TITLE: Nvidia, SK hynix Partner on AI-Optimized SSD Targeting 10x Speed

Nvidia and SK hynix Advance AI Storage with High-Performance SSD

SK hynix Inc. and Nvidia Corp. are expanding their collaboration beyond high-bandwidth memory to develop a next-generation solid-state drive tailored for artificial intelligence workloads, executives said this week.

The partnership focuses on creating an SSD that could achieve up to 10 times the performance of existing enterprise-grade drives, with a target of 100 million input/output operations per second (IOPS). SK hynix Vice President Kim Cheon-seong announced the initiative at a recent industry conference, describing it as a major architectural advancement to bridge the gap between memory and storage in AI systems.

Current AI inferencing models, which process vast datasets in real time, face significant bottlenecks in data retrieval. High-bandwidth memory (HBM) and dynamic random-access memory (DRAM) used in Nvidia's AI graphics processing units struggle to handle the continuous access to model parameters at scale. The new SSD aims to function as a pseudo-memory layer, leveraging advanced NAND flash and controller technologies optimized for AI computations rather than general-purpose storage.

Project Details and Timeline

Nvidia refers to the effort as "Storage Next," while SK hynix calls it "AI-NP," shorthand for AI NAND Performance. The project is currently in the proof-of-concept phase, with plans to complete a prototype before the end of 2026.

Achieving 100 million IOPS would represent a dramatic improvement over today's top enterprise SSDs, which typically reach several million IOPS. This leap requires innovations in NAND architecture, controller design and integration with AI accelerators. The companies' prior cooperation on HBM supply for Nvidia's GPUs provides a foundation, but this venture shifts focus to flash storage innovation.

Industry analysts note that the drive's design emphasizes throughput and energy efficiency, key challenges in scaling AI infrastructure. By enabling NAND flash to play a more active role in machine learning tasks, the SSD could reduce reliance on costlier memory types while supporting larger models.

Market Implications

The announcement comes amid surging demand for storage in cloud computing and AI data centers. NAND flash supplies are already strained, with reports of data centers hoarding SSDs as hard drive availability dwindles. A specialized AI SSD could exacerbate this pressure, potentially leading to shortages similar to those seen in DRAM markets.

SK hynix, a leading memory producer, sees the project as a way to differentiate in the competitive NAND sector. Nvidia, dominant in AI hardware, benefits from enhanced storage solutions that complement its GPUs. Observers predict the technology could influence broader AI ecosystems, including edge computing and hyperscale environments.

No pricing or commercial rollout details were disclosed, but the initiative underscores the intensifying race to optimize hardware for generative AI and large language models. Both companies declined further comment beyond the conference remarks.