SK Hynix locks in long-term supply agreements with 10 key clients and ramps HBM production to meet AI GPU demand.
SK hynix posted the second-highest operating profit margin among global tech companies in the second quarter, achieving unprecedented earnings. The company is expected to see steep growth in the second half as it maximizes the benefits of the artificial intelligence boom. Leveraging its dominance in memory supply, SK hynix has expanded long-term agreements (LTAs) that lock in major clients, including big tech firms, for up to five years on favorable terms, raising the stability of its profits. With full-scale expansion of high-value high-bandwidth memory (HBM) production underway, the market expects third-quarter revenue to surpass 100 trillion won for the first time ever, with operating profit far exceeding the second quarter.
SK hynix announced in its second-quarter earnings conference call that it has concluded LTA negotiations with more than 10 clients, including key customers, and is continuing additional discussions with major clients. Rather than imposing a single pricing structure, the company is consulting with clients in various ways to respond to market volatility, aiming to secure mid- to long-term business stability.
An LTA represents a new contract type in which memory companies and big tech firms commit to three to five years of transactions at once, replacing the previous practice of signing new contracts quarterly. Big tech firms minimize contract renewals and avoid the burden of quarterly memory price increases. SK hynix recently signed a $500 billion strategic partnership with Nvidia that includes long-term supply. According to industry sources, LTAs currently represent an estimated half of SK hynix's total revenue.
SK hynix plans to maximize profitability by improving LTAs qualitatively. Rather than fully fixing supply prices under three-to-five-year contracts with big tech firms, the company sets only a base price and includes conditions that adjust the actual price within a certain range in line with memory market price fluctuations. This approach allows SK hynix to reflect memory price increases to some degree even during the contract period, though not to the extent of previous quarterly contracts.
With memory supply expected to remain short until 2030, contract terms are likely to favor SK hynix in line with upward market price trends. An industry official noted that price adjustments are possible depending on detailed LTA conditions, and for big tech firms, securing memory supply itself takes priority over contractual details. SK hynix also plans to add deposit requirements to its LTAs, requiring clients to remit a certain amount in advance, which strengthens contract binding and accelerates fundraising for investment.
Alongside strengthening LTAs, SK hynix is pursuing full-scale ramp-up of sixth-generation HBM (HBM4) as a major business strategy for the second half. The company recently raised yield to the level of previous-generation HBM3E and began mass production. HBM4 is the latest HBM product to be mounted in Nvidia's new AI accelerator "Vera Rubin" and others, with demand from big tech firms concentrated over the remainder of this year and next year. Kim Ki-tae, vice president in charge of HBM sales and marketing at SK hynix, stated: "We are smoothly negotiating next year's HBM supply volumes and prices with major clients. As DRAM prices are rising sharply, this could also have some impact on HBM price negotiations."
SK hynix's HBM market share in the first quarter reached 58 percent, higher than Samsung Electronics and Micron at 21 percent each, though the gap narrowed from 69 percent in the same quarter last year. Samsung Electronics accelerated its pursuit by becoming the first in the industry to mass-produce and ship HBM4 in February. In response, SK hynix is focusing on timely supply and volume expansion of HBM4. The company is also supplying samples of seventh-generation HBM (HBM4E) and preparing for mass production next year.
Expanding HBM production increases not only revenue but profitability. While current memory prices are driven by general-purpose DRAM rather than HBM3E, prices are expected to soar starting with HBM4 as big tech firms competitively develop high-performance AI accelerators. HBM4E is known to cost 2 million won per unit, twice the cost of HBM4 at 1 million won.
Eugene Investment & Securities forecasts SK hynix's average selling price (ASP) for HBM this year at $1.51 per gigabit, below general-purpose DRAM at $1.61. However, next year when HBM4 and HBM4E are commercialized, HBM ASP will reach $3.22, surpassing general-purpose DRAM at $2.15. Beyond HBM, SK hynix is broadening its product lineup with "near-GPU storage," which mounts NAND flash near the graphics processing unit to assist AI data management.
SK hynix has focused its investment strategy on profitability, planning to invest all 40 trillion won raised through its recent American depositary receipt (ADR) issuance in building advanced fabs in Yongin and Cheongju, aiming to preempt high-margin next-generation memory. Over the mid to long term, the strategy is to control the pace of fab investment in line with demand to prevent oversupply and price declines. This year's capital expenditure will be executed conservatively relative to earnings growth, at the high 40 trillion won range.
Financial consensus estimates for SK hynix's third-quarter revenue this year are 102.87 trillion won, expected to surpass 100 trillion won in a quarter for the first time. Third-quarter operating profit is also forecast at 80.58 trillion won.