Saturday, August 8, 2026
DarkSubscribe
AI Infrastructure · News & Analysis
HomeChips & HardwareReport
Chips & Hardware · Report

SK Hynix announces $38 billion investment for comprehensive AI-focused chip manufacturing expansion.

Massive supply-chain commitment validating memory as critical infrastructure bottleneck; sustained demand for HBM and AI chips.
Trade pressSlicast · August 8, 2026 · US · Source: Google News
importance 93

South Korean memory manufacturer SK hynix has approved a 54 trillion won ($38 billion) investment in two new semiconductor fabrication plants, signaling confidence that demand for AI infrastructure will remain robust for years to come. The company will allocate 35.2 trillion won toward construction of its Y2 DRAM fab in Yongin and 19.1 trillion won for the M17 NAND flash fab at its Cheongju campus. Production from both facilities is expected to begin between 2028 and 2029.

This expansion reflects SK hynix's assessment that memory—not processors—represents the industry's most acute constraint on AI infrastructure growth. The company's decision comes on the heels of record quarterly performance, with second-quarter 2026 revenue reaching 79.3 trillion won and net income of 60.5 trillion won, driven primarily by surging demand for high-bandwidth memory (HBM) chips. SK hynix currently commands 58 percent of the global HBM market, ahead of Samsung and Micron with 21 percent each.

Memory has emerged as a critical bottleneck in AI development. Sophisticated accelerator devices require multiple HBM stacks integrated adjacent to processors using advanced packaging technologies such as TSMC's CoWoS. Simply increasing GPU production proves insufficient without corresponding gains in HBM supply. According to Omdia projections, memory chips will represent over 50 percent of global semiconductor market revenue by 2026, while TrendForce forecasts that DRAM and HBM supplies will remain constrained through 2027 despite industry-wide capacity expansions, citing limitations in advanced packaging capability.

The manufacturing bottleneck is already reshaping product roadmaps across the industry. Nvidia, for instance, is reconsidering its next-generation Rubin Ultra accelerator design, moving from the initially planned 12-Hi HBM4e configuration to 8-Hi HBM and HBM4 solutions due to concerns over reduced DRAM availability in 2027 and its potential impact on HBM module production.

Operationally, the Y2 DRAM fab will handle HBM and next-generation DRAM production, with construction commencing in July 2027 and the cleanroom opening targeted for June 2029. The M17 fab breaks ground in February 2027, with its cleanroom expected to be operational by December 2028. These projects represent a core element of SK hynix's broader expansion strategy, which includes 600 trillion won committed to the Yongin Semiconductor Cluster and 100 trillion won for Cheongju expansion, alongside ongoing construction of the neighboring Y1 fab.

Market analysts project that HBM will remain scarce for years to come, given the timeline before new manufacturing capacity comes online. HBM represents a specialty product distinct from standard DRAM, employing stacking technology, through-silicon vias (TSV), and complex packaging processes that make production exceptionally difficult.

SK hynix management has characterized this investment as groundwork for a sustained shift rather than a response to transient demand, citing Omdia forecasts indicating 19 percent compound annual growth in global DRAM and NAND demand through 2030. The company believes that memory production capacity will prove decisive in determining how rapidly AI hardware can scale across the industry in the coming years.

Read the original
SK Hynix announces $38 billion investment for… · Slicast