SK Hynix and Nvidia have announced a collaboration to develop a new SSD designed for AI workloads, targeting performance levels that could be as much as ten times higher than existing enterprise SSDs.
The collaboration was disclosed by SK hynix Vice President Kim Cheon-seong at a recent industry conference, according to Korean media outlet Chosun. He confirmed that the two companies are developing a proof-of-concept AI SSD aimed at improving data movement in AI infrastructure.
The project uses separate internal codenames at each company: Storage Next at Nvidia and AI-NP (AI NAND Performance) at SK hynix. Development is still at an early stage, with a prototype targeted before the end of 2026.
SK hynix has set an ambitious target of 100 million IOPS for the drive, well beyond the capabilities of current high-end enterprise SSDs. Reaching that level would represent a major architectural shift, narrowing the gap between memory and storage in AI systems.
The project targets a key bottleneck in modern AI data centers. Large inference models require frequent, high-speed access to extensive parameter data, placing heavy pressure on HBM and DRAM. Storage Next proposes a cost-effective pseudo-memory layer using NAND flash, combined with a controller optimized for AI computation rather than general-purpose storage.
The partnership also points to possible changes in the memory supply chain. Demand for NAND flash is already climbing as cloud and AI services expand, and a successful AI-focused SSD could add significant new pressure. Some analysts warn that if these high-performance NAND solutions become standard in AI systems, the market could face a supply crunch similar to past DRAM shortages.
The two companies are targeting two of the biggest limits facing today’s AI infrastructure: throughput and power efficiency. By pairing advanced NAND with a purpose-built controller, SK hynix and Nvidia aim to give flash storage a more active role in machine learning, one that traditional memory struggles to deliver at scale without prohibitive cost.
Maybe you would like other interesting articles?