Enterprise SSDs
The rollout of PCIe 6.0 SSDs is happening faster than expected.
From February to July 2026, within just half a year, the PCIe 6.0 SSDs from the three major memory manufacturers—Micron, Samsung, and Kioxia—have all been fully launched. Coincidentally, all of them target the enterprise market exclusively, with not a single consumer product.
Meanwhile, executives at Silicon Motion Technology Corp. recently announced that the company's consumer PCIe 6.0 SSD platform is expected to launch by the end of 2027.
On one hand, enterprise products are being launched one after another; on the other hand, the consumer market has to wait at least another year and a half. How was this time gap formed? What technical routes have the three manufacturers taken? What other factors are restricting the popularization of consumer PCIe 6.0 SSDs?
1 Three Major Manufacturers Launch PCIe 6.0 SSDs
Samsung
On July 8, Samsung Semiconductor officially initiated mass production of its first PCIe 6.0 compliant PM1763 SSD. This product was first unveiled at NVIDIA's GTC conference in March 2026, exhibited alongside HBM4 and SOCAMM2, and has completed compatibility verification for NVIDIA's Vera Rubin platform.
The PM1763 is equipped with Samsung's 9th-generation V-NAND flash memory and a self-developed controller built on a 4nm process. The 16TB version achieves sequential read speeds of up to 28,400 MB/s and sequential write speeds of 21,900 MB/s—approximately double that of the previous generation PM1753. Transferring a 40GB large language model takes only about 1.4 seconds.
In terms of thermal management, the PM1763 adopts direct-to-chip liquid cooling technology, specifically optimized for liquid-cooled server environments. Energy efficiency is improved by 1.8 times compared to the previous generation. For security, it supports post-quantum cryptography and the TEE (Trusted Execution Environment) Device Interface security protocol. It is available in 4TB, 8TB, and 16TB capacities, compatible with enterprise form factors such as E1.S and E3.S.
Jang-seok Choi, Vice President of Samsung Electronics, stated in the announcement that the PM1763 has completed end-to-end compatibility verification for next-generation AI platforms, helping customers flexibly expand storage capacity and optimize AI computing efficiency.
Kioxia
On July 30, Kioxia launched the CM10 series—the first enterprise SSD featuring 10th-generation BiCS FLASH TLC memory with 332 stacked layers.
The CM10 series supports the PCIe 6.0 x4 interface with a per-lane speed of 64 GT/s, doubling the bandwidth of PCIe 5.0. Compared to the previous CM9 series, sequential read performance is improved by approximately 92%, and random read performance by about 85%.
Capacities range from 1.60TB to 61.44TB. It is available in E3.S, E1.S, and 2.5-inch form factors. The E3.S and E1.S (9.5mm thickness) versions support PCIe 6.0 and 10th-generation BiCS FLASH; the 2.5-inch version supports only PCIe 5.0 and uses 8th-generation BiCS TLC flash memory.
For thermal management, it supports direct cold-plate liquid cooling while remaining compatible with air cooling. Security specifications include SIE, SED, and FIPS 140-3 encryption options, supporting CNSA 2.0 recommended post-quantum cryptography and SPDM 1.4 device authentication.
In terms of application positioning, the CM10 series is optimized specifically for AI inference and large-scale context caching—directly targeting NVIDIA's CMX architecture. Kioxia explained that as AI model parameter scales expand to trillions and context windows extend to millions of tokens, migrating part of the KV cache from GPU memory to high-speed flash memory can provide infrastructure designers with an additional context data storage layer.
Currently, the CM10 series has been sampled to select customers and was officially showcased at the FMS (Flash Memory Summit) conference held in Santa Clara, California, USA, from August 4 to 6.
Micron
Going back to February 2026, Micron Technology officially announced that the 9650 NVMe PCIe 6.0 SSD has entered mass production. The 9650 series achieves sequential read speeds of up to 28,000 MB/s, sequential write speeds of 14,000 MB/s, and random read performance of up to 5.5 million IOPS. Compared to the previous generation 9550, sequential read and write speeds are improved by 100% and 40%, respectively.
The series offers two configurations: the PRO version targets read-intensive scenarios with capacities from 7.68TB to 30.72TB; the MAX version targets mixed workload scenarios with capacities from 6.4TB to 25.6TB. Both models have a maximum power consumption of 25W. Equipped with Micron's self-developed G9 TLC NAND flash memory, it delivers significantly higher performance per watt than PCIe 5.0 solutions at the same power consumption.
At the GTC conference in March 2026, Micron announced that the 9650 series, along with HBM4 and SOCAMM2, has entered mass production, all centered around the Vera Rubin AI platform.
02 Controller Chips in Place
The performance ceiling of an SSD largely depends on the capabilities of its controller chip. As of the first half of 2026, controller manufacturers' PCIe 6.0 deployments are fully underway.
Phison Electronics showcased its first PCIe 6.0 SSD controller, the PS5303-X3, at Computex Taipei 2026. This controller supports the PCIe 6.0 x4 interface and NVMe 2.3 protocol, achieving sequential read and write speeds of up to 28,000 MB/s and random read and write speeds of up to 6.8 million IOPS, doubling that of PCIe 5.0 solutions. It supports a maximum single-drive capacity of 2PB. With a full-load power consumption of only 7W, the energy efficiency ratio reaches 4,000 MB/s/W, twice that of PCIe 5.0 solutions. This controller will be sampled to customers in December 2026 and will enter mass production and shipment by mid-2027.
Silicon Motion's enterprise PCIe 6.0 controller, the SM8466, is built on a 4nm process, supports 16 NAND channels, achieves sequential reads of 28 GB/s and random read/write of up to 7M IOPS, and supports single-drive capacities of up to 512TB. It is planned for mass production within 2026, with the timeline aligning with the commercialization of AMD EPYC Venice processors and NVIDIA's Vera Rubin GPU platform.
InnoGrit Technology released the PCIe 6.0 controller IG5686 "Crestone" at Computex Taipei. It features sequential reads of 28 GB/s, sequential writes of 22 GB/s, random reads of 7 million IOPS, and a maximum single-drive capacity of 256TB. It fully supports SLC/MLC/TLC/QLC NAND and SCM, with NAND interface speeds up to 4800 MT/s. It targets enterprise scenarios such as AI training servers and high-performance databases. InnoGrit also simultaneously launched the CXL 3.1 Type-3 memory expansion controller IG5676, designed to build low-cost, high-bandwidth memory pooling solutions.
DapuStor's IPO fundraising investment plan disclosed in April 2026 shows an intended investment of approximately 958 million RMB in next-generation controller chips and enterprise SSD R&D, with the company's PCIe 6.0 products expected to be released and enter mass production in 2027. In its July earnings forecast, the company confirmed that the relevant R&D is progressing smoothly according to the fundraising plan.
03 Why is the First Stop for PCIe 6.0 the Enterprise Market?
The debut products from Samsung, Kioxia, and Micron all target the enterprise sector, and the mass production schedules of controller manufacturers are also primarily anchored in data centers. This phenomenon is no coincidence; behind it lies the almost "thirsty" demand for storage bandwidth from AI data centers.
The training and inference of large AI models are essentially a race in data throughput. As model parameters scale from tens of billions to trillions, and context windows extend from thousands to millions of tokens, GPU computing power is increasing, but the speed of data movement is becoming the new bottleneck.
HBM capacity is limited, and DRAM costs are too high. Once the data channel between the CPU and GPU becomes congested, no matter how strong the computing power is, it can only wait. The per-lane speed of PCIe 6.0 reaches 64 GT/s, doubling the bandwidth of PCIe 5.0, which perfectly addresses this pain point—the faster data is moved from storage to the GPU, the higher the actual utilization rate of computing power. Therefore, for AI data centers, PCIe 6.0 is not just a nice-to-have upgrade, but a critical necessity to alleviate storage bottlenecks.
How large is the scale of this demand? Global market research firm TrendForce estimates that the supply-demand gap for NAND Flash bits in 2026 will be -4% to -5% for the whole year, maintaining an industry-wide shortage pattern. Servers have become the core growth engine for storage demand; from 2026 to 2027, the proportion of server NAND bit demand will climb from 44.2% to 51.1%. Meanwhile, demand for smartphones and laptops remains weak, with smartphone production estimated to drop 15% to 20% year-on-year in 2026, and laptop shipments to decline by about 10% annually.
The industry believes that the prosperity of the consumer electronics market still has a critical impact on the overall supply-demand balance, but NAND manufacturers' capacity is continuously shifting towards the server sector. Under this supply-demand landscape, it is an extremely rational choice for memory manufacturers to prioritize the enterprise market for their debut PCIe 6.0 products. Enterprise customers are highly sensitive to the efficiency returns brought by performance improvements, and their price tolerance is far higher than that of the consumer market. Using the same controller chip and the same batch of NAND particles, manufacturing them into enterprise SSDs yields a much higher gross margin than making them into retail products.
04 How Long Until Consumer Arrival?
Enterprise PCIe 6.0 has entered the mass production stage, but the wait for the consumer market may be longer than expected.
The industry points out that the popularization of consumer PCIe 6.0 depends on two conditions: first, the availability of controller chips, and second, support from client platforms. Regarding controllers, enterprise products will enter mass production sequentially from late 2026 to 2027, while consumer controllers will be much slower—Silicon Motion's consumer platform is expected to launch by the end of 2027, Phison has not announced a clear timeline, and InnoGrit's IG5686 is explicitly targeted at the enterprise sector. The controller is the core of an SSD; without consumer controllers in place, consumer SSDs are out of the question.
In terms of platforms, current mainstream client platforms from AMD and Intel only support up to PCIe 5.0, with no clear support plans for PCIe 6.0 yet.
Furthermore, the industry generally believes that consumer scenarios do not have an urgent need for extreme read and write speeds. The sequential read speed of PCIe 5.0 (approximately 10-14 GB/s) is already hard to fully utilize in daily use, and the 28 GB/s of PCIe 6.0 offers limited perceptible benefits in scenarios like game loading and file copying. For the comprehensive popularization of consumer PCIe 6.0, the industry generally expects it to happen around 2030.
The enterprise PCIe 6.0 SSDs from the three major manufacturers are all in place, but the rollout of consumer products depends on when the controller chips and client platforms mature. For at least the next year and a half, PCIe 6.0 will remain exclusive to the enterprise market.
#NAND #EnterpriseSSD #Micron #Kioxia #Samsung