix commits $51 billion to one new plant. Micron broke ground on a $9 billion project in Hiroshima. Samsung posted an 18-fold profit increase in the same wave.
Nanya sits far smaller than the three dominant DRAM makers. Its bet rests on carving a niche rather than matching their volume. The firm pushes customized high-bandwidth memory variants for AI servers and edge applications. Domestic brokerages forecast server-related products could exceed 60% of sales by 2027, up from near zero recently. A partnership with Etron Technology, Piecemakers Technology and Formosa Advanced Technologies aims to deliver a specialized ultra-high-bandwidth DRAM product by the end of this year. The company works on 10nm-class third- and fourth-generation processes. Validation of 1C 16Gb DDR5 finishes in the second half of 2026. Pilot production of the 1D version starts in the second quarter.
Customers fund the shift. In March 2026 Nanya raised NT$78.72 billion, roughly $2.5 billion, through a private placement. SanDisk Technologies committed about $970 million. SK Hynix’s Solidigm unit, Kioxia and Cisco each contributed around $500 million. The investors secured multi-year supply agreements. “The proceeds will be used to invest in NTC’s factory facilities and production equipment for advanced memory manufacturing to address the surge in computational demand driven by next-generation AI,” the company stated at the time. Shares jumped 10% to the daily limit following the announcement. Yahoo Finance covered the share surge and deal details.
Strategic alignment runs deep. Nanya’s buyers include Nvidia, Qualcomm and Google. These firms lock in supply amid tight markets. Three SSD makers — SanDisk, Kioxia and Solidigm — joined to guarantee DRAM access for their own products. The arrangement reflects a broader industry pattern. When shortages bite, customers buy equity to protect future deliveries. Nanya’s ties to the Formosa Plastics group provide stability, yet its past reliance on low-margin PC and consumer DRAM left it exposed to price wars.
Redirecting output toward AI servers offers escape. DDR4 and low-power DDR4 still account for about 70% of shipments. DDR5 contributes roughly 10% of revenue today. But AI server architectures pack far more memory per system. One rack can hold DRAM equivalent to 4,500 smartphones, according to industry observers tracking Nvidia designs. Edge AI, AI PCs, smartphones and robots add further pressure. Chairman Tsou highlighted how HBM and RDIMM components boost overall DRAM content in servers. The trend extends beyond cloud data centers.
Analysts temper enthusiasm. The new capacity arrives years from now. First wafers start late 2027 or early 2028. By then the shortage may ease. Global suppliers plan their own expansions. A synchronized wave of new fabs risks tipping the market from scarcity to oversupply. Memory cycles have punished overbuilders before. Debt incurred during the boom could weigh heavily if prices collapse.
Nanya acknowledges the hazards. Its executives stress that AI-driven demand appears sustainable. Cloud spending leads. Consumer segments show selective softness. The firm licensed advanced process technology from Micron in prior years, aiding its technology roadmap. It avoids direct competition in leading-edge HBM stacks dominated by SK Hynix. Instead it targets customized solutions for specific AI workloads.
Profitability transformed fast. Full-year 2025 revenue reached NT$66.59 billion, up 95%. Net profit after tax hit NT$6.61 billion. The second quarter of 2026 alone generated more profit than all of 2025. Such figures attract attention. Yet sustaining 79.5% gross margins demands continued pricing power. ASPs, or average selling prices, climbed sharply as supply lagged.
The 5A fab represents more than bricks and tools. It signals Nanya’s attempt to reinvent itself inside Taiwan’s semiconductor cluster. Construction began years ago in Nanling Technology Park. Volume production follows equipment move-in. The plant uses 1B technology, the company’s second-generation 10nm-class node, for DDR5, DDR4 and LPDDR4 initially.
Broader market forecasts reinforce the wager. Some analysts project the global HBM market alone could reach $127 billion by 2028. SK Hynix controls about half today. Challengers scramble. Nanya’s customized approach and strategic investor base position it as a supporting player rather than a front-line combatant. Its edge lies in agility and long-term supply commitments.
Still, questions linger. Can $6.2 billion close the experience gap with larger rivals? Will yields and ramp schedules meet aggressive timelines? How quickly can the company shift its product mix toward higher-value AI memory? Executives provided no guarantees. The 2027 budget remains preliminary. Board review could adjust numbers.
For now momentum favors the bold. Nanya posted losses not long ago. Its transformation mirrors the industry’s larger reordering. Memory no longer serves as a simple commodity. It forms a strategic bottleneck in AI infrastructure. Suppliers that secure capacity, technology and customer contracts stand to gain. Those that hesitate risk marginalization.
The concrete pours. The tools arrive in 2027. Demand, everyone hopes, waits on the other side. But memory’s history teaches caution. Booms invite busts. Nanya’s $6 billion commitment tests whether this cycle differs. Its customers, investors and executives believe the appetite for intelligence will outrun the time required to build new capacity. The coming years will render judgment.
WebProNews is an iEntry Publication