How HBM Helped SK hynix Overtake Samsung in Market Value

SK hynix overtook Samsung Electronics by ordinary-share market capitalization on June 22, 2026, becoming South Korea’s most valuable listed company under that measure. Reuters’ account of the June 22 close put SK hynix at 2,080.4 trillion won, or about $1.35 trillion, against 2,066.7 trillion won for Samsung’s ordinary shares; including Samsung’s preferred shares raised its combined value to 2,246.4 trillion won.

The qualification remains essential: SK hynix achieved a striking valuation reversal, but it did not become larger than Samsung under every market-cap calculation. What has strengthened since that milestone is the financial case behind the rerating. SK hynix subsequently posted record second-quarter revenue and operating profit as it began mass shipments of its latest high-bandwidth memory generation.

A valuation reversal, not a verdict on corporate scale

Market capitalization reflects the stock market’s valuation of a company’s equity. It is not a direct ranking of revenue, physical assets, workforce or the breadth of a business, and different share classes can change the comparison.

Samsung remains a diversified electronics group with businesses spanning smartphones, televisions, home appliances, displays, contract chipmaking and memory. SK hynix is far more concentrated on memory products. Its higher ordinary-share valuation therefore shows how strongly investors have valued exposure to the memory required by AI systems, rather than proving that SK hynix has surpassed Samsung across the wider electronics industry.

That concentration cuts both ways. A focused producer benefits disproportionately when memory prices, factory utilization and sales of premium products rise together. It is also more exposed when customers reduce inventories, new capacity creates oversupply or a weak economic cycle pushes memory prices down.

HBM changed what a memory supplier could sell

Conventional DRAM has historically been vulnerable to commodity economics. Manufacturers commit large sums to fabrication capacity, while compatible products from different suppliers can often compete primarily on price and availability. Profits can rise sharply during shortages and fall just as quickly when production overtakes demand.

High-bandwidth memory changes that competitive equation. HBM places multiple DRAM dies in a vertical stack, linked through through-silicon vias and micro-bumps. The stack is installed close to a processor through an interposer, creating a wide interface that supplies much more bandwidth without requiring a large collection of separate memory packages around the chip.

The technology had a commercial life before the current AI investment cycle. AMD’s 2015 technical account of its SK hynix partnership documents HBM-equipped Fiji graphics processors with up to 4GB of memory, a 4,096-bit interface and 512GB-per-second bandwidth, as well as the use of an interposer, through-silicon vias and micro-bumps.

That history makes the rise of SK hynix more than a story about fortunate timing. The company and its partners had to develop stack design, packaging, manufacturing and validation capabilities while HBM served a comparatively limited graphics market. Those accumulated capabilities became much more valuable when AI accelerators began demanding enormous volumes of data from memory.

Why AI memory is harder to substitute

An AI accelerator’s processors and memory operate as a system. Training and inference repeatedly move model parameters and intermediate results, so processor performance can be constrained when the memory subsystem cannot deliver data quickly enough. HBM addresses that bottleneck by supplying high bandwidth close to the processor.

Selling it requires more than producing DRAM dies that meet a capacity specification. A completed stack must satisfy performance, power, thermal, packaging and manufacturing-yield requirements, while also working within the customer’s accelerator package. Qualification therefore creates technical and commercial barriers that are higher than those surrounding interchangeable commodity memory.

This integration can give an established supplier better visibility into customer demand and more differentiation. Replacing one qualified HBM product with another may require package changes and additional validation, limiting a buyer’s ability to switch suppliers at short notice. The advantage is not permanent, but it gives proven production capability considerable economic value while supply remains constrained.

Record results gave the rerating an earnings foundation

SK hynix’s July 29 earnings release records second-quarter revenue of 79.3187 trillion won and operating profit of 60.5426 trillion won, both quarterly highs. First-half revenue crossed 100 trillion won for the first time, long-term agreements had been finalized with around ten customers, and HBM4 mass shipments began during the quarter.

Those figures connect the market-cap milestone to operating performance rather than enthusiasm alone. A richer product mix, higher memory prices and demand from AI infrastructure customers produced a level of profitability that would have been difficult to imagine when HBM was confined mainly to premium graphics products.

The figures also require caution. The release identifies them as preliminary results based on consolidated K-IFRS and notes that the independent review had not been completed. More fundamentally, one exceptional quarter cannot establish that memory-industry cyclicality has disappeared.

The turnaround was built through several cycles

SK hynix began as part of Hyundai and later became a heavily indebted memory producer under creditor control after a proposed sale to Micron collapsed. Its eventual acquisition by SK Group provided a new owner, but ownership alone did not create its present position.

The company continued investing in specialized memory and advanced packaging through periods when the immediate return was uncertain. That persistence mattered because semiconductor advantages depend on accumulated manufacturing knowledge: stable yields, repeatable packaging, customer qualification and the ability to deliver large volumes cannot be assembled instantly when demand appears.

The result was a shift from distressed commodity producer to a central supplier of AI memory. HBM did not eliminate the risks of a concentrated business; it changed the market’s assessment of what that concentration was worth.

The lead is valuable, but it is not protected forever

Samsung and Micron remain formidable competitors with capital, engineering resources and existing customer relationships. HBM specifications will continue to evolve, and leadership in one generation does not guarantee the same position in the next. Production yields, qualification schedules and packaging capacity can alter the competitive order.

Capacity expansion presents another tension. SK hynix needs enough output to serve growing AI deployments, yet memory manufacturers have repeatedly damaged industry returns by adding more supply than customers ultimately required. Long-term agreements can improve visibility, but they cannot remove technology transitions, customer concentration or the risk of slower infrastructure spending.

The defensible conclusion is therefore narrower than the underdog mythology. SK hynix built an early commercial base in stacked memory, preserved the required engineering and manufacturing capabilities, and reached the AI investment cycle with products that customers could not easily replace. Its June 2026 ranking depended on the share classes counted, but the financial transformation behind it was real.

SK hynix Overtook Samsung—But Only Under One Market-Cap Measure