Memory Makers Accelerate Construction of New Facilities, Yet Extra Capacity Benefits Only the AI Elite

Memory Manufacturers Mobilize to Tackle Shortages Amid AI Demand

As the demand for memory products surges, a daunting challenge has emerged: severe shortages that heavily favor sellers. With the rapid expansion in artificial intelligence (AI) technologies, companies are eagerly seeking long-term agreements (LTAs) with leading memory manufacturers, such as Micron. Compounding this issue, the consumer sector continues to see unstoppable demand, urging suppliers to concentrate on boosting production capabilities. Recent reports from The Wall Street Journal and Korean media indicate that memory manufacturers are intensifying their efforts, committing to invest hundreds of billions in capacity expansion projects.

Focusing on Micron, which has made headlines for its ambitious investments, the company has earmarked a staggering $200 billion for its future growth. As outlined in a report by the WSJ, one of Micron’s flagship projects includes developing a 450-acre facility in Boise, Idaho. This site is set to house the largest cleanroom facility in the United States, projected to boost production to between 150, 000 and 200, 000 wafers per month (WPM), thereby enhancing Micron’s worldwide capacity by approximately 40%.

Impact of increased DRAM costs on smartphone Bill of Materials (BoM) — Image Credits: SK hynix

In addition to the Boise investment, Micron’s facility in New York marks a historic private investment for the state, estimated at $100 billion. This project includes plans for four additional 600, 000-square-foot cleanrooms, which will further enhance DRAM production capabilities. However, these expansions raise an important question: when will these facilities begin contributing to the market? The Boise site is expected to reach full capacity by the second half of 2027, while the New York project may take until 2045 to be fully operational.

Meanwhile, Korean memory suppliers are not sitting idly by. A recent report from Chosun Biz revealed that companies such as We hynix are accelerating their production schedules to take advantage of the ongoing supercycle in memory demand. We hynix plans to commence test operations at its Yongin fab shortly, which promises to be a marked advancement over Micron’s endeavors. The Yongin facility is set for an investment of $85 billion, with initial operations anticipated as early as February or March, ahead of previous projections.

Samsung is also stepping up its efforts by fast-tracking its plans for the Pyeongtaek P4 fab, aiming for completion by the fourth quarter of 2026 instead of early 2024. Once operational, this facility is projected to deliver 100, 000 to 120, 000 WPM, potentially easing the supply constraints by 2027.

Bar chart comparing total DRAM capacity across NVIDIA architectures — Visualization of DRAM capacity trends across NVIDIA architectures | Image Credits: Wccftech

In the context of AI, the importance of DRAM is rapidly evolving. Notably, the LPDDR standard, once predominantly utilized in consumer products, is now gaining traction in NVIDIA’s advanced rack-scale offerings. As AI applications increasingly rely on high memory bandwidth, manufacturers are racing to implement cutting-edge technologies, including High Bandwidth Memory (HBM) and SOCAMM modules. This uptick in DRAM capacity is poised to address the soaring demands from the AI sector.

While there is hope that memory conditions may eventually stabilize for consumers, it is crucial to understand that a significant portion of the expanded capacity will likely prioritize the booming AI sector. Consequently, persistent memory shortages may loom over the consumer market for the foreseeable future.

Source & Images