Kioxia unveils an insane new SSD that could feed GPUs millions of IOPS and break AI memory limits
- Kioxia GP Series SSD provides GPUs with faster access to memory beyond HBM limits
- Storage class memory closes the performance gap between DRAM and conventional NAND flash storage
- XL-FLASH prioritizes low latency and millions of random IOPS over sequential speed
Kioxia has introduced a new type of SSD designed to work as a direct memory expansion for GPUs.
The new Kioxia GP series, announced at Nvidia GTC 2026, is not a replacement for existing storage, but rather an additional level in the memory hierarchy.
Its primary role is to provide a larger pool of quickly accessible data to GPUs, thus acting as an overflow for expensive and capacity-limited high-bandwidth memory.
Article continues below
Memory-intensive AI models are driving the change
The drive leverages Storage Class Memory (SCM), a class of technology that falls into the performance gap between traditional NAND flash memory and system DRAM.
This concept was popularized years ago by Intel’s now-discontinued Optane technology, which aimed to bridge the same gap but failed.
Kioxia’s version, dubbed XL-FLASH, prioritizes low latency and high I/O operations per second over raw sequential throughput, enabling finer-grained data access at just 512 bytes.
This development is a direct response to a fundamental problem with current AI infrastructure: GPU memory is simply not large enough for the models it is intended to run.
As AI models scale to billions of parameters and pop-ups scale to millions of tokens, the demand for memory to store things like the KV cache has exceeded the physical limits of HBM.
Nvidia’s Storage-Next initiative, supported by this SSD, was created to address this exact bottleneck by encouraging storage vendors to create drives that GPUs can directly connect to.
“Kioxia fully supports the NVIDIA Storage-Next initiative and will provide SSDs specifically designed to effectively address the need for GPU-accessible memory,” said Makoto Hamada, Senior Director of the SSD Division at Kioxia Corporation.
While the GP series targets millions of IOPS to power GPUs with data, the industry as a whole is pursuing even more ambitious performance goals.
The idea of reaching 100 million IOPS is a known industrial pipe dream, which could require the invention of entirely new classes of memory.
Other companies are also targeting the niche market left by Optane.
For example, the InnoGrit N3X SSD uses Kioxia’s XL-Flash in SLC mode to provide extreme endurance.
This drive can support up to 50 full writes per day for five consecutive years.
Kioxia itself has already signaled its intention to achieve even higher performance.
The company is targeting 10 million IOPS using SLC NAND, suggesting that the GP series is just one step in a longer race to eliminate storage latency from processors.
It remains to be seen whether the Kioxia GP Series SSD will succeed where previous SCM attempts have failed.
Follow TechRadar on Google News And add us as your favorite source to get our news, reviews and expert opinions in your feeds. Make sure to click the Follow button!
And of course you can too follow TechRadar on TikTok for news, reviews, unboxings in video form and receive regular updates from us on WhatsApp Also.
