SK Hynix Soars 13% in Record U.S. Debut

SK Hynix closed at $168.01 on Friday, up 13% from its $149 offering price, after raising $26.5 billion in the largest U.S. share sale ever by a foreign company. The South Korean memory maker now sits at a $1 trillion market capitalization—the world's 16th most valuable company, according to market data—and controls 58% of the global high-bandwidth memory market that every AI accelerator depends on. The stock rose 13% in its first day of trading on Nasdaq, and SK Hynix is the leader in the high-performance memory that's used in AI chips from Nvidia , CNBC reported.

The debut caps a sevenfold rally over the past year, driven entirely by a product most investors couldn't name two years ago: high-bandwidth memory, or HBM. SK Hynix's valuation has risen more than sevenfold over the past year as demand for AI infrastructure has caused a shortage in computer memory and sent prices skyrocketing , according to CNBC. Annual revenue almost tripled from 2023 to 2025, when sales reached about $65 billion, and analysts polled by LSEG expect that number to more than triple again to about $235 billion in 2026 . The company's first-quarter operating margin hit 72%—higher than Nvidia's 65%—making it the most profitable semiconductor manufacturer in history by that measure, industry data shows.

Can Three Companies Really Control the Entire AI Buildout?

They already do. Samsung, SK Hynix, and Micron – the three companies that control over 95% of global DRAM production – have systematically reallocated manufacturing capacity toward high-bandwidth memory (HBM) chips used in AI accelerators, leaving consumer-grade DRAM and NAND flash in critically short supply , according to IDC. The economics are brutal: a single silicon wafer provides 3x as much commodity DRAM as HBM , Fortune reported. That means every wafer SK Hynix dedicates to AI chips is three fewer memory modules for laptops, smartphones, or cars.

The result is what insiders are calling "RAMageddon." TrendForce said it expects average DRAM memory prices to rise between 50% and 55% this quarter versus the fourth quarter of 2025, and analyst Tom Hsu told CNBC that type of increase for memory prices was "unprecedented" . Data center demand for DRAM surged to around 50% of global consumption in 2025, up sharply from 32% five years earlier, and by 2030, AI servers are projected to account for more than 60% of global consumption , Bloomberg Intelligence estimates.

Chairman Chey Tae-won told CNBC on Friday that even when SK Hynix announced it would double capacity within five years, customers said they still need more. "All my customers said that, 'Well, that's not enough, man, and, well, we need more,'" Tae-won said. "The demand is enormous, exponentially, so I don't really see" signs that HBM demand is shrinking , he added.

The $26.5 billion raised on Friday will fund an aggressive expansion: a cluster of chip fabrication plants in Yongin that will cost $390 billion , plus a $4 billion advanced packaging plant in Indiana , CNBC reported. But new fabs take years. The new U.S.-based fabs will not be ready in time to alleviate the 2026 crisis, as facilities are not slated for mass production until 2028 or later , according to industry analysts.

What Happens When Memory Becomes the Bottleneck?

The shortage is already reshaping who gets to build AI. Large cloud providers and frontier labs with long-term supply agreements and the capital to pre-pay for future HBM allocation have secured priority access, while mid-size AI companies and startups building their own infrastructure face the longest wait times and the least negotiating leverage on pricing and delivery timelines , according to analysis from Skycrumbs. The competitive landscape is now bifurcated between the "compute-rich" and the "compute-poor," and startups that secured early Blackwell allocations are seeing their valuations skyrocket, while those stuck on older H100 clusters are finding it increasingly difficult to compete on inference speed and cost , industry reports indicate.

Nvidia's Blackwell B200 and GB200 chips—sold out through mid-2026, according to CEO Jensen Huang—each require eight HBM chips containing 12 individual DRAM dies. A single B300 GPU consumes 96 DRAM dies, and a fully configured DGX B300 system with eight GPUs requires 768 DRAM dies just for the HBM modules alone, not counting the system memory , according to tech analysis. Huang described the demand for the B200 and GB200 chips as "insane," and industry reports indicate a staggering backlog of 3.6 million units from the world's largest cloud providers alone .

The packaging process itself has become a chokepoint. TSMC North America packaging solutions head Paul Rousseau told CNBC that its most advanced method, Chip on Wafer on Substrate or CoWoS, is increasing at a stunning 80% compound annual growth rate, and Nvidia has reserved a majority of the most advanced capacity available at TSMC .

SK Hynix's competitors are scrambling. Samsung Electronics has struggled to meet NVIDIA's qualification standards for its 12-layer HBM 3 E chips due to yield and performance issues, relegating the world's largest memory maker to a tertiary position , according to industry analysis. Micron has raised its 2026 capital expenditure to $20 billion and is investing $7 billion in a new HBM assembly facility in Singapore, but at most, Micron can only meet two-thirds of the medium-term memory requirements for some customers , a company executive said.

What Changed This Week

SK Hynix's Nasdaq debut marks the first time U.S. retail investors can directly buy into the company supplying the majority of the world's AI memory. The $26.5 billion raised eclipses Alibaba's 2014 New York listing and confirms that memory—once a sleepy commodity business—now commands software-like margins and valuations. The listing also crystallizes a structural shift: AI infrastructure companies have effectively locked up HBM supply well into 2027, leaving consumer electronics and automotive sectors competing for scraps at inflated prices.

What to Watch

Regular trading under the ticker SKHY begins Monday, July 13, after the stock switches from its temporary SKHYV symbol. Option trading is expected to start two business days after the debut, allowing investors to bet against the stock—a test of whether the market believes the AI memory boom is sustainable or cyclical. Micron's factories in Boise, Idaho, are scheduled to start producing memory in 2027 and 2028, which could begin to ease supply constraints. And TSMC's Arizona advanced packaging facilities, expected to break ground this year, will determine whether the U.S. can build a complete AI chip supply chain on domestic soil by the end of the decade.