SK Hynix, the renowned South Korean memory chip manufacturer, has recently unveiled its collaboration with NVIDIA to engineer high-performance solid-state drives (SSDs) tailored specifically for AI inference tasks. This innovative product is designed to shatter the performance limitations of current enterprise-grade SSDs, with an ambitious goal of achieving up to a tenfold increase in performance.

The project, internally codenamed 'AI-NP,' leverages the cutting-edge PCIe Gen 6 interface. Initial samples of these groundbreaking SSDs are slated for release by the end of 2026, boasting an impressive 25 million IOPS (Input/Output Operations Per Second). The second-generation iteration is projected to commence mass production by the end of 2027, delivering a staggering performance of up to 100 million IOPS—a more than 30-fold enhancement compared to existing offerings.

At present, both NVIDIA and SK Hynix are immersed in proof-of-concept endeavors, with a keen focus on optimizing energy efficiency and speed during extensive data throughput operations. Moreover, SK Hynix is spearheading the standardization of high-bandwidth flash memory (HBF), a technology that operates on a principle akin to HBM (High Bandwidth Memory) by stacking NAND layers to augment bandwidth. The Alpha version of HBF is anticipated to make its debut by the end of January 2026.