SK Hynix supply deals aim to steady AI memory pricing

SK Hynix said it has signed around ten long-term supply agreements with key customers, putting SK Hynix supply deals at the center of its plan to reduce swings in memory pricing while AI demand keeps capacity tight. The South Korean memory maker did not name the customers or disclose contract values, but said many are AI players and that some agreements run as long as five years.

The disclosure came with second-quarter results showing how sharply the memory cycle has turned in SK Hynix’s favor. The company reported quarterly revenue of ₩79.3 trillion, or $54.5 billion, up 257% from a year earlier. Operating profit rose 557% year over year to ₩60.5 trillion, or $41.6 billion. Net income exceeded revenue, which the company attributed to asset sales.

Pricing did much of the work. SK Hynix said the average price paid by customers for DRAM rose 30%, while NAND pricing increased 50%, alongside higher shipments. The company still guided for slower shipment growth in the third quarter, with NAND volumes expected to rise by a low single-digit percentage and DRAM shipments forecast to increase by about 10%.

What are the SK Hynix supply deals meant to do?

SK Hynix says the agreements are designed to make demand more predictable and reduce exposure to short-term memory price swings. Company president Song Hyeon-jong said customer relationships are moving from transaction-by-transaction buying toward longer strategic partnerships, which he described as evidence of continued demand from the AI ecosystem.

Song said the contracts include long-term volume commitments and mechanisms such as deposits. In the company’s framing, those deposits strengthen contract execution and give SK Hynix better visibility into future demand. The more practical reading is that large customers are offering firmer commitments so SK Hynix can fund and plan capacity without relying entirely on spot-market signals.

SK Hynix also said HBM4 shipments will rise in volume, lifting average selling prices and earnings. Executives pointed to heavy spending on additional manufacturing capacity but said they do not see an oversupply risk because AI customers are still consuming large amounts of memory and are expected to keep doing so.

Park Seong-hwan, SK Hynix’s head of investor relations, argued that memory and storage demand should remain strong even if major AI companies end up with more data center capacity than they need. He said hyperscalers leasing out data centers shows higher use of existing AI infrastructure, and said more efficient AI models could increase demand for AI services rather than reduce memory needs.

The company made a similar claim about agentic AI, saying in its earnings release that demand is shifting so both AI memory and conventional memory are growing at the same time. SK Hynix did not provide customer-level forecasts to support that claim.

Investors were not clearly persuaded by the combination of record profitability and longer customer commitments. SK Hynix’s share price was down about 5% at the time of reporting, suggesting the market is still weighing whether today’s AI-driven memory demand can justify the scale of new manufacturing capacity being built.

This story draws on original reporting from The Register.