SK Hynix operating profit hits 60.5 trillion won despite HBM4 delays

The tension between record profitability and manufacturing delays underscores the critical bottleneck in the AI hardware stack. As the primary memory supplier for Nvidia, SK Hynix’s execution directly impacts the deployment timelines for next-generation AI infrastructure. While the company holds a 50% market share in High Bandwidth Memory, rising production costs—including a 10% price hike from lithography suppliers—are testing margins as Samsung and Micron ramp their own AI-focused memory. Watch for AMD’s August 4 earnings report and HBM4 qualification updates as key signals for the health of the broader AI hardware ecosystem.

The global competition for High Bandwidth Memory (HBM) has intensified as all three major memory makers—SK Hynix, Samsung, and Micron—race to secure slots for Nvidia’s Rubin GPU platform. Per EE Times (January 2026), SK Hynix remains the leader with over 50% market share, having unveiled a 16-layer HBM4 device at CES 2026 that utilizes proprietary Mass Reflow Molded Underfill (MR-MUF) technology to meet strict height requirements. However, Micron has positioned itself as a credible second-source supplier, reporting in June 2026 that its HBM4 shipments for the Vera Rubin platform began in March 2026, with yields improving ahead of expectations.

Capacity expansion is now the primary lever for these manufacturers to meet unprecedented demand. Per Chosun (July 2026), SK Hynix moved up the mass production schedule for its M15X fab in Cheongju and is investing heavily in the Yongin Semiconductor Cluster, where the first cleanroom is expected to open in early 2027. This urgency is mirrored by Samsung, which per Tom’s Hardware (September 2025), finally passed Nvidia’s qualification for 12-layer HBM3E after multiple setbacks, allowing it to begin volume shipments for the 2026 cycle.

The market environment remains constrained, with supply for calendar 2026 effectively sold out across the industry. Per S&P Global (June 2026), structural supply constraints are intensifying as HBM production requires more than three times the wafer capacity of standard DRAM. This shortage has prompted major cloud providers to explore non-conventional agreements, including offers to directly fund production lines and purchase AI infrastructure tools to ensure their own hardware supply through 2027.