SK Hynix Lists on Nasdaq as SKHY: The $26.5B AI Memory Play

SK Hynix Lists on Nasdaq as SKHY on 10 July 2026

SK Hynix, the South Korean chipmaker that manufactures roughly 60 per cent of the world's high-bandwidth memory, listed American depositary receipts on Nasdaq on 10 July 2026 under the ticker SKHY. The company priced 177.9 million ADRs at $149 each, raising approximately $26.5 billion — the largest ADR offering in Wall Street history. That record came less than a month after SpaceX completed an $85.7 billion capital raise in June 2026, the largest stock listing ever recorded globally. SKHY opened at $170, a 13 per cent premium to the issue price, and SK Hynix's chairman told CNBC on the day of listing that demand from AI infrastructure builders was enormous. The offering was oversubscribed by approximately seven times, with cornerstone investors Baillie Gifford and Coatue Management committing a combined $7 billion in expressed interest ahead of pricing.

The listing converts one of the world's most strategically critical semiconductor companies from a Korea Stock Exchange-only stock into a directly accessible US-traded security. For institutional investors running dollar-denominated portfolios, the listing removes the currency conversion overhead and settlement friction that previously made South Korean equities expensive to hold at scale. By bridging this accessibility gap, SK Hynix is targeting the elimination of what analysts call the Korea discount — the longstanding suppression of South Korean semiconductor valuations relative to American peers such as Micron.

What High-Bandwidth Memory Is and Why AI Cannot Run Without It

High-bandwidth memory is a form of DRAM in which multiple chips are stacked vertically and connected through thousands of microscopic copper pathways called through-silicon vias. The stacked configuration places memory physically adjacent to the processor — in the case of AI accelerators, on the same chip package as the GPU die — and connects them through an interposer that delivers memory bandwidth far beyond what conventional DRAM provides over a standard bus.

For AI inference and training, memory bandwidth is one of the two primary performance ceilings alongside raw compute throughput. Large language models hold billions of parameter weights in memory and read them into the processor for every token generated. NVIDIA's H100 and H200 GPUs use multiple stacks of HBM3e per chip, delivering memory bandwidth in excess of three terabytes per second per device. The number of HBM stacks an AI accelerator receives is constrained not by DRAM wafer supply but by advanced packaging capacity — which is why HBM has remained both strategically critical and persistently supply-constrained through the current AI data centre buildout.

SK Hynix holds approximately 60 per cent of the global HBM market and is NVIDIA's lead supplier for HBM3e — the memory configuration used in the H100, H200, and Blackwell GPU families. HBM is sold out well into 2027, reflecting the pace at which hyperscalers including Microsoft, Google, Amazon, and Meta are adding AI compute capacity. SK Hynix reported Q1 2026 revenue exceeding 52 trillion Korean won, with an operating profit margin above 70 per cent — margins made possible by HBM's commanding price premium over commodity DRAM.

The Capital Raise: What $26.5 Billion Is Funding

SK Hynix stated that the proceeds from the Nasdaq ADR offering are earmarked for three specific capacity investments. The first is Phase 1 of the Yongin Semiconductor Cluster in South Korea, a multi-facility campus designed to expand advanced memory production for AI data centre demand through the end of the decade. The second is the Cheongju P&T7 advanced-packaging plant, which will assemble HBM stacks and CoWoS interposer packages for AI accelerator customers. The third is procurement of ASML EUV lithography systems, which SK Hynix requires to pattern next-generation HBM dies at the tighter geometries needed for HBM4 and beyond.

All three investments are capacity-expansion bets on a single thesis: demand for HBM from hyperscale AI data centres will continue to grow faster than SK Hynix can currently supply. The company has stated publicly that every unit of HBM it produces through at least 2027 is already committed to customers.

What the SKHY Listing Means for India's AI Infrastructure Plans

For India's AI infrastructure ecosystem — being built under the IndiaAI Mission's shared compute programme and through private GPU cloud providers including Tata, Reliance, and emerging domestic GPU operators — the SK Hynix listing has two practical implications.

The first is supply trajectory. HBM scarcity is one of the binding constraints on the number of NVIDIA H100 and H200 GPUs that can be manufactured and delivered globally. IndiaAI Mission's 45,000-GPU shared compute facility and the private GPU investments accompanying it all depend on continued HBM output expansion. SK Hynix's $26.5 billion capacity programme is constructive for that pipeline: more Yongin and Cheongju output means more HBM available for NVIDIA and AMD to package into GPUs for global customers, including Indian infrastructure buyers.

The second implication is for engineering teams building software on those GPUs. HBM bandwidth is the reason certain model architectures and batch sizes run efficiently on accelerators — and why memory-efficient inference techniques such as quantisation, KV cache compression, and flash attention matter at production scale on India-hosted GPU infrastructure. Understanding the hardware bottleneck that SK Hynix is paid to solve helps engineering teams make smarter choices about model size, batching strategy, and inference stack configuration when the cost per GPU-hour is their primary constraint.

The Bottom Line

SK Hynix listed on Nasdaq as SKHY on 10 July 2026, raising $26.5 billion — the largest ADR in Wall Street history. The company priced 177.9 million ADRs at $149, opening at $170 on a seven-times-oversubscribed book anchored by Baillie Gifford and Coatue. SK Hynix holds approximately 60 per cent of the global HBM market and is NVIDIA's lead supplier for the high-bandwidth memory that powers the H100, H200, and Blackwell AI accelerators. Q1 2026 revenue exceeded 52 trillion Korean won at an operating margin above 70 per cent. The $26.5 billion proceeds will fund the Yongin semiconductor cluster, the Cheongju P&T7 packaging plant, and ASML EUV lithography equipment — all aimed at expanding the HBM supply that AI builders worldwide, including India's IndiaAI Mission programme, depend on to deploy AI compute at scale.