NVIDIA may have secured new multi-year memory supply agreements with SK hynix and Micron as the global memory shortage continues to reshape the semiconductor industry and raise concerns about the pace of AI infrastructure expansion.
According to a report from Edgewater Research, the potential agreements are believed to cover key memory products, including HBM and DRAM, both of which are critical for NVIDIA’s AI accelerators and data center hardware. The move comes at a time when demand for high-performance memory is surging, driven by artificial intelligence, cloud computing, and next-generation server deployments.
The report also notes that NVIDIA had previously reduced its DRAM memory forecast, suggesting that its upcoming Rubin AI chips could require less memory than originally expected. Even so, the company appears to be taking steps to lock in future supply as memory availability remains tight across the industry.
The current memory shortage has already had a noticeable impact on pricing. DRAM and high-bandwidth memory costs have climbed as manufacturers struggle to meet demand from AI chipmakers, server companies, and consumer electronics brands. Because building new semiconductor fabrication facilities is both expensive and time-consuming, the supply crunch is not expected to ease quickly.
New memory production capacity can take several years to come online. This creates a difficult situation for chipmakers, which must decide whether to invest billions of dollars in expansion while knowing that demand could shift by the time those facilities are operational. This pattern has long made the memory industry highly cyclical, with periods of shortage often followed by oversupply.
However, the rise of AI may be changing that pattern. Long-term supply agreements are becoming more common as major customers try to secure components years in advance. These contracts give memory producers greater confidence in future demand, making it easier for them to justify large investments in new capacity.
Industry executives have already warned that the memory shortage could continue for years. Earlier comments from PC manufacturer MSI suggested that supply constraints may last through 2026 and have already affected the consumer market. AMD has also indicated that memory prices may not fully stabilize until 2028, highlighting how long-lasting the current imbalance could become.
Edgewater Research suggests that memory pricing for the 2027 calendar year may not be finalized until late 2026. This means more long-term agreements could emerge across the industry as companies negotiate future pricing and supply commitments.
The shift toward longer contracts appears to be gaining momentum. Reports indicate that major memory suppliers such as Samsung and SK hynix are moving away from traditional one-year deals and are instead favoring agreements that span three to five years. This approach helps protect suppliers from sudden market swings while giving customers more predictable access to critical components.
For NVIDIA, securing reliable HBM and DRAM supply is especially important. Its AI GPUs depend heavily on advanced memory to deliver the performance required for training and running large language models, recommendation systems, and other demanding workloads. As hyperscalers and enterprises continue investing in AI infrastructure, memory availability could become just as important as GPU production itself.
If NVIDIA has indeed entered new long-term agreements with SK hynix and Micron, the deals would underline how competitive the AI hardware supply chain has become. The battle is no longer only about who can design the fastest chips, but also about who can secure enough advanced memory to build them at scale.
With AI demand showing no signs of slowing, the memory market is likely to remain under pressure. Long-term supply agreements may become a defining feature of the next phase of the semiconductor industry, helping major players manage uncertainty while shaping pricing, capacity, and availability for years to come.
