Samsung and SK hynix have moved DRAM (working memory) and HBM (AI accelerator memory) supply into 5 year, advance paid deals with a 12 month renegotiation window, restructuring how memory chip downturns get absorbed.
Samsung and SK hynix, the world's two largest memory-chip makers, are no longer selling the bulk of their DRAM, HBM, and enterprise SSD supply as spot-priced commodities. Instead, they are routing the majority of those volumes into long-term contracts that run five years, are paid for in advance, and carry a 12-month renegotiation clause.
The architecture is prepaid downside insurance for the next memory recession: cash up front, multi-year term, and a 12-month window that lets the buyer reopen the deal. The escape hatch is real, and the buyer's pool outside Samsung is thin enough that a missed renegotiation from a single hyperscaler could still drag the cycle back into spot pricing.
DRAM is the working memory that lets servers and AI accelerators shuffle data in and out of processors. HBM, or high-bandwidth memory, is a stacked version of the same category, built specifically to feed the fastest AI chips. Both categories have ridden the generative-AI buildout into an extended shortage, and both are now being routed through a contract structure that Korean trade press calls the LTA, or long-term agreement. TheElec reported that Samsung has been pushing minimum five-year LTAs with AI customers, with advance-payment terms baked in, deepening those LTAs across HBM and enterprise SSDs, not just DRAM. TrendForce has separately tracked the same shift, noting that Samsung and SK hynix have deepened LTAs as buyers pay premiums to secure supply, a structure the research firm now treats as a pricing variable in its own right.
The advance-payment condition is the part that makes this a model redesign rather than a marketing move. A buyer that posts cash against a five-year commitment changes the working-capital profile of the supplier, and converts what used to be cyclical revenue into something closer to a prepaid annuity. The 12-month renegotiation window reintroduces the cycle on the buyer's side. Once a year, the customer can reopen the contract on price and volume. Korea JoongAng Daily framed this as the central tension in the new architecture, asking whether pricing, profit, and cycle risk can be rebalanced inside a single five-year term. The supplier gets the cash and the commitment. The buyer gets the supply, with a yearly tripwire back to spot pricing.
The second leg of the redesign is the contract cap. SK hynix has reportedly removed the price cap in its memory LTAs, diverging from Micron's approach, according to TrendForce. A capped contract is a hedge for the buyer; an uncapped contract is closer to a floating subscription. TrendForce treats the divergence as a competitive choice: SK hynix is willing to absorb more cycle risk in exchange for longer customer lock-in, while Micron is keeping the price ceiling as a backstop. Samsung's posture sits in the middle, and Seoul Economic Daily reported that Samsung now plans to supply roughly 70% of its memory under long-term contracts, a figure that converts the architectural choice into a market share number.
Outside Samsung, the realistic LTA counterparties are SK hynix, Micron, and China's CXMT, a state-backed DRAM maker that has spent the last several years building a domestic alternative to the Korean duopoly. CXMT DDR5 is reportedly priced above Samsung's, which leaves it as a backstop in name more than in practice. A 12-month renegotiation window sounds protective on the supplier side until the same window is read from the buyer's side, where a hyperscaler that hits a model-training slowdown can reopen the contract, trim volume, and force the supplier back into the spot market with a year of prepaid capacity still on the books.
The Korean memory industry has tried cyclical fixes before: capacity discipline, inventory write-downs, quiet supply cuts. This is the first time the two biggest players have tried to change the contract that the cycle rides on. The next test is whether the five-year term holds, the advance payment sticks, and the renegotiation window stays open for renewal. The first renegotiation cycle will show which side the model actually favors.