[Conference Call] SK hynix Rebuts 'AI Investment Slowdown' Concerns... "Supply Surplus Unlikely" (Comprehensive)

DRAM and NAND Shipments to Expand Further in Q3 Compared to Q2

On AI Investment Slowdown Concerns: "Moving Into Full-Fledged Monetization Phase"

LTA: "Will Operate at Appropriate Level Considering Market and Environment"

"Reviewing Ad

SK hynix addressed concerns about a slowdown in artificial intelligence (AI) infrastructure investment by projecting that demand will increase as access to AI services expands. The company analyzed that as the number of users grows, the monetization of AI infrastructure will expand in earnest, which will in turn drive continued chip purchases. Despite forecasts of a semiconductor supply glut leading to price declines, SK hynix stated that the likelihood of such a scenario is low, since their investments are based on clear demand visibility.

On July 29, during the company’s earnings conference call for the second quarter of this year, Sung Hyun-jong, President of SK hynix, announced, “In the second quarter, we continued to see robust demand driven by the expansion of AI infrastructure investment combined with a tight supply environment, which led to ongoing price increases. We also set a new record for quarterly sales, continuing the momentum from the previous quarter.”

SK hynix forecast this year’s demand for DRAM to grow by around the mid-20% range and NAND by the upper 10% range. The company also noted that advanced processes essential for HBM and AI server memory are becoming increasingly challenging, and that constructing new production facilities takes time, making a quick resolution of supply shortages unlikely. Accordingly, SK hynix plans to increase DRAM shipments by about 10% quarter-over-quarter in the third quarter, and expand NAND shipments by a low single-digit percentage. For HBM4, which began mass production and shipments in the second quarter, full-scale production will ramp up in the second half of the year; meanwhile, HBM4E development will continue based on samples supplied to major customers.

The company is also expanding investments to strengthen its production capacity. SK hynix is accelerating the mass production timeline at the Cheongju M15X plant and is working to expand manufacturing capacity following the cleanroom opening at the first fab in Yongin in early 2027. As a result, this year’s total capital expenditures are expected to reach the upper-40 trillion won range. The company is also moving forward in phases with investments in the advanced packaging facility P&T7 and the new NAND production base M17.

During the Q&A session, numerous questions were raised about concerns over a slowdown in AI investments. In response, an SK hynix representative stated, “We are well aware of the concerns that the review of data center leasing businesses by some big tech firms and the emergence of highly efficient AI models may signal a slowdown in AI infrastructure investments. However, rather than seeing this as a contraction, we view it as a process of increasing utilization and accelerating monetization of the large-scale AI infrastructure that has been built so far.”

Addressing concerns about a possible semiconductor supply oversupply, the company stated, “As our capacity expansion will be implemented flexibly, the odds of mid- to long-term investment immediately leading to a supply glut are limited.” The company emphasized, “Capacity expansion is based on visibility of secured market demand. Actual equipment investment and production operations will proceed in phases, taking into comprehensive consideration customer demand visibility and investment efficiency.”

Regarding observations that the pace of semiconductor price increases had slowed in the second quarter, SK hynix explained, “In the second quarter of this year, the shipment timing of some high-value-added products was postponed to the second half, and changes in the product mix appear to have influenced the overall blended average selling price (ASP). In the second half of the year, the volume of HBM4 shipments will expand significantly, and shipments of standard DRAM based on the sixth-generation (1c) 10nm-class process will also rise, resulting in bit growth (shipment growth rate) that is expected to surpass that of the first half of the year.”

On long-term supply agreements (LTA) with customers, the company did not disclose specific contract details. The representative stated, “The standard contract term is typically five years, but specific terms can vary by customer and product. As for the pricing structure, we are exploring various models with our clients, rather than relying on a single format, to accommodate price volatility.” He added, “We plan to manage LTAs at an appropriate level considering market conditions and customer needs. This will enhance downside stability for our performance while ensuring flexibility to pursue growth opportunities in favorable market environments.”

Discussing domestic and international investment strategies, the company remarked, “Our mid- to long-term investment direction is to make timely investments in line with AI demand, while implementing capital expenditures (CAPEX) based on supply-demand dynamics. Moving forward, rather than distinguishing between domestic and overseas production bases, we will comprehensively consider electricity, water, personnel, supply chains, the semiconductor ecosystem, and customer accessibility to determine the optimal locations.”

Responding to questions on HBM development strategy, the company explained, “We began mass production and supply of HBM4 to major customers starting in the second quarter of this year. The production yield and quality have already reached levels close to those of the previous generation (HBM3), which is considered mature. We are steadily ramping up our supply capacity.” SK hynix continued, “Next-generation HBM4E development is progressing smoothly as planned, with a goal of entering full-scale mass production in 2027. In addition to hybrid bonding, we are developing ‘IHBM’ technology to effectively address thermal issues in next-generation products such as HBM5.”

Regarding NAND flash strategy, SK hynix noted, “As the AI market shifts from training to inference-centric workloads, NAND is rapidly becoming a core component of the AI memory hierarchy.” The company added, “Our NAND strategy for the AI era is not about choosing between SLC, MLC, TLC, or QLC, but about providing an optimal portfolio tailored to the customer workload. With this comprehensive portfolio, we will address all AI demand and expand the NAND product lineup to secure new growth engines.”

SK hynix stated there are no concrete plans at this time to further expand U.S. ADRs (American Depositary Receipts). Regarding additional shareholder returns, the company explained, “We are currently reviewing various options for further shareholder returns and plan to make an announcement within this year.”

This content was produced with the assistance of AI translation services.