On November 11, 2026, Kioxia announced an object storage configuration, developed in collaboration with Dell Technologies, that fits roughly 157PB of raw capacity into a standard 42U rack. It uses 640 high-capacity SSDs of 245.76TB each, installed in 16 Dell PowerEdge servers. Dell announced a high-density storage configuration in August that held 9.83PB in a 2U chassis, and the new setup extends that design to the rack level. However, 157PB is the sum of the SSDs' nominal capacities and differs from the amount of data that can actually be stored. Judging the benefits of deploying it requires considering the space needed for data protection, power consumption, cooling equipment, and more.
157PB in a 42U Rack, With 640 SSDs of 245.76TB Each
The configuration places a total of 640 Kioxia LC9 series SSDs in 16 Dell PowerEdge R7725xd servers and runs Dell's object storage software, ObjectScale. ObjectScale is a storage platform for storing and managing unstructured data such as images, video, and documents used in AI training and big data analytics. According to Kioxia's November 11 announcement, the configuration breaks down as follows.
| Item | Announced details |
|---|---|
| SSD | Kioxia LC9 series, 245.76TB each |
| Number of SSDs | 640 in total |
| Servers | 16 Dell PowerEdge R7725xd |
| Storage software | Dell ObjectScale |
| Rack | Standard 42U rack |
| Raw capacity | Approx. 157PB |
The 157PB figure is a simple sum of the capacities of the installed SSDs. With 640 SSDs of 245.76TB each, the calculated raw capacity is as follows.
245.76TB × 640 drives = 157,286.4TB = 157.2864PB
This matches the announced figure of roughly 157PB. The capacities here are in decimal units: 1TB is 10¹² bytes, and 1PB is calculated as 1,000TB.
It is important to note that 157PB does not represent the capacity actually available for use. Storage systems need some space for formatting and management information, and they also use capacity for redundancy to prevent data loss in the event of a failure. For example, usable capacity varies with the protection scheme adopted, such as data replication or erasure coding. The announcement does not disclose the effective capacity after such settings are applied, so 157PB cannot be taken as the amount of data that can be stored.
Kioxia calls this configuration a "world first," but that is a claim based on the company's own research as of November 11, 2026. Although the system is said to fit in a standard 42U rack, details of the placement of servers, network equipment, and other components have not been published. It is unclear whether servers occupy all 42U or how much space is allocated to networking gear, so the raw capacity calculation should be distinguished from the rack's implementation conditions.
Extending a 9.83PB, 2U Design to a Full Rack
The 245.76TB SSD used in this announcement is not a new product. In an official blog post on August 12, 2026, Dell introduced a high-density storage configuration that placed 40 245.76TB NVMe SSDs in a 2U system for 9.83PB of raw capacity. Dell explained at the time as well that usable capacity after formatting would differ. The distinguishing feature of the new announcement is that it extends this high-density SSD design across 16 servers and presents it as a rack-scale storage system using ObjectScale.
The SSD counts also show how the two configurations relate. Distributing 640 SSDs across 16 servers works out to 40 SSDs per server. With 40 SSDs of 245.76TB, the raw capacity is 9.8304PB, matching the 9.83PB Dell announced in August. However, the August article lists the server as the PowerEdge R7725, whereas this configuration uses the R7725xd. The SSD count and capacity are common to both, but the overall system specifications are not necessarily identical.
Supporting this density is Kioxia's high-capacity enterprise SSD, the LC9 series. The series uses 3D QLC NAND from eighth-generation BiCS FLASH and achieves its large capacity by stacking 32 NAND dies. The 32-die stacking here refers to dies stacked within a package and is different from the number of layers in the NAND memory cells themselves.
The maximum capacity of 245.76TB is available only in the E3.L form factor; the maximum capacity for the 2.5-inch and E3.S models is 122.88TB. Even within the same LC9 series, the capacity that can be installed depends on the form factor. When considering a high-density configuration like this one, it is therefore necessary to check not only SSD capacity but also the form factors the server supports and how many drives it can hold.
Power, Cooling, and Data Protection Determine the Real Space Savings
Kioxia's aim with this configuration is to reduce installation space compared with using HDDs or lower-capacity SSDs, while also gaining advantages in power efficiency and cooling. AI training and big data analytics require storing large volumes of images, video, and other unstructured data, so growth in storage capacity translates directly into data center floor space and power demand. If raw capacity per rack can be increased, the number of racks and amount of equipment needed to secure the same capacity may be reduced. However, the 157PB figure indicates only storage capacity; it does not mean AI training speed or GPU utilization will improve.
Evaluating actual power savings requires power consumption data as well as capacity density. Kioxia cites improved TB/W, a measure of capacity per unit of power, as an advantage, but it has not published measured power consumption for this rack configuration, the power required for cooling, or the reduction rate compared with existing systems. The reductions in power use and cooling load are therefore, for now, benefits claimed by the manufacturer and have not been confirmed by independent measurement. When comparing against existing HDD systems or lower-capacity SSD configurations, it is necessary to evaluate power consumption across the entire system, including servers and network equipment, under conditions that secure the same effective capacity.
The LC9 series also includes features that improve write efficiency and reliability inside the SSD. One is FDP (Flexible Data Placement), which aims to extend SSD lifespan by controlling data placement appropriately and thereby reducing write amplification. Write amplification is a phenomenon in which the amount of data actually written inside the SSD exceeds the amount written by the host, leading to wear on the NAND. The LC9 series also offers die failure protection, which keeps the SSD operating even if a NAND die fails.
However, these features improve reliability inside the SSD and do not guarantee data protection for the entire rack. In real storage systems, data redundancy and recovery methods must be designed to account for server and network equipment failures as well as individual SSD failures. Effective capacity and recovery time after a failure also vary with the protection scheme adopted in ObjectScale, but the announcement does not give specific settings or recovery performance.
The significance of this announcement lies in showing a configuration that uses 245.76TB high-capacity SSDs to increase the density of object storage at the rack level. For companies judging the actual benefits of deployment, however, the 157PB raw capacity alone is not enough. Once the effective capacity after data protection, the power consumption and cooling requirements of the whole system, recovery time after failures, and deployment and operating costs are known, it will be possible to evaluate concretely how far existing storage facilities can be reduced. Going forward, it will be important for measured data covering these operating conditions to be published.