← Back to feed

For the first time in history, all three major memory chip makers carry trillion-dollar valuations simultaneously. Samsung crossed $1 trillion on May 6. Micron followed on May 27, surging 19% in a single session — its biggest single-day gain since 2011 — to a market cap of roughly $1.095 trillion. SK Hynix got there the next day, jumping 11% to $1.08 trillion. The driver in every case: high-bandwidth memory, the stacked DRAM that sits beside every Nvidia AI accelerator, and a structural shortage that Goldman Sachs is calling one of the strongest memory upcycles ever.

The Numbers Behind the Rally

SK Hynix reported Q1 2026 revenue of 52.576 trillion won ($35.6B) and operating profit of 37.61 trillion won — a 72% operating margin, the highest in company history. Micron’s fiscal Q1 showed $13.64B in revenue, up 56.6% year over year, with its Cloud Memory Business Unit generating $5.28B at a 66% gross margin. Both companies have sold out their entire DRAM, NAND, and HBM production through 2026 as of April. Micron guided fiscal Q3 revenue to approximately $33.5B at margins approaching 81%.

SK Hynix went from a $100 billion valuation to $1 trillion in 16 months.

Why HBM Changed the Game

High-bandwidth memory is not interchangeable with conventional DRAM. Producing it requires approximately three times the manufacturing resources of standard memory. It must be stacked vertically, packaged with precision, and tested against GPU-specific thermal and bandwidth requirements. Only three companies can do it at scale: SK Hynix, Samsung, and Micron.

SK Hynix controls 57% of global HBM revenue and has secured approximately 70% of HBM4 orders for Nvidia’s upcoming Vera Rubin AI platform. The company is working with Nvidia and TSMC on HBM4 base die design — a partnership that effectively makes it a co-developer, not just a supplier. The global HBM market was $35 billion in 2025 and is forecast to reach $100 billion by 2028.

Hyperscalers are projected to spend $715 billion on AI infrastructure in 2026, up more than 70% from an already-record 2025. Every server in that buildout needs HBM. The demand math is not complicated.

The Structural Squeeze

The problem is physics and time. New fab construction takes three to four years from groundbreaking to volume production. AI infrastructure buildouts are accelerating faster than fab timelines can track.

SK Hynix’s SK Group Chairman said in March 2026 that global chip wafer shortages will likely persist until 2030, with supply deficits potentially exceeding 20%. SK Hynix has committed KRW 31 trillion ($21.5B) to its Yongin mega-cluster in South Korea, with the first cleanroom opening accelerated to February 2027 — nine months ahead of schedule. It also broke ground on a $3.87 billion advanced packaging facility in West Lafayette, Indiana.

Micron is betting on domestic manufacturing: roughly $200 billion in planned US investment, including a second leading-edge fab in Idaho, and advanced HBM packaging capabilities aimed at capturing CHIPS Act incentives and locking in US government procurement.

The Market Shift

For years, memory was treated as a commodity with brutal cyclicality. The AI wave ended that characterization for the top HBM suppliers. Pricing power is real — data center operators are paying 30–60% premiums over 2024 DRAM levels. HBM prices surged more than 40% year on year per TrendForce. Analysts at HSBC are forecasting a 40% sequential rise in SK Hynix DRAM average selling prices in Q2 2026.

The concentration at the top is extreme. Samsung, SK Hynix, and Micron control over 90% of global DRAM production. Samsung and SK Hynix together account for roughly 50% of South Korea’s KOSPI index. The next phase of the story depends on whether Samsung closes the HBM capability gap, whether Nvidia’s Vera Rubin platform lands on schedule, and whether the February 2027 Yongin cleanroom opening holds.

Memory was supposed to be cheap and abundant. AI made it a chokepoint, and the market is pricing accordingly.