Tuesday, July 14, 2026
DarkSubscribe
AI Infrastructure · News & Analysis
HomeChips & HardwareReport
Chips & Hardware · Report

SK Hynix announces $64 billion buildout with no-cap pricing strategy ahead of Nasdaq IPO.

HBM supply commitment signals aggressive capacity expansion; pricing flexibility accelerates memory scarcity relief.
Trade pressSlicast · July 5, 2026 · US · Source: Google News
importance 70

SK Hynix is tearing up the rulebook in the memory-chip business. The South Korean giant has switched to a pricing model with no ceiling and contract terms stretching as far as five years, a radical departure from industry norms. The move signals extraordinary confidence in the AI-driven demand for High Bandwidth Memory (HBM), where it already controls roughly 62% of the market. The strategy underpins a $64.4 billion investment blitz across two new factories in Cheongju, central Korea.

CEO Kwak Noh-jung laid out the full scope of the capital programme. The bulk — 80 trillion won — will go into a NAND flash fabrication facility called M17, with construction starting in 2027 and completion targeted for 2029. The remaining 20 trillion won is earmarked for a next-generation packaging plant, P&T7, which should be ready by the end of next year. That plant will handle the complex stacking required for HBM, the specialised memory that powers Nvidia's AI accelerators. The projects are part of a broader national semiconductor megacluster worth 2.1 quadrillion won.

The new pricing policy is the real headline, however. By scrapping price ceilings for long-term customers, SK Hynix can capture the full upside of the red-hot spot market for HBM chips. The first quarter of 2026 already delivered a punchy operating margin of 72%, and the entire HBM production for that year is sold out. A June partnership with Nvidia for its upcoming Rubin platform has cemented the company's pole position, with analysts projecting a 70% share of that market.

Investors responded with a 10.88% jump in the stock on Friday, closing at 2,425,000 won — though the move came after a 9% plunge earlier in the week. The whipsaw captures the tension between extraordinary long-term promise and near-term jitters about the sustainability of the AI spending spree.

Still, the bullish case sits uncomfortably alongside well-documented risks. The memory industry has always been brutally cyclical, and the sheer scale of the new investment could backfire if the AI infrastructure buildout slows. A cooling in capital spending by cloud giants such as Meta would leave the group exposed to overcapacity.

Competitors are not standing still. Samsung and Micron are ploughing money into their own HBM efforts, and some market watchers expect a meaningful price correction after 2026. Technical challenges also loom: HBM production involves stacking multiple memory layers and connecting them microscopically, a process prone to yield losses. Any delays in ramping HBM4 volume would quickly dim the outlook.

The legal backdrop adds another layer of uncertainty. SK Hynix, alongside Samsung and Micron, faces a US class-action lawsuit over alleged price-fixing in DDR3 and DDR4 memory.

All eyes are now on the company's landmark Nasdaq listing, scheduled for July 10. The group aims to raise $29.4 billion through American depositary receipts, which would rank as one of the largest tech floats in history. The order book opens on July 6, and the final offer price will be set on July 9. That price will serve as the first real vote of confidence — or scepticism — on the entire strategy.

The stock's year-to-date gain of 258% already reflects sky-high expectations. At current levels, the shares trade 18.5% above their 50-day moving average, and the 14-day relative strength index sits at a neutral 51.6, having unwound the overbought condition from previous weeks. Yet the distance to the 52-week peak of 2,987,000 won, set on June 25, remains nearly 19% — a reminder that even after the latest bounce, the market is not wholly convinced.

Read the original
SK Hynix announces $64 billion buildout with… · Slicast