SK HYNIX US IPO — KEY FACTS
● Listing date: July 10, 2026 — NYSE
● Offering size: $29.4 billion — largest US equity listing since SpaceX's $75B in June
● What SK Hynix makes: HBM3E high-bandwidth memory — the memory stack inside every Nvidia H100, H200, and B100 AI accelerator
● HBM revenue share: Over 40% of SK Hynix total revenue — up from under 5% in 2022
● Market position: Dominant HBM supplier — approximately 50% market share ahead of Samsung and Micron
● Why it matters: First major AI infrastructure equity offering after SpaceX. Demand signals will indicate whether public markets will support the Anthropic and OpenAI IPOs expected in Q4 2026.
What SK Hynix Makes and Why It Matters for AI
SK Hynix is not a name most people outside the semiconductor industry know. It should be. Every Nvidia GPU powering Claude, GPT-5.5, Grok, and Gemini — every H100, H200, and B100 AI accelerator — contains SK Hynix HBM3E high-bandwidth memory stacked directly on top of the GPU die. HBM is not optional for AI training. Standard DDR5 memory cannot deliver the memory bandwidth that transformer models require at scale. HBM3E at 1.2TB/s bandwidth per stack is the only memory technology that keeps frontier AI accelerators from being bottlenecked by memory access rather than compute.
SK Hynix's HBM revenue share went from under 5% of total company revenue in 2022 to over 40% by mid-2026. That is one of the fastest major revenue mix shifts in semiconductor history. The driver: Nvidia shipped approximately 3.6 million H100 and H200 units in 2025, each requiring 6-8 HBM3E stacks. Every Colossus cluster, every Amazon Trainium2 farm, every Google TPU v5 alternative that uses HBM is a SK Hynix revenue event. Anthropic pays SpaceX $1.25 billion per month for Colossus compute. That compute runs on chips that run on SK Hynix memory. The AI spend chain runs through SK Hynix whether or not the AI company knows it.
Why the Listing Matters Beyond SK Hynix
The SK Hynix US IPO is the first major AI infrastructure equity offering since SpaceX's $75 billion IPO in June. Both Anthropic (S-1 filed June 1, $965B target valuation) and OpenAI (S-1 filed June 8, $830B-$1T target) are planning IPOs before the end of 2026. Before they can test public market appetite for AI software companies, SK Hynix tests public market appetite for the infrastructure layer those companies depend on. If SK Hynix trades well above its offering price on July 10, it signals that institutional investors remain willing to pay premium multiples for AI infrastructure exposure. If it struggles, it raises questions about whether the AI spending boom has already been priced into public markets.
The timing is also significant relative to the enterprise AI spending revolt documented by CNBC in June — several companies switching from Anthropic and OpenAI to DeepSeek to cut costs. If enterprise AI spending is rationalising, the demand for AI accelerators (and therefore HBM) could moderate. SK Hynix's order book for HBM3E through 2026 is reportedly fully committed, meaning near-term demand is locked. The question the IPO tests is whether public market investors believe the AI infrastructure spending at this scale continues into 2027 and beyond.
The HBM Market — SK Hynix's Position vs Samsung and Micron
| Company |
HBM market share |
Status |
| SK Hynix |
~50% ✓ market leader |
Dominant Nvidia HBM3E supplier. US listing July 10. |
| Samsung |
~35% |
Qualified for HBM3E in early 2026 after quality delays. Growing share. |
| Micron |
~15% |
Amazon custom silicon ($20B ARR) uses Micron HBM. Growing fast in US-based AI infrastructure. |
SK Hynix's 50% HBM market share is sustainable in the near term because HBM manufacturing requires yield expertise that takes years to build. Samsung's HBM3E qualification delays in 2025 illustrate the difficulty: HBM stacking requires stacking 12 DRAM dies with through-silicon vias at tolerances that standard DRAM manufacturing does not require. SK Hynix built this capability first and has the highest yields. The risk to market share is Samsung's scale — if Samsung resolves its yield issues and begins producing HBM3E at full throughput, SK Hynix's pricing power compresses.
What to Watch on July 10
First-day trading performance: A premium above the $29.4B offering price signals institutional confidence in sustained AI infrastructure spending. A flat or negative open raises questions about AI capex rationalisation entering H2 2026.
Order book commentary: SK Hynix management will likely address HBM3E order visibility through 2027 in the listing day press release and early analyst calls. Forward demand visibility is the most important data point for assessing whether the AI memory boom extends.
Signal for Anthropic and OpenAI IPOs: Both companies are planning Q4 2026 IPOs. A strong SK Hynix debut supports the thesis that AI infrastructure has durable public market demand. A weak debut raises the question of whether the AI IPO window that opened with SpaceX is already closing.
Sources: LLM Stats, SK Hynix investor relations, Reuters · Related: Anthropic IPO: Freshfields hired, $510B global VC record → · July 3: OpenAI 5% equity stake, Anthropic overtakes OpenAI → · Today's full AI news digest →