Industry executives and experts share their predictions for 2026. Read them in this 18th annual VMblog.com series exclusive.
By Avi Shetty, Senior Director, AI Market Enablement and Partnerships, Solidigm
As we move into 2026, the enterprise technology industry will undergo continued transformation as the rapid evolution of artificial intelligence (AI) and high-performance computing reshapes data storage. With the unrelenting pursuit of GPU power, storage solutions must not only keep pace but also innovate to address the unique challenges posed by increasingly advanced computing workloads. From enhanced SSD architectures to the rise of liquid cooling technologies, advancements in storage throughout 2026 and beyond will redefine how data is accessed and managed in next-generation systems.
Storage Evolving to Keep Pace with Increasing GPU Power
As GPUs get more powerful, storage must become more efficient to match lower power consumption with higher performance.
Storage challenges inside the server and across clusters are multifaceted. Inside the server, modern accelerators can consume data faster than most NVMe SSDs can deliver. This exposes limits in queue depth, latency consistency, and PCIe bandwidth when GPUs demand millions of small, random reads per second for applications like retrieval-augmented generation (RAG), vector search, and fine-grained training workloads. Across clusters, the challenge escalates: feeding thousands of GPUs requires large fleets of SSDs that can deliver extreme parallelism, perfect wear-levelling under punishing read-heavy AI workloads, and sustained performance without Quality of Service (QoS) drops.
The key to overcoming these challenges lies in the size and scale of storage solutions and ensuring efficient access to data. It’s not just about single-drive speeds, but solution-level enablement that can handle the demands of powerful GPUs.
Storage Bottlenecks and Performance Slowdowns
As GPU power continues to increase, storage performance will be the bottleneck due to several factors, including:
- Power availability limitations
- Thermal capabilities of air-cooling systems
- Liquid cooling flow rate availability
Performance slowdowns will emerge wherever the storage path can’t keep pace with GPU data demands. Local NVMe drives will hit limits on IOPS (Input/Output Operations Per Second), queue depth, and tail latency under AI workloads. As accelerators become more powerful, storage efficiency will become the defining factor in system performance, rather than compute power.
However, storage will not be the only bottleneck; the scope, scale, and interaction with the host system will have similar impacts. Therefore, it’s crucial to ensure that internal and external networking can efficiently move data from storage to compute resources.
The Next Generation of SSDs will Adapt to Meet AI Storage Needs
SSDs will evolve into purpose-built AI drives, acting less like storage and more like high-bandwidth memory extensions.
To support workloads that demand both fast access and high endurance near GPUs, SSDs will need to evolve in several key ways. First, controllers will be redesigned for much deeper parallelism to service extreme random-read IOPS without tail-latency spikes. Storage architectures will also evolve in how they levrage different aspects of SSDs to survive read-heavy AI workloads.
PCIe and NVMe stacks will become more streamlined to shave microseconds off the data path, and we will see smarter firmware, telemetry, and data-placement algorithms that prefetch and stage data exactly where GPUs need it.
As has always been the case, SSDs will provide the necessary latency for data movement to satisfy the growing need for data. By migrating more resident data to SSDs from other media, such as HDDs, data availability is drastically improved.
The Rise of Liquid-Cooled Storage
I envision a future where every component of a server design is fanless and liquid-cooled. For example, working with NVIDIA, Solidigm has led the effort to introduce the industry’s first liquid-cooled eSSD with single-sided cold-plate technology. Key drivers of this shift include the removal of previously required fans when liquid cooling is used for GPUs /CPUs and the inability of air cooling to meet future cooling demands as storage power and performance increase.
Liquid cooling allows SSDs to run at higher sustained performance in the same footprint as previously air-cooled solutions. The adoption of liquid cooling removes a significant constraint in server design-the need for airflow. By removing heat far more efficiently than air, liquid cooling enables storage to keep pace with the thermal and density demands of next-gen AI systems.
With fully liquid-cooled solutions, we can better optimize how systems are designed, unencumbered by the limitations of traditional fan-based cooling. This shift enables denser server designs and accelerates the transition from legacy form factors such as U.2 to EDSFF.
##
ABOUT THE AUTHOR
Avi Shetty is a Senior Director at Solidigm dedicated to driving AI ecosystem partnerships and establishing the company as a thought leader through strategy and marketing initiatives. His goal is to foster a cohesive AI ecosystem that supports innovation and growth. Over the last 18 years, he has held leadership roles across strategy, marketing, and NAND design engineering within Intel’s storage group. Prior to Intel, Avi started his career as a Design Engineer at Tanner Research. Avi received his Masters in Electrical Engineering from University of Southern California (USC) and Masters in Business Administration from University of California, Davis.
