This is data-center hardware, not something that will turn up in a Windows desktop. It still matters to anyone planning storage for Windows Server, Azure Local, SQL Server archives or object-store clusters. The same trade-offs apply wherever flash gets packed into a 1U chassis: density, bandwidth, endurance and cost.
What Kioxia actually announced
Kioxia's own announcement lists the following:
- The drive is PCIe 5.0 compliant but runs at Gen 4 x4. Kioxia's footnote gives the speed as up to 16 GT/s x4.
- It is NVMe 2.0e compliant.
- It supports the Open Compute Project Datacenter NVMe SSD Specification (v2.6).
- It has a single-port design.
- Capacities are 15.36TB and 30.72TB, with a validated architecture for up to 122.88TB.
- It uses BiCS FLASH generation 8 QLC.
Kioxia says samples are "currently sampling" to select customers. The company's note says the samples are for functional checks and that specifications may differ in mass production. Kioxia will show the drives at the 2026 OCP Global Summit in San Jose, October 12–15.
The 122.88TB figure needs a caveat
Tom's Hardware notes that Kioxia has not said when 122.88TB drives will arrive. It also has not said when the series enters mass production. StorageReview makes the same point: the largest sampling configuration is a quarter of the 122.88TB the architecture supports. Treat the top capacity as a roadmap signal.
Capacity labels also deserve care. Kioxia defines a terabyte as 1 trillion bytes. An operating system that reports in powers of two (1TiB = 2^40 bytes) will show a smaller number, and formatted capacity varies further. So "122.88TB" does not mean 122.88TiB of usable space.
"PCIe 5.0" with a Gen 4 asterisk
The interface is the detail most likely to confuse buyers. Kioxia lists PCIe 5.0 compliance but specifies 16 GT/s across four lanes, which is Gen 4 signaling. Tom's Hardware puts the theoretical ceiling at roughly 7.88 GB/s in each direction. StorageReview says that is about half what a Gen 5 x4 drive could reach.
StorageReview argues that this is a reasonable trade for the class. For a read-heavy QLC drive spread across many slots, the server's aggregate PCIe and network budget matters more than any single drive's speed. That is the reviewer's analysis, not a Kioxia claim. ServeTheHome makes a related point: although PCIe Gen 6 servers are starting to arrive, storage applications using QLC NAND will probably be slower to move to Gen 6 than drives in GPU nodes.
Why E1.L, and why 1U
E1.L is the long EDSFF "ruler" format. Tom's Hardware gives its size as 318.75 mm long, 38.4 mm high and 9.5 mm thick. ServeTheHome describes the shape as extending the depth for longer PCBs, and therefore more NAND sites, to fit more flash capacity into a 1U chassis. More packages on one board means more terabytes behind each front-panel slot.
ServeTheHome also frames the form factor's value in terms of front-panel space. You either use less faceplate area for a given capacity or fit more drives. Server compatibility still depends on the chassis, backplane and system design. Kioxia has not published a universal compatibility list.
The QLC trade-off
QLC stores four bits per cell, so a drive can reach a given capacity with fewer NAND packages than TLC. StorageReview notes the cost of that density shows up in write endurance and sustained write speed. That is why Kioxia pitches the LD4 at read-heavy jobs such as AI and machine-learning data repositories and object storage. StorageReview's specification table lists power loss protection and end-to-end data protection.
Kioxia America's Neville Ichhaporia said QLC is compelling for read-intensive applications where high capacity and high read performance serve AI workloads. That is the vendor's positioning, and no independent testing has been published.
What is still missing
Tom's Hardware points out that Kioxia has not disclosed several figures that hyperscale buyers would want:
- sequential read and write speeds
- random IOPS
- latency
- endurance ratings (DWPD or TBW)
- power consumption
Without those, nobody can work out cost per terabyte, drive-writes budgets or watts per rack from the announcement alone. StorageReview also reports that Kioxia notes the drive does not meet every OCP requirement. Kioxia's own release says only that it supports the OCP v2.6 specification. Ask for the exact list of OCP provisions before you rely on it.
Practical takeaways
- Treat 15.36TB and 30.72TB as the real products for now. They are sampling and not yet in mass production.
- Plan network and PCIe budgets around Gen 4 x4 per drive. Do not assume Gen 5 throughput.
- Keep QLC for read-heavy tiers. Write-heavy databases and log pipelines need endurance data that Kioxia has not yet published.
- Request the full datasheet and qualification data. That means performance, endurance, power, thermals, firmware features and the OCP provisions supported.
- Watch the OCP Summit. Kioxia is due to demonstrate the drives there, and may supply more detail.
Kioxia has shown a credible density play for hyperscale servers, but it is a sampling-stage announcement with the key numbers still unpublished.