Samsung’s PM1763 PCIe 6.0 SSD Enters Mass Production for Vera Rubin

Samsung Electronics began mass production of the PM1763 on July 8, its first PCIe 6.0 enterprise SSD and the storage device built for NVIDIA’s Vera Rubin AI platform. The 16-terabyte drive hits sequential read speeds of up to 28,400 MB per second and sequential writes of 21,900 MB per second, more than double the performance Samsung quoted for its predecessor. Samsung says the drive can move a 40-gigabyte large language model in roughly 1.4 seconds.

Spec numbers tell one part of the story. Samsung also won the storage slot for NVIDIA’s Vera Rubin, the AI server platform that entered full production in June. TrendForce’s Q1 2026 numbers put Samsung at 35.1% of the global enterprise SSD market, up from 33.8% the prior quarter, while SK hynix including Solidigm slid from 30.2% to 23.1%. The 12-percentage-point gap between the two widened in a single quarter, and the PM1763 is built to keep that lead from closing.

What the PM1763 Puts on the Rack

Samsung’s official PM1763 mass production announcement positions the drive as a key storage solution for high-performance AI platforms, built on the company’s 9th-generation V-NAND and a newly developed 4-nanometer controller. The drive ships in 4TB, 8TB, and 16TB capacities. PCIe 6.0 doubles per-lane bandwidth over PCIe 5.0, from 4 GB/s to 8 GB/s, and Samsung leans on that jump to deliver 28,400 MB/s sequential read and 21,900 MB/s sequential write on the top-capacity model.

For AI workloads, throughput shows up as time-to-data. The 16TB configuration moves a 40GB large language model in approximately 1.4 seconds, a figure Samsung highlights as the way to reduce the latency gap between accelerator and storage. The drive also improves power efficiency by more than 1.8 times compared to its predecessor, the PM1753. Samsung frames that gain as a data-center operating-cost lever, and it lands directly on the procurement scoresheets cloud service providers are running for Vera Rubin.

Liquid cooling is where the PM1763 earns its place in dense AI racks. Samsung built the drive for direct-to-chip liquid-cooled server environments, the architecture Vera Rubin NVL72 racks depend on for sustained peak performance under intensive workloads. Security matches the operating envelope: the drive supports post-quantum cryptography (PQC) algorithms to defend against future quantum attacks, plus the TEE Device Interface Security Protocol (TDISP) to lock down data paths in virtualized environments. Those features address two of the biggest procurement concerns from cloud providers.

• Sequential read: up to 28,400 MB/s (16TB)
• Sequential write: up to 21,900 MB/s (16TB)
• 40GB LLM transfer time: approximately 1.4 seconds
• Power efficiency vs PM1753: more than 1.8 times
• Interface: PCIe 6.0 (doubles PCIe 5.0 per-lane bandwidth)

Why NVIDIA’s Vera Rubin Slot Is the Prize

Storage has become a first-class citizen in NVIDIA’s Vera Rubin platform. Vera Rubin, the third generation of NVIDIA’s MGX rack-scale systems, entered full production after being announced at the GTC Taipei keynote on June 1, 2026. NVIDIA describes the architecture as delivering 10x agent inference scale across the rack, a setup where the storage tier has to keep up with the compute. Samsung says the PM1763 has successfully completed validation for next-generation AI platforms, the quiet load-bearing detail in the release.

A design win at the storage slot on a Vera Rubin rack is a multi-year revenue commitment, because data-center operators standardize drives across large fleets and swap generations slowly. Samsung already ships HBM4 memory and SOCAMM2 modules into the same platform. The PM1763 plugs into a memory-plus-storage stack the company has been assembling for over a year, giving it three of the silicon lanes Vera Rubin pulls from.

How Far Ahead Samsung Already Is

The lead is not hypothetical. TrendForce’s Q1 2026 enterprise SSD revenue breakdown shows Samsung at $7.05 billion in eSSD revenue, up 92.8% quarter-on-quarter, with the company holding 35.1% of the global market. That share gain came from a market that itself grew 86.1% in a single quarter to $18.46 billion in combined top-five revenue. Samsung’s $7.05 billion alone was more than SK hynix’s $4.64 billion and Micron’s $3.09 billion combined.

The gap widened because Samsung shipped. The company migrated broadly to 236-layer NAND and pushed out 176-layer QLC products at scale, as confirmed in the Q1 2026 vendor-by-vendor eSSD market analysis. TrendForce reports eSSD contract prices jumped by approximately 80% in the quarter, the steepest single-quarter move the segment has recorded, as inventory at major suppliers fell to historic lows. SK hynix’s market share dropped from 30.2% to 23.1% despite its own revenue rising 42.5% quarter-on-quarter. Micron gained 2.4 percentage points by reallocating NAND capacity away from smartphones and retail, lifting its share from 13% to 15.4%.

Kioxia and SanDisk round out the top five. Kioxia’s revenue rose to approximately $2.22 billion as its 218-layer products qualified with North American customers, and the company is now pushing 245TB QLC drives into validation for the second half of 2026. SanDisk delivered nearly $1.47 billion, with QLC eSSDs entering volume shipment during the shortage. Neither has a PCIe 6.0 drive in mass production, and that is the lane where the next round of design wins is being decided.

The Bar Rivals Now Have to Clear

Micron’s parallel mass production of PCIe Gen6 SSDs is the clearest counterpoint to Samsung’s announcement, and the two are now racing to the same Vera Rubin slot. Micron began volume production of its 9650 drive in March, hitting up to 28 GB/s sequential read and up to 5.5 million random read IOPS, designed specifically for data-intensive AI workloads. Solidigm, under SK hynix, is accelerating its transition to 240-layer NAND while SK hynix itself develops 375-layer TLC products, both targeting AI inference demand.

None of those competitors has matched Samsung’s combination of PCIe 6.0 interface, direct-to-chip liquid-cooling optimization, and post-quantum cryptography in a single shipping drive. The longer rivals take to land a validated PCIe 6.0 product, the more quarters Samsung collects at full average selling prices while the eSSD market re-rates upward. SK hynix’s recent NVIDIA AI memory co-design deal gives it another lane into the Vera Rubin stack, but storage is a separate bid.

Why the Storage Tier Got Promoted

The PM1763 launch lands inside a market that has structurally promoted storage to a primary tier. DRAM capacity limits and rising DRAM costs are forcing AI architects to treat high-performance SSDs as a memory layer rather than a passive repository. TrendForce puts it plainly, calling SSDs a core memory tier within AI infrastructure, and that shift is what pulled eSSD contract prices up about 80% in a single quarter.

Omdia, separately, sizes the eSSD market at $24.1 billion last year and projects it to reach $154 billion this year, a sixfold jump that implies sustained momentum through 2026. Demand is being driven by cloud service providers stockpiling drives ahead of Vera Rubin deployments. AI training sets are large enough that the storage tier now accounts for a meaningful share of total rack cost, alongside HBM and accelerators.

NVIDIA’s own storage play reinforces the trend. At GTC 2026, the company introduced Inference Context Memory Storage (ICMS), an AI-native infrastructure tier designed for agentic workloads running on Vera Rubin. The PM1763 is built to sit at that tier. A first wave of Vera Rubin deployments is already being staged at hyperscale, including one AI lab’s $1.1bn Vera Rubin cloud bet that put NVL72 GPUs behind a reinforcement-learning superlearner.

PCIe 6.0 is the throughput unlock that lets storage sit closer to the accelerator. The interface doubles PCIe 5.0’s per-lane bandwidth from 4 GB/s to 8 GB/s, so a x4 link hits a theoretical 32 GB/s versus 16 GB/s previously. Samsung’s 28,400 MB/s read on the PM1763 nearly saturates that ceiling on a single drive. Direct-to-chip liquid cooling is what keeps the drive there continuously, since PCIe 6.0 drives push heat envelopes that air cooling cannot dissipate in dense AI racks.

What Customers Pay for Faster, Cooler, Quantum-Safe

The bill for moving up the stack is showing up in cloud capex. TrendForce reports that eSSD contract prices jumped about 80% in Q1 2026 as inventories at major suppliers hit historic lows. CSP procurement teams were buying ahead of Vera Rubin launches, and the supply response could not keep pace. Samsung, SK hynix, and Micron all raised prices during the quarter, with Samsung’s 92.8% revenue growth combining both volume and price.

Cooling is becoming a procurement specification rather than an afterthought. Direct-to-chip liquid cooling, the architecture Vera Rubin NVL72 racks depend on, was a niche configuration three years ago. Samsung built the PM1763 to operate in that environment continuously at peak throughput, which is what the 1.8x power-efficiency gain is paying for at the rack level. Operators that standardize on liquid-cooled AI servers are now filtering their storage shortlist by what survives direct-to-chip immersion.

Post-quantum cryptography is the newest line item. The PM1763 supports PQC algorithms aimed at quantum-era attacks and TDISP for virtualized data-path protection, features that earlier-generation PCIe 5.0 drives in this class did not offer. Samsung did not break out a price premium for PQC, but security feature lists are increasingly part of hyperscaler request-for-proposal scoring. The PM1763’s lock on Vera Rubin means Samsung collects that premium as long as the platform ships, which NVIDIA expects to do at scale through 2026.