Samsung and SK Hynix move AI customers to 3–5-year contracts with a price floor

Samsung and SK Hynix are changing the rules of the game: Samsung Electronics, SK Hynix, and other suppliers are pushing their largest customers to commit to long-term contracts for 3-5 years. The new agreements include minimum price thresholds and advance payments, which should help the industry scale up production of scarce chips faster.

Why the market is changing

HBM memory has become one of the most critical points in the AI hardware supply chain today. It is used alongside accelerators on which modern models are trained and run, and demand for such components has long been growing faster than production capacity worldwide. As cloud platforms and GPU compute providers ramp up purchases, memory manufacturers have a rare opportunity for the semiconductor market to dictate new terms to major customers.

This is not just about longer contracts. Suppliers want to lock in volumes for years to come and simultaneously protect themselves from a scenario where customers try to renegotiate prices after the next round of capacity expansion. That's why contracts include a minimum price threshold: even if the market cools down later, the base cost level remains fixed.

For buyers, this means less flexibility, but a better chance of securing the volumes of memory they need in the midst of shortage.

What changes for customers

The new scheme is particularly important for the largest cloud companies, AI infrastructure operators, and accelerator makers that need predictable supplies for years to come. When HBM becomes a bottleneck, having money to purchase is not enough: you need to reserve capacity in advance, negotiate delivery schedules, and share some risks with the manufacturer. That's why the market is gradually shifting from short buying cycles to more rigid strategic agreements right now.

• Contract terms are extended to 3-5 years
• Agreements include a minimum price threshold
• Manufacturers push for advance payments from customers
• Major customers reserve future supply volumes in advance
• Priority goes to those who lock in demand ahead of competitors

In essence, this is a transition from ordinary component procurement to a model of joint planning in the industry. If a major customer could previously negotiate based on changing market conditions, now the bet is on guaranteed access to memory. For companies building AI services, the risk of being left without needed chips could be more costly than agreeing to less favorable pricing terms. This is why HBM manufacturers currently hold a strong negotiating position.

Why advance payments are needed

The industry views advance payments under the new contracts as a source of funding for scaling production. HBM production is not a story about launching an additional production line in a few weeks. This requires investments in equipment, process tuning, improving yields, and coordination with packaging and integration partners.

When demand is already confirmed by multi-year contracts, manufacturers can invest in expansion more easily without the risk that the market will suddenly turn. For customers, such a model doesn't look meaningless either. Yes, they tie up capital in advance and agree to price constraints, but in return they get higher predictability in a critical part of the supply chain.

In AI infrastructure, downtime due to memory shortage can derail the launch of new clusters, delay server shipments, and hurt the revenue of services tied to computing. Against this backdrop, an advance becomes not so much an overpayment as a fee for reliability and a place in the queue.

What this means

The AI chip market is beginning to resemble industries where scarce capacity is reserved well in advance and paid for even before actual delivery. For cloud platforms and GPU giants, this is a signal: the battle is no longer just for the accelerators themselves, but for the memory around them in the coming years. For HBM manufacturers, it's a chance to secure high utilization and finance the next stage of growth.