Samsung begins mass production of AI server SSD for Nvidia's Vera Rubin

Samsung Electronics is mass-producing an enterprise solid-state drive (SSD) for Nvidia's next-generation AI platform, Vera Rubin, the Korean chip giant announced Wednesday. 

Samsung already supplies high bandwidth memory (HBM) for the platform, and the new drive extends its lineup into AI server storage.

The company said it has started mass production of the PM1763, a PCIe 6.0-based enterprise SSD. It unveiled the drive at Nvidia's GTC 2026 developer conference in March as a memory solution for Vera Rubin, alongside its sixth-generation HBM4 and the Small Outline Compression Attached Memory Module 2, also known as Socamn2.  

Enterprise SSDs store large volumes of data in AI servers and feed it quickly to GPUs and CPUs. As generative AI models have grown larger, data transfer speed, along with computing power, has become a factor that determines a system's performance.

The PM1763 uses Samsung's ninth-generation V-NAND and a newly developed 4-nanometer controller. In its 16TB configuration, it reaches sequential read speeds of up to 28,400MB per second and write speeds of up to 21,900MB per second, more than twice as fast as its predecessor, the PM1753. At that speed, the drive can transfer a 40GB large language model file in about 1.4 seconds, Samsung said.

Power efficiency is more than 1.8 times better than the previous generation. Samsung designed the drive for the liquid-cooled environments common in AI servers, and it supports security features including post-quantum cryptography, which guards against future quantum computing threats.

"Built on an industry-leading performance, the PM1763 has successfully completed validation for next-generation AI platforms and is well positioned to support evolving AI infrastructure requirements," said Choi Jang-seok, the vice president and head of Memory Product Planning at Samsung Electronics. "As AI models continue to grow in size and complexity, the PM1763 will serve as a key solution that enables customers to efficiently scale memory capacity and optimize AI operations."

BY PARK YOUNG-WOO [[email protected]]