Samsung Electronics has begun mass production of enterprise solid-state drives (eSSD) to supply for Nvidia's next-generation artificial intelligence (AI) platform "Vera Rubin." The company is expanding its supply scope beyond high bandwidth memory (HBM) to storage for AI servers, strengthening its push into the next-generation AI infrastructure market.
Samsung Electronics said on the 8th it will mass-produce the PCIe (Peripheral Component Interconnect Express) 6.0-based eSSD "PM1763," optimized for AI infrastructure. PCIe 6.0 adopts PAM4 signaling to provide twice the data transfer bandwidth of PCIe 5.0.
Samsung Electronics said at Nvidia's annual AI developer conference "GTC 2026" in March that it would supply the next-generation AI platform Vera Rubin with the sixth-generation HBM, HBM4, SOCAMM2, and the PM1763. The mass production of the PM1763 means Samsung Electronics' supply of memory solutions for Nvidia's next-generation platform has moved into the actual product stage.
The role of eSSDs in AI servers is growing. As data needed for AI model training and inference increases, reducing the time that graphics processing units (GPUs), central processing units (CPUs), and AI accelerators wait for data has become important. Enterprise SSDs are used as key components that quickly store and retrieve large-scale data to reduce delays in AI tasks.
The PM1763 is equipped with ninth-generation V-NAND and a new controller built on a 4-nanometer process. It features improved data processing efficiency through fast read speeds and controller architecture optimization. Samsung Electronics expects this product to become a core storage solution for next-generation AI platforms.
The product comes in three capacities: 4TB, 8TB, and 16TB. Of these, the 16TB model delivers the industry's best performance, according to Samsung Electronics. For the 16TB model, sequential read speed is up to 28,400MB per second, and sequential write speed is up to 21,900MB per second. That is about twice as fast as its predecessor, the "PM1753."
Samsung Electronics said the PM1763 can transfer a 40GB large language model (LLM) in about 1.4 seconds. A 4-bit quantized Llama 3 70B model corresponds to about 40GB. Faster AI model transfer speeds can reduce data latency between accelerators and processors and improve task processing efficiency.
Cooling performance is also tailored to next-generation AI server environments. The PM1763 uses D2C (Direct-to-Chip) liquid cooling, which attaches a cold plate directly to the device. It is designed to maintain peak performance for extended periods even under heavy loads.
Power efficiency has improved by more than 1.8 times over the previous model. Because AI data centers bear high server power usage and cooling expense, improved eSSD power efficiency also leads to lower operating costs.
Security features have also been strengthened. Samsung Electronics applied a PQC (Post-Quantum Cryptography) encryption algorithm to the PM1763 to counter Quantum Computing hacks. PQC is a technology that addresses weaknesses in existing encryption algorithms vulnerable to quantum computer attacks.
The product also applies TDISP (TEE Device Interface Security Protocol) technology, which protects data paths in virtualized environments. TDISP is a PCI-SIG standard technology that blocks unauthorized external intervention when SSD resources are allocated to host resources and a connection is formed.
Market research firms expect the eSSD market to grow rapidly as AI servers expand. TrendForce analyzed that due to DRAM capacity limits and rising expense, eSSDs are becoming the core memory tier of AI infrastructure. Omdia projected that the global eSSD market will grow from $24.1 billion last year to $154 billion this year.
Samsung Electronics holds the No. 1 position in the eSSD market with a 35.1% share. Leveraging the mass production of the PM1763, the company plans to strengthen its leadership in the high-performance storage market for AI servers. Choi Jang-seok, senior vice president and head of product planning at Samsung Electronics' Memory Business, said, "The PM1763, based on the industry's best performance, meets the requirements of global clients' next-generation AI platforms and has successfully completed product validation," adding, "this product will be a key solution that expands memory capacity to help clients' AI models operate efficiently."