PCIe Gen6 SSDs Are Here:
Doubling the Speed of Enterprise Storage

📐 Part 1: The PCIe 6.0 Specification – What Changed?

The PCIe 6.0 specification was finalised back in 2022, but bringing it to market required overcoming significant engineering challenges: new controller designs, signal integrity at higher frequencies, and platform support[reference:0].

The key technical leap is the shift from NRZ (Non-Return-to-Zero) to PAM4 (Pulse-Amplitude Modulation with 4 levels) signalling[reference:1]. This allows each lane to carry two bits per cycle instead of one, doubling the raw data rate without increasing the clock frequency. The result: each PCIe 6.0 lane runs at 64 GT/s, giving a standard x4 SSD link a theoretical one-way bandwidth of ~32 GB/s—double that of PCIe 5.0's ~16 GB/s[reference:2].

Gen5 drives were already pushing the limits of their interface, with the fastest models topping out at around 14–14.5 GB/s[reference:3]. The new Gen6 drives now saturate the x4 link, achieving 28+ GB/s reads—fully utilising the available bandwidth[reference:4].

🔬 Part 2: Samsung PM1763 – The First to Mass Production

On July 8, 2026, Samsung announced it had begun mass production of the PM1763, the industry's first PCIe 6.0-based enterprise SSD[reference:5][reference:6]. This drive is optimised for next-generation AI and HPC server environments[reference:7].

🔧 PM1763 Specifications

The PM1763's performance is a significant step up from its predecessor, the Gen5 PM1753, which topped out at 14,500 MB/s reads[reference:26]. Samsung claims the new drive can transfer a 40GB large language model in approximately 1.4 seconds—down from 2.7 seconds on the previous generation[reference:27][reference:28].

The drive is also built with security in mind for AI workloads. It supports post-quantum cryptography (PQC) algorithms to protect against future quantum computing attacks, as well as TDISP (TEE Device Interface Security Protocol), which secures data pathways in virtualised environments[reference:29][reference:30].

🧪 Part 3: Micron 9650 – The First to Market

Micron was the first to announce a PCIe Gen6 SSD, with the Micron 9650 unveiled at Computex 2026 and already in mass production[reference:31][reference:32]. While Samsung's PM1763 is also in production, Micron's 9650 holds the distinction of being the world's first PCIe 6.0 data centre SSD[reference:33].

🔧 Micron 9650 Specifications

Micron's 9650 is clearly read‑optimised for AI inference and training workloads[reference:50]. The drive's write performance—14 GB/s sequential and 900K IOPS random—is more modest, reflecting the asymmetric nature of many AI workloads[reference:51].

🌡️ Part 4: The Cooling Challenge – Liquid Cooling Becomes Standard

Both Samsung and Micron have optimised their Gen6 SSDs for liquid cooling. This is a notable shift from previous generations, where air cooling was sufficient for most enterprise drives.

Micron's E1.S form factor is not just liquid‑cooling capable, but liquid‑cooling is all but required to achieve the rated performance[reference:52]. The drive is designed to be paired with high‑performance GPU server environments, where liquid cooling is similarly becoming the norm[reference:53]. Micron's performance figures assume a 25‑Watt power consumption—comparable to Gen5 drives, but delivering double the performance per watt[reference:54][reference:55].

Samsung has similarly optimised the PM1763 for direct‑to‑chip (D2C) cooling technology, which places cooling plates directly on the SSD's core components to sustain peak performance under intensive workloads[reference:56][reference:57]. The company also claims a >1.8× improvement in power efficiency over the Gen5 PM1753, helping reduce data centre operating costs[reference:58].

🚀 Part 5: Why This Matters for AI Infrastructure

The arrival of PCIe Gen6 SSDs comes at a critical moment for AI infrastructure. As GPU compute speeds continue to climb, storage I/O bandwidth has become an ever‑larger bottleneck[reference:59]. Large language models are growing in size and complexity, and the time taken to load model weights from storage to GPU memory directly impacts training and inference efficiency[reference:60].

With Gen6 drives, a 40GB LLM can be transferred in approximately 1.4 seconds on the Samsung PM1763, compared to 2.7 seconds on Gen5[reference:61]. This reduction in data latency helps minimise idle GPU time and improve overall AI processing efficiency[reference:62].

For data centre operators, the improved power efficiency means more throughput per watt—a critical metric in power‑constrained environments[reference:63][reference:64]. And with both drives supporting post‑quantum cryptography, they are future‑proofed against the eventual arrival of quantum computers that could break today's encryption[reference:65].

🔮 Part 6: What's Next – Consumer Drives and the Gen6 Ecosystem

While enterprise drives are now shipping, consumer PCIe Gen6 SSDs are still some way off[reference:66]. The ecosystem is still maturing: there are no shipping server platforms with PCIe Gen6 connectivity yet[reference:67]. NVIDIA and AMD are reportedly targeting 2027 for Gen6 server platforms[reference:68].

Other NAND manufacturers are also on the roadmap. SK hynix, Kioxia, and Solidigm have PCIe 6.0 on their roadmaps, but Samsung and Micron are the first to bring drives to market[reference:69].

For consumers, the wait will be longer. The high cost of Gen6 controllers, the cooling requirements, and the lack of consumer platforms mean Gen6 SSDs will likely remain an enterprise‑only technology for the next 12–18 months.