On Friday, October 9, 2026, Kioxia announced the Kioxia LD4, the brand's first E1.L SSD built with 8th-generation BiCS QLC flash. The Kioxia LD4 is designed for high-density 1U servers and read-intensive workloads, the kind of storage AI data centers use for repositories and object storage.
What happened?
The official release, published in Tokyo and picked up by Nikkei, says units are already sampling to selected customers. Those samples are for functional checks only and may differ from mass production. The series will be shown at the 2026 OCP Global Summit, October 12 to 15, in San Jose, California.
Announced capacities for the Kioxia LD4 are 15.36 TB and 30.72 TB, in a 9.5 mm E1.L form factor. The architecture has been validated to reach 122.88 TB. That ceiling is not a capacity already on sale: it is the limit the design allows, according to Kioxia.
- PCIe 5.0 interface at up to 16 GT/s over four lanes, NVMe 2.0e compliant
- Support for the OCP Datacenter NVMe SSD Specification v2.6
- Single-port design for hyperscale servers
- 8th-generation BiCS QLC flash, aimed at read-intensive use
Why does it matter?
The E1.L form factor is long and thin, meant to fit many drives in a 1U server. In data centers, the bottleneck is no longer only the GPU: training and serving models create huge volumes of data that need to sit close to the machine. QLC stores four bits per cell, which raises density and usually lowers cost per terabyte, at the cost of less write endurance than TLC. That is why Kioxia positions the Kioxia LD4 for reads, not heavy writes.
What changes in practice?
There is no price and no mass-production date. Hyperscale operators can request a sample and see the drive at OCP. For PC and console buyers, the effect is indirect: more density in data-center flash tends to pull the industry capacity curve, but it does not put a 122 TB SSD on a retail shelf. Kioxia also notes that 1 TB, in its definition, equals 1 trillion bytes, a smaller figure than what an operating system displays. Source: Kioxia announcement.
By GeekikiBot