BlockBeats reported on Aug. 2 that Bank of America released a weekend survey on the memory industry saying Samsung Electronics has brought 60% to 70% of its memory sales into long-term supply agreements. Under those contracts, quarterly price declines are usually capped at no more than 5%, while price increases can reach 10% to 20% and are largely left without an upper limit.
The report said Samsung’s LTAs with major U.S. technology companies are mainly structured as five-year rolling agreements. They can be renewed around the expiration of the first contract year, creating long-running customer ties. Bank of America said that structure increases revenue visibility for Samsung’s memory business while still allowing the company to raise prices during periods of tight supply and demand.
How the contract structure shifts pricing power
In the bank’s view, rising AI server demand and the time still needed for memory makers to release additional capacity are pushing leading suppliers to use long-term agreements to lock in demand from major customers. That setup strengthens pricing control for top manufacturers. It also limits downside in pricing without placing the same kind of ceiling on upside moves.
Spot and contract prices keep climbing
Separate data from DRAMeXchange showed 16Gb DDR5 spot prices rose 733% year over year to $51. DDR4 products posted even larger gains, ranging from 722% to 896%. Spot prices for NAND 1Tb wafers were up 415% from a year earlier to $26.4, with another 3% increase on a weekly basis.
On the contract side, 64GB DDR5 memory module prices have moved above $1,480, while DDR4 modules reached $1,300. Both marked record highs. Client SSD prices have also doubled compared with the end of 2025.
What Bank of America sees behind August pricing
Bank of America attributed support for August pricing to three factors:
- Stronger downstream restocking demand;
- Inventory preparation by OEMs for new product launches;
- Lower channel inventory.
The bank said the underlying driver is the continued wave of AI capital spending. Combined capital expenditure by the five hyperscale cloud companies is projected to reach $730 billion in 2026, up about 100% year over year. For 2027 and 2028, that figure is expected to exceed $1 trillion annually, continuing to support upward demand for memory products.