- Sandisk’s 256TB SSD is completely skipped, raising concerns about the workload performance of the whale
- Faster speed claims still do not verify without public reference points or IOPS performance numbers
- Direct writing QLC can sacrifice speed in exchange for greater reliability and data integrity
Sandisk has announced a 256TB SSD, the UltraqlC SN670, which will be sent in the first half of 2026.
This model represents the largest SSD ever revealed by the company, marking a bold step towards high density storage solutions adapted for AI and hyperscale infrastructure.
Although the company plans to launch the 128TB version to the evaluators in a matter of weeks, the complete commercial availability remains months away.
An architecture built for the scale, not the speed
In essence, the SN670 is based on a BICS NAND 3D architecture of 218 layers and presents a 2TB CBA dice (CMOS directly linked to the matrix).
It is connected through a PCIE Gen5 NVME interface and is part of the new Ultraqlc Sandisk platform.
Unlike conventional SSDs that cushion the data through pseudo-SLC caches, this model uses a “direct writing QLC” approach.
This simplifies the writing process and makes the unit safer for the loss of power, but also introduces compensation, especially when it comes to performance under heavy or short loads.
Without a SLC cache, the SN670 can suffer slower cutting writings, an inconsistent performance under load and an increase in controller demands, which makes it less receptive during intensive or unpredictable workloads.
However, Sandisk states that SN670 offers more than 68% of faster random readings and 55% fastest random deeds compared to an SSD QLC Gen 5 128 TB leader.
Sequential reading speeds are more than 7% better, while sequential writing speeds improve in more than 27% in internal comparisons.
Sandisk has emphasized benefits as the dynamic frequency scale, which is said to improve yield by up to 10 percent at the same power level
He also states that the data retention profile could reduce recycling wear by up to 33%.
Both characteristics are intended to improve longevity and reduce energy consumption.
However, none of these statements is backed by revealed performance data, such as reading/writing speeds or resistance figures.
Internally, the UltraqlC SN670 is compatible with a custom controller and firmware, which according to Sandisk allows a better latency and bandwidth, but without real reference comparisons or IOPS, these statements remain marketing -based projections.
It is worth noting that the previous iterations of the Sandisk business units using QLC Nand showed limitations compared to TLC -based models.
In this case, native QLC programming latencies could reach 800–1200 microseconds, several times slower than SLC -based designs.
Sandisk may be depending on optimizations such as great DRAM shock absorbers or advanced Die, but such architectural details have not yet been confirmed.
The final product will initially arrive in form U.2, and more variants are expected later in 2026.
For now, Sandisk’s 256TB unit is a symbolic leap towards the future data infrastructure, not a realistic option for conventional users.
Through blocks and files
