SK Hynix listed on Nasdaq today (July 10, 2026) under the when-issued ticker SKHYV — switching to regular trading as SKHY on Monday, July 13. The offering priced at $149 per American Depositary Share, raised $28.07 billion across 177.9 million ADSs, was 7× oversubscribed, and is the largest foreign company listing in US stock market history.
For AI builders, the interesting question is not the stock price. It is what the proceeds fund, and what that means for the memory supply that underpins every GPU workload you run.
What SK Hynix makes
SK Hynix is a Korean DRAM and NAND manufacturer. That description undersells its position in the AI stack.
Every NVIDIA AI accelerator — H100, H200, B100, B200, the GB200 NVLink domain — requires High Bandwidth Memory (HBM) stacked directly on the die. HBM is not a commodity component sourced from multiple interchangeable vendors. It is a precision-manufactured, thermally and electrically critical part of the chip package. NVIDIA validates specific HBM suppliers for each GPU SKU. As of mid-2026, SK Hynix supplies the HBM3E stacks inside NVIDIA’s H200 and B200 — the GPUs powering most frontier AI training and inference at scale.
SK Hynix’s 2026 HBM production capacity is entirely sold out. Every wafer is contracted, most of it to NVIDIA.
The market share number that matters
In Q1 2026, SK Hynix held 56.4% of global HBM revenue market share, per the company’s SEC F-1 filing. The market is three players: SK Hynix (~55%), Samsung (~22%), Micron (~21%). Samsung and Micron successfully qualified HBM3E for NVIDIA platforms in 2025, which dropped SK Hynix’s share from the 90% it held in 2023 — but SK Hynix remains the category leader with the longest production history in HBM and the earliest HBM4 mass production.
HBM4 — the next generation — entered mass production at SK Hynix’s Cheongju M15X fab in February 2026. HBM4 doubles interface bandwidth versus HBM3E and is designed for next-generation accelerators beyond the B200. SK Hynix is first to volume; Samsung’s qualification timeline for HBM4 remains uncertain as of July 2026.
What the $28 billion funds
The F-1 prospectus is explicit: 100% of proceeds go to manufacturing expansion and EUV lithography hardware procurement. No operating expenses, no debt retirement, no acquisitions.
The primary capital destination is the Yongin Semiconductor Cluster — a multi-fab complex in Gyeonggi Province, South Korea, originally planned to reach full build-out in 2045. SK Hynix has accelerated that timeline to 2033, pulling the completion forward by 12 years.
Phase 1 details:
| Parameter | Value |
|---|---|
| First cleanroom equipment-in | February 2027 (pulled 3 months early from May 2027) |
| Capacity added at Phase 1 completion | +360,000 wafers/month |
| Total SK Hynix monthly HBM capacity after Phase 1 | ~900,000 wafers/month |
| Full cluster investment (all phases) | KRW 600 trillion |
| Full cluster completion (accelerated) | 2033 |
Phase 1 going online in February 2027 is the next meaningful inflection point in HBM supply. Until then, capacity is structurally constrained by what exists today.
Why this matters for builders
You pay for HBM every time you run inference or training. It is embedded in the $/hr or $/token your cloud provider charges for A100, H100, H200, and B200 instances. The price of compute is in part a function of HBM supply — and HBM supply is in part a function of what SK Hynix, Samsung, and Micron can manufacture and qualify on NVIDIA’s schedule.
The Yongin Phase 1 milestone (February 2027) is load-bearing for compute pricing:
- Until February 2027, global HBM supply is roughly fixed. The existing Icheon and Cheongju fabs are running at full utilization. SK Hynix has already sold 2026 capacity. Any demand increase (new GPU SKUs, expanded hyperscaler orders, sovereign AI buildouts) competes for the same pool.
- After February 2027, 360K additional wafers/month come online. That is roughly a 67% capacity increase at the Yongin Phase 1 fab alone. If Samsung and Micron HBM4 qualifications proceed on schedule, 2027-2028 could see the first meaningful oversupply condition in HBM since AI demand accelerated in 2023.
- If HBM tips toward oversupply in 2027-2028, GPU instance spot prices and reserved capacity rates face downward pressure. If demand continues to outpace supply (which it has for three consecutive years), the Yongin expansion keeps HBM scarce.
No one can predict with precision which way this resolves. The 7× oversubscription on the SKHY ADR offering suggests the market is currently betting on continued scarcity.
The ADS structure
Each SKHY ADS represents 1/10th of one Korean-listed share (KRX: 000660). At the $149 offering price, one SKHY ADS gives you 1/10th share exposure. This structure is standard for Korean company US listings — it allows Korean shares (priced in Korean Won at the equivalent of ~$1,490 per share) to trade at an accessible per-unit price on Nasdaq.
When-issued trading (SKHYV) runs today, July 10. Regular trading (SKHY) begins Monday, July 13. When-issued shares represent a commitment to deliver — they cannot be settled until the full listing process completes on Monday.
What SK Hynix is not
SK Hynix does not make GPUs, AI accelerators, or inference chips. It does not sell AI services. It sells the memory that sits inside the chips that run those services.
This makes SK Hynix an upstream play — further from the AI application layer, more correlated to raw silicon demand volume than to any particular AI product category. If AI training and inference workloads continue to grow, HBM demand grows with them regardless of which foundation model or inference provider wins. If compute demand stalls, SK Hynix’s revenue stalls before application-layer AI companies feel it.
The supply chain view
The critical path for AI capability today runs through a small number of physical bottlenecks:
- EUV lithography machines (ASML, ~20/year global production, long lead times)
- Advanced packaging (TSMC CoWoS capacity, HBM stacking lines)
- HBM wafers (SK Hynix, Samsung, Micron — three players, qualification lag of 12–18 months per generation)
SK Hynix’s $28 billion goes into items 2 and 3. The Yongin fab investment does not materially change the EUV constraint, but it adds capacity at the packaging and HBM wafer levels — the links in the chain where SK Hynix can actually move the needle.
For builders doing multi-year infrastructure planning: the earliest date Yongin Phase 1 production can hit customer allocations is late 2027, assuming February 2027 cleanroom-ready and a ~6-9 month ramp to full production throughput. Plan your reserved capacity commitments accordingly.
SK Hynix’s Nasdaq debut is, on one level, a capital markets event. On another level, it is the company that supplies the most critical component inside every production-grade AI accelerator telling you exactly where the next round of manufacturing investment is going and when it will be ready. The Yongin Phase 1 timeline — February 2027 equipment-in — is the number to watch.