Global Reliable Electronic Components Distributor - ICs Supplier

SK Hynix Bets on AI Inference Market with 800% DRAM Capacity Expansion

2025-11-21
As artificial intelligence applications shift from model training to large-scale inference, the demand for general-purpose DRAM is surging. SK Hynix has keenly captured this market shift and plans to significantly expand its capacity for sixth-generation 10-nanometer DRAM (1c DRAM). This strategic adjustment reflects the strong demand for cost-effective general-purpose DRAM in AI inference applications and also shows that SK Hynix is shifting its strategic focus from high-bandwidth memory (HBM) to a broader AI memory market.

Capacity Expansion Plan

According to South Korean media reports, SK Hynix plans to increase the monthly capacity of its sixth-generation 10-nanometer DRAM from the current approximately 20,000 300mm wafers to 160,000–190,000 wafers by 2026. This represents an increase of 8–9 times and will account for more than one-third of its total DRAM capacity. The expanded 1c DRAM will mainly be used to produce products such as GDDR7 and SOCAMM2 to meet the order demands of large technology companies like Nvidia.
SK Hynix Bets on AI Inference Market with 800% DRAM Capacity ExpansionSK Hynix's capacity expansion plan not only includes adding equipment in existing factories but also involves the construction of new factories. The company plans to add 140,000 wafers per month next year by upgrading processes at its Icheon factory. In addition, SK Hynix is also considering further increasing its monthly capacity to 160,000–170,000 wafers. This expansion plan demonstrates SK Hynix's strong confidence in the "AI memory supercycle."

Market and Technology Background

SK Hynix's expansion plan is carried out against the backdrop of a boom in AI infrastructure investment. With the popularization of AI applications, the demand for high-performance DRAM is also steadily growing, especially for high-capacity DDR5 module products. SK Hynix plans to accelerate the migration to its most advanced 1c DRAM process, which will establish a "complete" DRAM product line for server, mobile, and graphics applications.
In addition, SK Hynix also holds a dominant position in the HBM market, with its HBM4 product set to begin shipping in the fourth quarter of 2025. The company has already locked in all customer demand for next year's DRAM and NAND production. This indicates that SK Hynix is in a favorable position in the AI memory market, able to meet the market demand for both HBM and general-purpose DRAM.

Industry Competition and Market Outlook

In the DRAM market, SK Hynix's main competitor, Samsung, is also making efforts to regain its market leadership. According to reports, Samsung plans to expand its monthly capacity for 1c DRAM to 200,000 wafers by the end of 2026. This target includes achieving a monthly capacity of 60,000 wafers by the end of 2025, adding another 80,000 wafers in the second quarter of 2026, and then another 60,000 wafers in the fourth quarter.
SK Hynix's expansion plan and market strategy show its long-term confidence in the AI memory market. Compared with HBM, which requires complex stacking processes, 1c DRAM has higher production efficiency and can more quickly respond to the explosive demand for AI memory in the market. In addition, SK Hynix also plans to further expand the production of HBM4 through its new factory M15X to meet the needs of customers like Nvidia.

Semicon Chip distributors

As a professional semiconductor chip distributor, Semicon is dedicated to providing a wide range of IC chips, covering multiple fields such as analog, digital, power management, interface and communication, to meet the needs of different application scenarios. Semicon's popular chips include microcontroller chips, power devices and modules, power management and driver chips, analog and mixed-signal chips, etc. Here are several popular and carefully recommended IC models:
● ADIS16475-1BMLZ: A high-performance six-axis inertial measurement unit integrating a three-axis accelerometer and a three-axis gyroscope, suitable for high-precision navigation and attitude measurement.
● BTT60502ERAXUMA1: A high-power MOSFET, suitable for power management and motor drive applications, featuring low on-resistance and high withstand voltage characteristics.
● STM32F051C8T6: A high-performance 32-bit microcontroller featuring a rich array of peripheral interfaces and low power consumption, suitable for a wide range of embedded applications.
● 74AVC4TD245BQ,115: A high-speed dual-bus buffer that supports multiple data transmission modes and is suitable for communication and interface expansion.

Policy Pivot: U.S. Considers Allowing NVIDIA to Export H200 Chips to China

Google Launches Gemini 3: A New Era of AI Intelligence

Related Article

image
The MLCC market shows stark divergence: general‑purpose lead times stay at 14‑18 weeks, while high‑end AI/server models stretch beyond 20 weeks.
MLCC Market Split: General Stable, Premium Extended Lead Times
2026-08-31
image
Explore the key semiconductor component trends shaping 2026, including Memory IC, PMIC, MCU, Automotive IC, MOSFET, Analog IC and high-speed connectivity.
2026 Semiconductor Trends: Key ICs the Global Electronics Industry
2026-08-28
image
Navitas acquires Claros. It builds a rare grid-to-chip power architecture, creating an end-to-end pipeline from utility grid to AI processor cores.
Navitas to Acquire Claros, Extending AI Power Architecture from Grid to Chip
2026-08-27
image
Following its $20B Groq acquisition, NVIDIA's Groq 3 LPX enters mass production. Samsung foundry supplies 4nm wafers, Samsung Electro‑Mechanics provides FC‑BGA substrates.
NVIDIA Groq 3 LPX Enters Full Production: $20 Billion Acquisition Comes to Fruition
2026-08-26